A method and device for quickly evaluating the optimal adjustable capacity of a virtual power plant

Abstract

The present application discloses a method and device for quickly evaluating the optimal adjustable capacity of a virtual power plant, relating to the field of virtual power plants. The method includes that during the training process, the objective of the reconstruction layer is to determine a reconstruction strategy from a large number of feasible network topologies, while receiving the target value fed back by the evaluation layer, and iteratively updating the distribution network reconstruction strategy through interaction with the evaluation layer; the evaluation layer continuously receives the network structure determined by the reconstruction layer, and at the same time, considering the node characteristic constraints and network topology constraints, adopts a method for evaluating the adjustable capacity of a virtual power plant based on vertex search to evaluate the feasible region of the virtual power plant under a given topology until the optimal feasible region is found; during application, the optimal distribution network reconstruction strategy of the target distribution network is obtained by using an integrated network of a trained graph convolutional network and a bidirectional long short-term memory network, and then the optimal adjustable capacity of the virtual power plant is solved. The present application can quickly calculate the optimal adjustable capacity of a virtual power plant in a dynamic distribution network network structure.

Description

Technical Field

[0001] The present application relates to the field of virtual power plants, and in particular, to a method and device for quickly evaluating the optimal adjustable capacity of a virtual power plant. Background Art

[0002] With the rapid development of renewable energy technologies, the penetration rate of distributed resources with regulation potential, such as distributed photovoltaic, energy storage, temperature-controlled loads, and electric vehicles, in the distribution network has increased significantly. However, distributed resources are characterized by small scale, distribution, and heterogeneity, which pose significant challenges to the efficient utilization of the flexible regulation potential of individual distributed resources. As a feasible solution, the virtual power plant (VPP) emerged, using advanced communication and control technologies to aggregate distributed resources scattered in the distribution network into a whole, participating in power market transactions and grid ancillary services, thereby enhancing the flexibility and stability of the power grid. The determination of the adjustable capacity is a prerequisite for the virtual power plant to participate in ancillary services. However, a virtual power plant usually contains a large number of distributed resources with different operating response characteristics, so evaluating the adjustable capacity of a virtual power plant is a complex and challenging problem. In addition, the virtual power plant must consider the impact of network topology on its adjustable capacity to prevent issuing unexecutable dispatch commands or endangering the network security of the distribution network. In this case, the operating security constraints related to the network topology will limit the regulation capacity of the virtual power plant and hinder its full play of the regulation potential of distributed resources.

[0003] Currently, the evaluation methods for the adjustable capacity of virtual power plants can be divided into two types: bottom-up and top-down. Specifically, the bottom-up methods, such as the virtual energy storage modeling method, involve a detailed description of the operating characteristics of each distributed resource, and then gradually summarize through Minkowski summation to determine the overall regulation capacity of all distributed resources. In addition, approximate evaluation methods, including inscribed hyperboxes, inscribed homothetic polyhedra, and Chino polyhedron transformation, usually set specific shapes to approximately evaluate the adjustable capacity of the virtual power plant, sacrificing a certain degree of flexibility to improve the evaluation efficiency. However, it is difficult for these methods to incorporate network topology constraints into the evaluation process. On the contrary, top-down methods, such as the Fourier-Motzkin elimination method and the vertex enumeration method, evaluate from an overall perspective and essentially involve projecting from a high-dimensional polyhedron to a low-dimensional feasible region polyhedron. However, these methods usually generate a large number of new constraints or enumerate a large number of vertices, resulting in low computational efficiency.

[0004] Secondly, different network topologies can lead to different power flows and voltage distributions, which are crucial for maintaining power flow balance, ensuring voltage security, and complying with line capacity constraints. These changes have a direct impact on the operation flexibility and regulation ability of the virtual power plant. In some unreasonable network structures, it is easier to encounter problems such as line overload and voltage violation when aggregating distributed resources, making it difficult to fully utilize the regulation potential of distributed resources. Fortunately, most distribution networks have the ability to adjust their topologies, enabling the implementation of network reconfiguration strategies. Network reconfiguration strategies adjust the topology of the distribution network by changing the connection status of power lines and have been widely applied in aspects such as fault recovery, power flow optimization, loss reduction, and voltage improvement. However, network reconfiguration has not been applied to improve the adjustable ability of virtual power plants. Utilizing the advantages of network reconfiguration in enhancing the operation performance of the power grid may help to explore the maximum regulation ability of virtual power plants under dynamic network topologies.

[0005] Finally, the network reconfiguration problem is usually formulated as a non-linear integer programming problem, and its solution is inherently complex. Current solution methods can be roughly divided into mathematical programming methods and heuristic algorithms. Branch and bound, dynamic programming, and second-order cone programming are the earliest mathematical programming methods to accurately solve simple network topology reconfiguration problems. They do not depend on the initial network topology, but as the scale of the power grid increases, they face the challenge of the curse of dimensionality. Heuristic algorithms, such as branch exchange algorithms and particle swarm algorithms, although the solution efficiency has been slightly improved, it is difficult to guarantee the quality of the solution. In recent years, machine learning methods such as long short-term memory networks and graph convolutional networks have received increasing attention in solving optimization problems, especially voltage regulation and power flow calculation. Machine learning methods can not only achieve a fast mapping from input data to output results but also avoid the errors caused by simplification or relaxation of traditional physical models. However, applying machine learning methods to solve network reconfiguration to enhance the adjustable ability of virtual power plants remains to be explored.

[0006] The defects and difficulties existing in the above virtual power plant optimal adjustable ability evaluation process are mainly manifested in the following 1)-3).

[0007] 1) Difficulty in evaluating the adjustable ability of the virtual power plant: For the aggregation of the adjustable ability of the virtual power plant, bottom-up methods are difficult to consider the relevant distribution network safe operation constraints caused by the network structure during the process of aggregating distributed resources, while existing top-down methods cannot balance the solution accuracy and computational efficiency.

[0008] 2) The adjustable potential of the virtual power plant remains to be explored: Due to the need to ensure the safe operation of the distribution network, the relevant safe operation constraints brought by the network structure will limit the adjustable ability of the virtual power plant. Especially in some network structures with unreasonable initial designs, there is a technical gap in the research on introducing network reconfiguration to seek the maximum adjustable ability of the virtual power plant.

[0009] 3) Consider the complexity of solving the network reconstruction model: The network reconstruction model is usually formulated as a non - linear integer programming problem, which is relatively complex to solve. Traditional mathematical programming methods are difficult to ensure the solution efficiency of the reconstruction strategy, while heuristic algorithms cannot guarantee the solution quality. Summary of the Invention

[0010] The objective of this application is to provide a method and device for quickly evaluating the optimal adjustable capacity of a virtual power plant, so as to quickly calculate the optimal adjustable capacity of the virtual power plant in the dynamic distribution network structure.

[0011] To achieve the above objective, the following solutions are provided in this application.

[0012] In a first aspect, the present application provides a method for rapidly evaluating the optimal adjustable capacity of a virtual power plant, including: establishing a two-layer evaluation architecture including a reconstruction layer and an evaluation layer; generating a distribution network reconstruction strategy for each historical operating state in the reconstruction layer and transmitting it to the evaluation layer; in the evaluation layer, jointly constructing a historical state space with the node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the distribution network reconstruction strategy received this time; using the breadth-first search method to perform dimensionality reduction projection on the historical state space to obtain the feasible region of the output power of the virtual power plant under the distribution network reconstruction strategy received this time, and calculating the area of the feasible region this time; according to the area of the feasible region this time, using the output power of the common coupling point in the received distribution network reconstruction strategy as the decision variable, calculating the target value when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized, and transmitting both the target value this time and the area of the feasible region this time to the reconstruction layer; in the reconstruction layer, selecting the larger target value between the target value received this time and the target value received last time, and according to the area of the feasible region corresponding to the larger target value, using the line connection state of the distribution network as the decision variable, obtaining the distribution network reconstruction strategy when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized, and transmitting it to the evaluation layer, returning to execute the step of jointly constructing a historical state space with the node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the distribution network reconstruction strategy received this time in the evaluation layer until the target values received in two adjacent times are equal, and determining the latest distribution network reconstruction strategy in the reconstruction layer as the optimal distribution network reconstruction strategy for each historical operating state; using each historical operating state as the input and the optimal distribution network reconstruction strategy for each historical operating state as the label to train an integrated network of a graph convolutional network and a bidirectional long short-term memory network to obtain a strategy solving model; in the reconstruction layer, according to the current operating state of the target distribution network, using the strategy solving model to obtain the optimal distribution network reconstruction strategy of the target distribution network and transmitting it to the evaluation layer; jointly constructing the current state space with the node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network; in the evaluation layer, using the breadth-first search method to perform dimensionality reduction projection on the current state space to obtain the feasible region of the output power of the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network, and calculating the area of the feasible region of the target distribution network and transmitting it to the reconstruction layer; the area of the feasible region of the target distribution network represents the optimal adjustable capacity of the virtual power plant.

[0013] In a second aspect, the present application provides a device for rapidly evaluating the optimal adjustable capacity of a virtual power plant, including: an establishment module, a strategy generation module, a state space construction module, a dimensionality reduction projection module, an area calculation module, a strategy update module, a training module, a model application module, a constraint integration module, and an optimal adjustable capacity determination module.

[0014] A building module, which is used to build a two - layer evaluation architecture including a reconstruction layer and an evaluation layer. A strategy generation module, which is used to generate a distribution network reconstruction strategy for the distribution network in each historical operating state in the reconstruction layer and transmit it to the evaluation layer. A state space construction module, which is used to jointly construct a historical state space with the node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the distribution network reconstruction strategy received this time in the evaluation layer. A dimensionality reduction projection module, which is used to perform dimensionality reduction projection on the historical state space by using the breadth - first search method to obtain the feasible region of the output power of the virtual power plant under the distribution network reconstruction strategy received this time and calculate the area of the feasible region this time.

[0015] An area calculation module, which is used to calculate the target value when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized with the output power of the point of common coupling in the received distribution network reconstruction strategy as the decision variable according to the area of the feasible region this time, and transmit both the target value this time and the area of the feasible region this time to the reconstruction layer.

[0016] A strategy update module, which is used to select the larger target value between the target value received this time and the target value received last time in the reconstruction layer, and obtain the distribution network reconstruction strategy when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized with the line connection state of the distribution network as the decision variable according to the area of the feasible region corresponding to the larger target value, and transmit it to the evaluation layer, and call the state space construction module until the target values received in two adjacent times are equal, and determine the latest distribution network reconstruction strategy in the reconstruction layer as the optimal distribution network reconstruction strategy for each historical operating state.

[0017] A training module, which is used to train an integrated network of a graph convolutional network and a bidirectional long - short - term memory network with each historical operating state as the input and the optimal distribution network reconstruction strategy for each historical operating state as the label to obtain a strategy solving model.

[0018] A model application module, which is used to obtain the optimal distribution network reconstruction strategy of the target distribution network by using the strategy solving model according to the current operating state of the target distribution network in the reconstruction layer and transmit it to the evaluation layer.

[0019] A constraint integration module, which is used to jointly construct the current state space with the node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network. An optimal adjustable capacity determination module, which is used to perform dimensionality reduction projection on the current state space by using the breadth - first search method in the evaluation layer to obtain the feasible region of the output power of the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network and calculate the area of the feasible region of the target distribution network, and transmit it to the reconstruction layer; the area of the feasible region of the target distribution network represents the optimal adjustable capacity of the virtual power plant.

[0020] According to the specific embodiments provided by the present application, the present application has the following technical effects.

[0021] The present application provides a method and device for quickly evaluating the optimal adjustable capacity of a virtual power plant. In a two-layer evaluation framework, the goal of the reconstruction layer is to determine a reconstruction strategy from a large number of feasible network topologies, while receiving the evaluation results (target values) fed back by the evaluation layer, and iteratively updating the distribution network reconstruction strategy through interaction with the evaluation layer; the evaluation layer continuously receives the network structure determined by the reconstruction layer, and while considering node characteristic constraints and network topology constraints, adopts a method for evaluating the adjustable capacity of a virtual power plant based on vertex search to evaluate the feasible region of the virtual power plant under a given topology until the optimal feasible region is found; the graph convolutional network and the bidirectional long short-term memory network respectively extract the spatial dependence and long-term dependence of the input data, realize the mapping from the distribution network operation state to its optimal reconstruction strategy, and efficiently obtain the optimal network reconstruction strategy corresponding to the optimal adjustable capacity of the virtual power plant, so as to quickly calculate the optimal adjustable capacity of the virtual power plant in the dynamic distribution network structure. Brief Description of the Drawings

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0023] Figure 1 It is a schematic flow chart of a method for quickly evaluating the optimal adjustable capacity of a virtual power plant provided by an embodiment of the present application.

[0024] Figure 2 It is a schematic framework diagram of a method for quickly evaluating the optimal adjustable capacity of a virtual power plant provided by an embodiment of the present application. Detailed Description of the Embodiments

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0026] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0027] In view of the defects and difficulties existing in the evaluation process of the optimal adjustable capacity of a virtual power plant, in an exemplary embodiment, such asFigure 1 As shown in the figure, the present application provides a method for quickly evaluating the optimal adjustable capacity of a virtual power plant, including the following steps 101 to 110.

[0028] Step 101: Establish a two-layer evaluation architecture including a reconstruction layer and an evaluation layer.

[0029] Step 102: Generate a distribution network reconstruction strategy for the distribution network in each historical operating state in the reconstruction layer and transmit it to the evaluation layer.

[0030] Step 103: In the evaluation layer, jointly construct a historical state space with the node characteristics constraints and network topology constraints of the distributed resources in the virtual power plant under the received distribution network reconstruction strategy.

[0031] Step 104: Use the breadth-first search method to perform dimensionality reduction projection on the historical state space, obtain the feasible region of the output power of the virtual power plant under the received distribution network reconstruction strategy, and calculate the area of the current feasible region.

[0032] Step 105: According to the area of the current feasible region, use the output power of the point of common coupling in the received distribution network reconstruction strategy as the decision variable, calculate the target value when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized, and transmit both the current target value and the area of the current feasible region to the reconstruction layer.

[0033] The point of common coupling refers to the point of common coupling with the superior power grid.

[0034] Step 106: Select the larger target value between the current received target value and the previously received target value in the reconstruction layer, and according to the area of the feasible region corresponding to the larger target value, use the line connection state of the distribution network as the decision variable to obtain the distribution network reconstruction strategy when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized, and transmit it to the evaluation layer. Then return to execute Step 103 until the target values received in two adjacent times are equal, and determine the latest distribution network reconstruction strategy in the reconstruction layer as the optimal distribution network reconstruction strategy for each historical operating state.

[0035] Step 107: Use each historical operating state as the input and the optimal distribution network reconstruction strategy for each historical operating state as the label to train an integrated network of a graph convolutional network and a bidirectional long short-term memory network to obtain a strategy solving model.

[0036] Step 108: In the reconstruction layer, according to the current operating state of the target distribution network, use the strategy solving model to obtain the optimal distribution network reconstruction strategy of the target distribution network and transmit it to the evaluation layer.

[0037] Step 109: The node characteristic constraints and network topology constraints of the distributed resources in the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network together constitute the current state space.

[0038] Step 110: In the evaluation layer, a breadth-first search method is used to perform dimensionality reduction projection on the current state space to obtain the feasible domain of the virtual power plant output power under the optimal distribution network reconstruction strategy of the target distribution network, and the feasible domain area of ​​the target distribution network is calculated and transmitted to the reconstruction layer; the feasible domain area of ​​the target distribution network represents the optimal adjustable capacity of the virtual power plant.

[0039] This application defines the concept of a feasible domain representing the adjustable capacity of a virtual power plant, and considers node characteristics and network topology constraints at the same time. The breadth-first search method can be used to evaluate the feasible domain of the output power of a virtual power plant under a fixed topology; a two-layer evaluation architecture is used to explore the optimal adjustment capacity of a virtual power plant under a dynamic network topology. In the two-layer evaluation architecture, the goal of the reconstruction layer (also known as the upper layer) is to determine the reconstruction strategy from a large number of feasible network topologies, while receiving the evaluation results fed back by the evaluation layer (also known as the lower layer), and updating the reconstruction strategy through interactive iteration with the evaluation layer. The evaluation layer continuously receives the network structure determined by the upper layer, and uses a vertex search-based virtual power plant adjustable capacity evaluation method to evaluate the feasible domain of the virtual power plant under a given topology until the optimal feasible domain is found; finally, a data-driven fast solution model is established: the spatial dependency and long-term dependency of the input data are extracted respectively through graph convolution and bidirectional long short-term memory networks, and the mapping from the operating state of the distribution network to its optimal reconstruction strategy is realized, so as to efficiently obtain the optimal network reconstruction strategy corresponding to the optimal adjustment capacity of the virtual power plant. Among them, the long-term dependency refers to the relationship between the inputs that the model needs to consider when processing sequence data.

[0040] Although it is challenging to incorporate network reconstruction into the evaluation process of the feasible domain of virtual power plants, it is necessary to fully exploit the flexibility of distributed resources in virtual power plants. On the one hand, when the network topology is not ideal, the aggregation of distributed resources is more likely to violate the safe operation restrictions of the distribution network, such as line capacity or voltage safety constraints, resulting in reduced adjustability. On the other hand, determining the optimal feasible domain of the virtual power plant requires repeatedly solving the optimization problem until the termination condition is met, and each iteration must consider the network topology constraints. If the network topology is used as a variable in the evaluation process of the feasible domain of the virtual power plant, it is difficult to ensure the consistency of the network topology throughout the iteration process, which is likely to lead to erroneous evaluation results. To this end, a two-layer evaluation architecture is proposed. The two-layer evaluation architecture adjusts the distribution of the distribution network flow in combination with the distribution network reconstruction strategy, thereby improving the overall adjustability of the virtual power plant.

[0041] When considering the reconfiguration of the distribution network, the reconfiguration cost is an important economic indicator, which directly reflects the economic input required for network topology adjustment. Therefore, it is necessary to incorporate this economic indicator into the evaluation framework. In addition, improving the adjustable capacity of the virtual power plant is the ultimate goal of implementing the reconfiguration strategy. For this purpose, the area of the feasible region of the virtual power plant is introduced as an evaluation indicator for selecting the optimal feasible region. The larger the area of the feasible region, the better the adjustable capacity of the virtual power plant. The optimal network topology should strike a balance between economy and flexibility. Therefore, the two-layer evaluation framework comprehensively considers the distribution network reconfiguration cost and the adjustable capacity of the virtual power plant in its objective function.

[0042] The reconfiguration layer uses the line connection status as the decision variable to generate a feasible and economic network reconfiguration strategy. Its goal is to continuously optimize the objective function according to the evaluation results feedback from the lower layer, update the reconfiguration strategy, and transfer the new network topology to the evaluation layer of the lower layer. Finally, the optimal network topology is identified from among numerous feasible network topologies. In addition, the reconfiguration layer records the optimal network topologies obtained by the distribution network under different historical operating states. Then, these data are organized into a dataset for training an integrated network of a graph convolutional network and a bidirectional long short-term memory network, aiming to establish a fast mapping mechanism from the distribution network operating state to the optimal network topology.

[0043] Under the network topology given by the reconfiguration layer, the evaluation layer takes the operating state of the distribution network as the input, executes the virtual power plant adjustable capacity evaluation method based on breadth-first search, and obtains the target value and the area of the feasible region under the premise of satisfying the node characteristic constraints and the distribution network safe operation constraints. Then, the target value and the area of the feasible region are fed back to the reconfiguration layer to compare the advantages and disadvantages of different network reconfiguration strategies, and the reconfiguration strategy with a larger target value is given priority. To sum up, through the iterative interaction between the reconfiguration layer and the evaluation layer, the two-layer evaluation framework finally determines the optimal network reconfiguration strategy in the reconfiguration layer and evaluates the optimal feasible region of the virtual power plant under the optimal network topology in the evaluation layer.

[0044] In another exemplary embodiment of this application, the definition of the feasible region: The node characteristic constraints and network topology constraints of the distributed resources in the virtual power plant jointly constitute a state space , and this state space is a high-dimensional state space. The state space is expressed as Equation (1).

[0045] (1)

[0046] In the formula, represents the state space; represents the projection variable, , and respectively represent the active power and reactive power of the virtual power plant at the point of common coupling; represents the internal decision variable and state variable; , both represent the coefficient matrix; represents the constant vector; represents the two-dimensional real space, referring to is a two-dimensional vector; represents -dimensional real space, referring to is a -dimensional vector.

[0047] Then, project onto a low-dimensional state space determined only by the projection variable to concisely represent the feasible region of the virtual power plant on the P-Q coupling plane, expressed as Equation (2). (2)

[0048] (2)

[0049] In the formula, describes the feasible region of the virtual power plant's output power. Each point in this region can be achieved through at least one feasible combination of internal distributed resources and does not violate the node characteristic constraints or network topology constraints. Therefore, the projection from to describes the feasible region of the virtual power plant, which represents the flexible adjustable ability of the virtual power plant at the point of common coupling.

[0050] In another exemplary embodiment of the present application, the distributed resources in the virtual power plant are distributed on different distribution network nodes and have different operating characteristics, including distributed photovoltaic, energy storage, temperature-controlled loads, and electric vehicles. Node characteristics refer to the specific attributes and behavioral characteristics generated by the distribution network nodes due to the access of distributed resources, and these characteristics determine the performance of the nodes when evaluating the adjustable ability of the virtual power plant.

[0051] The node characteristic constraints include: photovoltaic node operating characteristics, energy storage operating characteristics, electric vehicle operating characteristics, and temperature-controlled load operating characteristics.

[0052] As a typical power source type of resource, photovoltaic adjusts the active or reactive power output through a grid-connected inverter and has a certain degree of flexibility. The operating characteristics of photovoltaic nodes are expressed by Equations (3)-(5).

[0053] (3)

[0054] (4)

[0055] (5)

[0056] In the formula, , and respectively represent the active power, reactive power and grid-connected inverter capacity of the PV at node at time ; and respectively represent the upper and lower limits of the active power of the PV at node ; and respectively represent the maximum power factor angle and the minimum power factor angle of the PV at node ;

[0057] Energy storage can maintain the supply-demand balance of the distribution network through rapid charge and discharge, thereby improving the reliability and stability of grid operation. The operating characteristics of energy storage are expressed by Equations (6) - (10).

[0058] (6)

[0059] (7)

[0060] (8)

[0061] (9)

[0062] (10)

[0063] In the equations, , and respectively represent the active power, reactive power and grid-connected inverter capacity of the energy storage at node at time ; and represent the charge and discharge power of the energy storage at node at time ; , respectively represent the maximum charging power and the minimum charging power of the energy storage at node ; , respectively represent the maximum discharging power and the minimum discharging power of the energy storage at node ; represents the state of charge of the energy storage at node at time , and respectively represent the upper and lower limits of the state of charge of the energy storage at node ; and respectively represent the charge and discharge efficiency; Indicates the time interval of the charging / discharging process; Indicates the node The state of charge of the energy storage at moment.

[0064] Electric vehicles can participate in the power regulation of the power grid by dynamically adjusting the charging time and charging power through a reasonable control strategy. The operating characteristics of electric vehicles are expressed by Equations (11)-(19).

[0065] (11)

[0066] (12)

[0067] (13)

[0068] (14)

[0069] (15)

[0070] (16)

[0071] (17)

[0072] (18)

[0073] (19)

[0074] In the equations, Indicates the energy state of the electric vehicle at node at moment, and Indicates the upper and lower limits of the energy state of the electric vehicle at node at moment; Indicates the maximum energy state technically allowed for the electric vehicle at node and and Indicates the energy state of the electric vehicle at node during the grid connection time and the off-grid time; and respectively indicate the grid connection time and the off-grid time of the electric vehicle; Indicates the maximum charging power technically allowed for the electric vehicle at node ; Indicates the charging efficiency technically allowed for the electric vehicle at node ; Indicates the electric vehicle at node at The actual charging power at a moment, and respectively represent the maximum charging power and the minimum charging power of the electric vehicle at node at a moment ; and respectively represent the upper and lower limits of the energy state of the electric vehicle at node at a moment ; represents the energy state of the electric vehicle at node at a moment ; and respectively represent the remaining energy when the electric vehicle at node is connected to the grid and the minimum energy required when it is off the grid.

[0075] Specifically, if the electric vehicle starts charging at the maximum charging power from the moment of grid connection, its maximum energy state at a moment can be calculated by Equation (11). Similarly, by ensuring the expected charging demand when the electric vehicle disconnects from the grid, the minimum energy demand at a moment can be deduced inversely by Equation (12). The charging trajectories between these two charging trajectories and satisfying the charging power constraint are all feasible charging schemes for the electric vehicle, fully reflecting its flexibility.

[0076] The temperature control load can adjust the temperature setpoint between the highest temperature at node and the lowest temperature at node without affecting the user's comfort, so as to cut peaks and fill valleys. An equivalent thermal parameter model of the temperature control load is established to describe the dynamic change of the indoor temperature caused by the refrigeration power, and the equivalent thermal parameter model is Equation (20).

[0077] (20)

[0078] Wherein, represents the refrigeration power of the temperature control load at node at a moment ; represents the indoor temperature at node at a moment ; represents the outdoor ambient temperature at node at a moment ; and respectively represent the equivalent thermal resistance and the equivalent heat capacity; represents the coefficient of performance of refrigeration.

[0079] As can be seen from Equation (20), the change in indoor temperature is continuous, reflecting the change in the heat stored in the room. Therefore, the temperature control load can be equivalently modeled as an energy storage device. Specifically, the rise or fall of the indoor temperature corresponds to the discharge or charge behavior of the energy storage device. When the node at the indoor temperature at time is

[0080] (21)

[0081] (22)

[0082] (23)

[0083] wherein, represents the energy already stored in the temperature control load of node at time , represents the maximum stored energy of the temperature control load of node at time , represents the state of charge of the temperature control load of node at time .

[0084] Then, the equivalent thermal parameter model [see Equation (20)] is approximated by a difference equation as Equation (24). Substituting Equation (23) into Equation (24), the temperature control load regulation characteristics shown in Equations (25)-(28) can be obtained.

[0085] (24)

[0086] (25)

[0087] (26)

[0088] (27)

[0089] (28)

[0090] wherein, represents the state of charge of the temperature control load of node at time ; represents the cooling power of the temperature control load of node at time ; Represents the node At The outdoor ambient temperature at the moment; 、 、 、 All represent the characteristic parameters of the equivalent energy storage model of the temperature control load; And Respectively represent the node The minimum cooling power and the maximum cooling power of the temperature control load.

[0091] In another exemplary embodiment of the present application, in addition to the node characteristic constraints, it is also necessary to consider the network topology constraints. The network topology constraints include power flow balance constraints, voltage security constraints, and line capacity constraints to ensure the accuracy of the evaluation results.

[0092] The power flow balance constraint is expressed by equations (29) - (32).

[0093] (29)

[0094] (30)

[0095] (31)

[0096] (32)

[0097] In the formula, And Respectively represent the active power and reactive power generated by the node At The moment; And Represent the active load and reactive load of the node At The moment; And Represent at The moment the active power flow and reactive power flow on the line ; And Respectively are the resistance and reactance between the node And the node ; Represents at The moment the square of the current on the line ; And Represent at The moment the active power flow and reactive power flow on the line ; And Respectively represent the node Set of parent and child nodes; Indicates at Time node The voltage at; Indicates at Time node The voltage at; Indicates the L2 norm.

[0098] The voltage security constraint is given by Equation (33).

[0099] (33)

[0100] In the equation, And Indicate the node At The upper and lower limits of the voltage at the time.

[0101] The line capacity constraint is given by Equations (34) - (35).

[0102] (34)

[0103] (35)

[0104] In the equation, And Respectively represent The active and reactive powers of the virtual power plant at the point of common coupling at the time; Represents the line The line capacity limit on; Represents the line capacity limit at the point of common coupling of the virtual power plant.

[0105] In another exemplary embodiment of the present application, The calculation of is a projection problem of a high-dimensional polyhedron, and the projection result will also be a polyhedron. Since And Are both bounded, making And Must be bounded polyhedra. As bounded polyhedra, And Can be completely characterized by the convex hull constructed by the finite vertex set . The basic idea of breadth-first search is to iteratively translate the cut surface of the obtained approximate polyhedron outwards to find new vertices. Subsequently, calculate the final Convex hull of, to obtain the bounded polyhedron , which represents the feasible region of the virtual power plant on the P-Q coupling plane. The key iterative process is as follows.

[0106] 1) Optimization problem M0: It is easy to observe that when the normalized outward normal vector of the cutting plane is multiplied by any point on the cutting plane, the result is always 1. If there are still undetected vertices outside this cutting plane, there must be an optimal solution greater than 1. . Therefore, considering the high-dimensional state space constraints composed of the node characteristics constraints and network topology constraints of distributed resources, the optimization problem M0 is formulated to translate each face of the obtained polyhedron as far outwards as possible to search for new vertices. Each optimal solution obtained by solving the optimization problem M0 represents the optimal vertex identified in the current state, as shown in Equation (36).

[0107] (36)

[0108] where, represents the movement width when searching for vertices along ; represents the unit outward normal vector for the cutting plane to search for new vertices, , and represent the direction vectors on the P-axis and Q-axis respectively; represents the node within the feasible region at time , which corresponds to the th incoming vertex during the vertex search process, represents the vertex set, , where and represent the active power and reactive power at the point of common coupling in the virtual power plant feasible region respectively.

[0109] 2) Termination condition: The relative movement width is introduced to quantify the contribution of the newly identified vertex to the improvement of the existing convex hull shape, as shown in Equation (37).

[0110] (37)

[0111] In the formula, represents the maximum movement width when searching for vertices along ; is defined as, represents the norm.

[0112] Generally, assuming a sufficiently small threshold , when , it indicates that the new vertex effectively improves the shape of the existing convex hull. In this case, the vertex is saved in . If is set to zero, then All vertices, thus obtaining the precise feasible region of the virtual power plant. Otherwise, a positive will result in the omission of some vertices in, which contribute less to the improvement of the convex hull shape, thus obtaining a conservative approximation of the feasible region of the virtual power plant.

[0113] 3) Algorithm process: First, initialize the direction vector set and the vertex set . First, search for the initial vertices along both ends of the P-Q coupling plane coordinate axes, that is, initialize as . Then, substitute each direction vector in into the optimization problem M0 to solve the corresponding initial optimal solution . These initial solutions are directly stored in to complete its initialization.

[0114] Subsequently, update the direction vector set . Calculate the unit outward normal vectors between all adjacent vertices in by Equation (38) and store them as the direction vectors for the next iterative search, thus updating .

[0115] (38)

[0116] Secondly, update the vertex set . By solving the optimization problem M0, search for new vertices along all updated direction vectors in . Then, compare the corresponding relative movement widths with . If is greater than , then save the vertex and update . Repeat the above two operations until all in the current iteration are less than . At this point, the termination condition is satisfied, stop the search process, and obtain the optimal .

[0117] Finally, solve the convex hull of the optimal to determine the feasible region of the virtual power plant on the P-Q coupling plane, that is, the adjustable capacity of the virtual power plant at the point of common coupling.

[0118] Then, the breadth-first search method is used in Step 104 above to perform dimensionality reduction projection on the historical state space to obtain the feasible region of the virtual power plant output power under the received distribution network reconstruction strategy this time, which can be specifically summarized as the following Steps 201 to 204.

[0119] Step 201: Determine the optimization problem as: .

[0120] Step 202: Establish the termination condition as: Less than the threshold .

[0121] Step 203: According to the optimization problem and the termination condition, use the breadth - first search method to perform dimensionality reduction projection on the historical state space to obtain the optimal vertex set.

[0122] Step 204: Calculate the convex hull of the optimal vertex set and determine it as the feasible region of the output power of the virtual power plant under the distribution network reconstruction strategy received this time.

[0123] In another exemplary embodiment of the present application, the reconstruction layer and the evaluation layer interact and cooperate to optimize through the objective functions [see Equations (39) and (40)], and finally determine the optimal network topology.

[0124] (39)

[0125] (40)

[0126] (41)

[0127] (42)

[0128] Among them, represents the objective value when the distribution network reconstruction cost is the smallest and the adjustable capacity is the largest in the evaluation layer; represents the objective value when the distribution network reconstruction cost is the smallest and the adjustable capacity is the largest in the reconstruction layer. and both represent weight coefficients, and ; represents the reconstruction cost, represents the cost of a single line switch action, represents the set of ordinary nodes in the distribution network, and respectively represent the line at time and time the binary variables of the switch state, where represents that the switch on the line is closed, represents that the switch is open; represents the area of the feasible region this time; and respectively represent the minimum cost and the maximum cost of network reconstruction under the initial network topology; and represent the minimum and maximum feasible region areas of the virtual power plant under the initial network topology, respectively. represents the feasible region area corresponding to the larger target value.

[0129] Specifically, the upper layer takes the line connection status as the decision variable to generate a feasible and economical network reconfiguration strategy. Its goal is to continuously optimize the objective function according to the evaluation results feedback from the lower layer, update the reconfiguration strategy, and transfer the new network topology to the evaluation layer of the lower layer. Finally, the optimal network topology is identified from numerous feasible network topologies. In addition, the upper layer records the optimal network topologies obtained by the distribution network under different historical operating states. Then these data are organized into a data set for training the data-driven solution method, aiming to establish a fast mapping mechanism from the operating state of the distribution network to the optimal network topology. To ensure that the reconfigured distribution network still maintains a radial network structure, the radial network structure constraints are formulated as equations (43)-(46).

[0130] (43)

[0131] (44)

[0132] (45)

[0133] (46)

[0134] In the formula, represents a binary variable, represents node is the parent node of node , represents node is not the parent node of node ; represents the set of generator nodes. Equation (43) indicates that the reconfigured network topology still needs to be a tree structure. Equation (44) ensures that in each branch, only one node can be the parent node of another node. Equation (45) restricts each node to have at most one parent node. Equation (46) specifies that the generator node cannot be a parent node.

[0135] Under the network topology given by the reconfiguration layer, the evaluation layer takes the operating state of the distribution network as the input, executes the virtual power plant adjustable capacity evaluation method based on breadth-first search, and obtains the evaluation index while satisfying the node characteristic constraints [see equations (3)-(28)] and the distribution network safe operation constraints [see equations (29)-(35)]. Then the evaluation index is fed back to the reconfiguration layer to compare the advantages and disadvantages of different network reconfiguration strategies, and priority is given to those with a larger Strategy. To sum up, through the iterative interaction between the reconstruction layer and the evaluation layer, the double-layer evaluation architecture finally determines the optimal network reconstruction strategy in the upper reconstruction layer and evaluates the optimal feasible region of the virtual power plant under the optimal network topology in the lower evaluation layer.

[0136] Then, step 105 above can be replaced by the following steps 301 to 302.

[0137] Step 301: Determine the objective function of the evaluation layer as: .

[0138] Step 302: According to the area of the feasible region in this time, using the active power and reactive power of the point of common coupling in the received distribution network reconstruction strategy as decision variables, and using the objective function of the evaluation layer, calculate the objective value when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized.

[0139] In step 106 above, according to the area of the feasible region corresponding to the larger objective value, using the line connection state of the distribution network as the decision variable to obtain the distribution network reconstruction strategy when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized can be replaced by the following steps 401 to 403.

[0140] Step 401: Determine the objective function of the reconstruction layer as: .

[0141] Step 402: Establish the radial network structure constraints as:

[0142] ;

[0143] ;

[0144] ;

[0145] .

[0146] Step 403: According to the area of the feasible region corresponding to the larger objective value, using the line connection state of the distribution network as the decision variable, and using the objective function of the reconstruction layer and the radial network structure constraints, obtain the distribution network reconstruction strategy when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized.

[0147] In another exemplary embodiment of the present application, efficiently identifying the optimal strategy from a large number of feasible network reconstruction strategies is the key to evaluating the optimal feasible region of the virtual power plant. Utilizing the graph structure characteristics of the distribution network, the graph convolutional network can effectively process this graph structure data and extract key spatial dependencies. In addition, by introducing a bidirectional long short-term memory network, the long-term dependencies of the input data can also be captured, thereby more comprehensively describing the reconstruction behavior of the distribution network. Therefore, integrating the graph convolutional network and the bidirectional long short-term memory network, an integrated network is proposed. The integrated network includes a spatially convolutional module, a sequence learning module, and a fully connected layer connected in sequence. Taking the operating state of the distribution network as input data, by extracting spatial dependencies and long-term dependencies from the input data, a mapping relationship between the operating state of the distribution network and the optimal network reconstruction strategy is constructed. The process of using this integrated network to solve the optimal distribution network reconstruction strategy becomes a data-driven fast solution method for the optimal reconstruction strategy.

[0148] (1) Spatially convolutional module.

[0149] The spatially convolutional module consists of multiple graph convolutional network layers. Through multi-layer convolutional operations, the spatially convolutional module gradually obtains the spatial dependencies between the nodes of the distribution network, which helps to explore how network topology changes affect the operating state of the distribution network and the adjustable capacity of the virtual power plant, thereby providing support for determining the network reconstruction strategy. For each graph convolutional network layer, the distribution network is initialized as an undirected weighted graph , where represents the undirected weighted graph, represents the set of nodes in the distribution network, represents the set of power lines, is the adjacency matrix, represents the adjacency matrix is a N × N -dimensional matrix. The admittance matrix of the distribution network is used as the adjacency matrix to effectively characterize the connection strength and mutual influence between nodes. Subsequently, the graph structure data is input into the graph convolutional network, where represents the input feature matrix, is the input feature matrix of the -th layer of the graph convolutional network, which is composed of data related to the operating state of the distribution network, including the power distribution, voltage distribution of the distribution network, the location and operating state of distributed resources connected to the distribution network, represents the dimension of the feature matrix, represents is a N × D -dimensional matrix. The The convolution operation of the layer graph convolutional network can be expressed as Equation (47).

[0150] (47)

[0151] Wherein, , is the identity matrix; is a diagonal matrix, representing the degree matrix of; represents the adjacency matrix after adding self-loops; is the trainable weight matrix of the layer graph convolutional network; is a non-linear activation function.

[0152] Suppose is the output of the spatial convolution module. The convolution operations of multiple graph convolutional network layers can be expressed as Equation (48).

[0153] (48)

[0154] (2) Sequence learning module.

[0155] The sequence learning module consists of multiple bidirectional long short-term memory network layers, aiming to further capture the long-term dependencies of the output of the spatial convolution module, where already contains spatial dependencies. The bidirectional long short-term memory network integrates the forward long short-term memory network and the backward long short-term memory network, and each long short-term memory network consists of a memory cell and three gates. For each long short-term memory network, the input sequence is the sequence data composed of the operation states of the distribution network, and the hidden state represents the corresponding historical optimal reconstruction strategy. The forget gate , the input gate and the output gate can all be improved by the previous hidden state and the current input . All gates will selectively add valuable information to or eliminate irrelevant information from the cell state . In addition, the cell state will also combine the information learned from the previous state and . Subsequently, the current optimal reconstruction strategy and It will be periodically fed into the next layer. With the bidirectional structure, the sequence learning module can extract both forward and backward information from the input data. Specifically, the sequence learning module can not only utilize the forward data to analyze the mapping relationship between the operation state of the distribution network and the optimal reconstruction strategy, but also improve the mapping mechanism through the backward data. This bidirectional ability can enhance the capture ability of long-term dependencies, as shown in equations (49)-(57) below.

[0156] (49)

[0157] (50)

[0158] (51)

[0159] (52)

[0160] (53)

[0161] (54)

[0162] (55)

[0163] (56)

[0164] (57)

[0165] In the formula, represents the candidate memory unit; , , and are the weight matrices corresponding to each gate; , , and are the bias matrices; and represent the outputs of the forward long short-term memory network and the backward long short-term memory network respectively; LSTM represents the steps defined by equations (49)-(54); , and are the trainable weight matrices and bias matrices when the forward and backward long short-term memory networks are combined.

[0166] (3) Fully connected layer.

[0167] The hidden state of the last bidirectional long short-term memory network layer is represented as , as the output of the sequence learning module, and is fed to the fully connected layer. At this stage, the spatial dependency and long-term dependency of the input data are extracted and integrated by the spatial convolution module and the sequence learning module respectively. A binary sequence is generated using the fully connected layer, which represents the control decision of the branch switch, thereby converting the spatial dependency and long-term dependency of the input data into a binary sequence. Mapping to the optimal network reconstruction strategy , as shown in formula (58). Subsequently, the obtained optimal network topology is passed to the lower layer of the two-layer evaluation architecture, and the virtual power plant adjustable capacity evaluation method based on breadth-first search is executed in the lower layer to determine the feasible domain with the largest area of ​​the virtual power plant, thereby completing the evaluation of the optimal adjustable capacity of the virtual power plant.

[0168] (58)

[0169] In the formula, and They represent the trainable weight matrix and bias matrix of the fully connected layer respectively. represents a non-linear activation function.

[0170] The key points of this application are as follows.

[0171] 1. A method for evaluating the adjustable capacity of virtual power plants based on breadth-first search is proposed: first, the concept of the feasible domain representing the adjustable capacity of virtual power plants is defined, and then the feasible domain of virtual power plants on the PQ coupling plane, i.e., the adjustable capacity of virtual power plants, is evaluated under a fixed topology by considering both node characteristic constraints and network topology constraints.

[0172] 2. A two-layer virtual power plant optimal feasible domain evaluation framework considering network reconstruction is proposed to explore the optimal adjustable capacity of virtual power plants under dynamic network topology. Through continuous interactive iteration between the upper and lower layers, the reconstruction strategy is continuously updated, and the virtual power plant adjustable capacity evaluation method based on breadth-first search is used to calculate the feasible domain of the virtual power plant under the current topology until the optimal feasible domain is found.

[0173] 3. A data-driven optimal reconstruction strategy fast solution method is proposed. The spatial dependency and long-term dependency of the input feature data are extracted through a graph convolutional network and a bidirectional long short-term memory network, respectively, to achieve a fast mapping from the distribution network operation state to its optimal reconstruction strategy, so as to efficiently obtain the optimal network reconstruction strategy corresponding to the optimal regulation capacity of the virtual power plant.

[0174] When considering both node characteristics and network structure, accurately evaluating the adjustable capacity of a virtual power plant is a very complex problem. Traditional evaluation methods either cannot consider network topology constraints during the evaluation process or cannot balance evaluation accuracy and computational efficiency. Unreasonable network structures are more likely to cause problems such as line overload and voltage violation during the aggregation of distributed resources, making it difficult for distributed resources to fully exert their adjustable potential. Therefore, network reconfiguration is considered to be introduced to improve the maximum adjustable capacity of the virtual power plant. The network reconfiguration problem is essentially a non-linear integer programming problem and is difficult to solve directly. Traditional solution methods either cannot solve complex networks or are difficult to guarantee the quality of the solution. The above problems pose higher requirements for evaluating the adjustable capacity of a virtual power plant considering node characteristics and network structure.

[0175] Aiming at the defects and difficulties existing in the evaluation process of the optimal adjustable capacity of the above virtual power plant, this application can quickly calculate the optimal adjustable capacity of the virtual power plant in a dynamic network structure. The proposed method for evaluating the adjustable capacity of a virtual power plant based on breadth-first search can not only consider node characteristics and network topology constraints during the evaluation process, but also ensure accurate solution of the adjustable capacity of the virtual power plant in a shorter time. Then, considering the potential limitation of network topology constraints on the adjustable capacity of the virtual power plant, a two-layer evaluation architecture for the optimal feasible region of the virtual power plant considering network reconfiguration is proposed to deeply explore the adjustment potential of distributed resources and seek the optimal adjustable capacity of the virtual power plant under dynamic network topologies. Finally, a data-driven fast solution method for the optimal reconfiguration strategy is proposed. By using a graph convolutional network and a bidirectional long short-term memory network to extract the spatial dependence and long-term dependence of input data respectively, the mapping from the operating state of the distribution network to its optimal reconfiguration strategy is realized, and the optimal network reconfiguration strategy corresponding to the optimal adjustable capacity of the virtual power plant is efficiently obtained, so as to realize the fast evaluation of the optimal adjustable capacity of the virtual power plant.

[0176] Based on the same inventive concept, the embodiment of this application also provides a device for quickly evaluating the optimal adjustable capacity of a virtual power plant for implementing the above-mentioned method for quickly evaluating the optimal adjustable capacity of a virtual power plant. The implementation solutions provided by this device to solve problems are similar to those recorded in the above method. Therefore, the specific limitations in one or more embodiments of the device for quickly evaluating the optimal adjustable capacity of a virtual power plant provided below can refer to the limitations on the method for quickly evaluating the optimal adjustable capacity of a virtual power plant in the above text and will not be elaborated here.

[0177] In an exemplary embodiment, a device for quickly evaluating the optimal adjustable capacity of a virtual power plant includes: a building module, a policy generation module, a state space construction module, a dimensionality reduction projection module, an area calculation module, a policy update module, a training module, a model application module, a constraint integration module, and an optimal adjustable capacity determination module.

[0178] The building module is used to build a two-layer evaluation architecture including a reconstruction layer and an evaluation layer. The policy generation module is used to generate a distribution network reconstruction policy for the distribution network in each historical operating state in the reconstruction layer and transmit it to the evaluation layer. The state space construction module is used to jointly construct a historical state space with the node characteristic constraints and network topology constraints of the distributed resources in the virtual power plant under the distribution network reconstruction policy received this time in the evaluation layer. The dimensionality reduction projection module is used to perform dimensionality reduction projection on the historical state space by using the breadth-first search method to obtain the feasible region of the output power of the virtual power plant under the distribution network reconstruction policy received this time, and calculate the area of the feasible region this time.

[0179] The area calculation module is used to calculate the target value when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized with the output power of the point of common coupling in the received distribution network reconstruction policy as the decision variable according to the area of the feasible region this time, and transmit both the target value this time and the area of the feasible region this time to the reconstruction layer.

[0180] The policy update module is used to compare the target value received this time with the target value received last time in the reconstruction layer, and obtain the distribution network reconstruction policy when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized with the line connection state of the distribution network as the decision variable according to the area of the feasible region with the larger target value, and transmit it to the evaluation layer, and call the state space construction module until the target values received in two adjacent times are equal, and determine the latest distribution network reconstruction policy in the reconstruction layer as the optimal distribution network reconstruction policy in each historical operating state.

[0181] The training module is used to train an integrated network of a graph convolutional network and a bidirectional long short-term memory network with each historical operating state as the input and the optimal distribution network reconstruction policy in each historical operating state as the label to obtain a policy solution model.

[0182] The model application module is used to obtain the optimal distribution network reconstruction policy of the target distribution network by using the policy solution model according to the current operating state of the target distribution network in the reconstruction layer and transmit it to the evaluation layer.

[0183] A constraint integration module is used to jointly constitute the current state space with the node characteristic constraints and network topology constraints of distributed resources in a virtual power plant under the optimal distribution network reconstruction strategy of a target distribution network. An optimal adjustable capacity determination module is used to perform dimensionality reduction projection on the current state space by using a breadth-first search method in the evaluation layer, obtain the feasible region of the output power of the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network, and calculate the area of the feasible region of the target distribution network, and transmit it to the reconstruction layer; the area of the feasible region of the target distribution network represents the optimal adjustable capacity of the virtual power plant.

[0184] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0185] Specific examples are used in this article to elaborate on the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for rapidly evaluating the optimal adjustable capacity of a virtual power plant, characterized in that: include: Establish a two-layer evaluation framework including reconstruction layer and evaluation layer; A distribution network reconstruction strategy is generated in the reconstruction layer under each historical operation state of the distribution network, and transmitted to the evaluation layer; At the evaluation layer, the node characteristic constraints and network topology constraints of the distributed resources in the virtual power plant under the distribution network reconstruction strategy received this time are combined to form a historical state space; The breadth-first search method is used to reduce the dimension of the historical state space, obtain the feasible domain of the virtual power plant output power under the distribution network reconstruction strategy received this time, and calculate the area of ​​the feasible domain this time; According to the feasible area of ​​this time, the output power of the common coupling point in the received distribution network reconstruction strategy is used as the decision variable to calculate the target value that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity, and transmit the target value and the feasible area of ​​this time to the reconstruction layer; In the reconstruction layer, the larger target value between the target value received this time and the target value received last time is selected, and according to the feasible domain area corresponding to the larger target value, the line connection state of the distribution network is used as the decision variable to obtain the distribution network reconstruction strategy that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity, and transmit it to the evaluation layer, and return to execute the step of forming a historical state space of the node characteristic constraints and network topology constraints of the distributed resources in the virtual power plant under the distribution network reconstruction strategy received this time in the evaluation layer, until the target values ​​received twice adjacently are equal, and the latest distribution network reconstruction strategy in the reconstruction layer is determined as the optimal distribution network reconstruction strategy under each historical operating state; Taking each historical operating state as input and the optimal distribution network reconstruction strategy under each historical operating state as a label, an integrated network of graph convolutional network and bidirectional long short-term memory network is trained to obtain a strategy solution model; In the reconstruction layer, according to the current operating state of the target distribution network, the optimal distribution network reconstruction strategy of the target distribution network is obtained by using the strategy solution model, and transmitted to the evaluation layer; The node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network together constitute the current state space; In the evaluation layer, the breadth-first search method is used to perform dimensionality reduction projection on the current state space to obtain the feasible domain of the virtual power plant output power under the optimal distribution network reconstruction strategy of the target distribution network, and the feasible domain area of ​​the target distribution network is calculated and transmitted to the reconstruction layer; the feasible domain area of ​​the target distribution network represents the optimal adjustable capacity of the virtual power plant.

2. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: The node characteristic constraints include: photovoltaic node operation characteristics, energy storage operation characteristics, electric vehicle operation characteristics and temperature control load operation characteristics; The operating characteristics of the photovoltaic node are: ； ； ； In the formula, , and Respectively represent nodes Photovoltaic Active power, reactive power and grid-connected inverter capacity generated at all times; and Respectively represent nodes The upper and lower limits of the active power of photovoltaic power; and Respectively represent nodes The maximum power factor angle and the minimum power factor angle; The energy storage operation characteristics are: ； ； ； ； ； In the formula, , and Respectively represent nodes Energy storage in Active power, reactive power and grid-connected inverter capacity at the moment; and Representation Node Energy storage in The charging and discharging power at each moment; , Respectively represent nodes The maximum charging power and minimum charging power of the energy storage; , Respectively represent nodes The maximum discharge power and minimum discharge power of the energy storage; Representation Node Energy storage in The state of charge at the moment, and Respectively represent nodes The upper and lower limits of the state of charge of the energy storage; and Respectively represent the charge and discharge efficiency; Indicates the time interval of the charging / discharging process; Representation Node Energy storage in The state of charge at the moment; The operating characteristics of electric vehicles are: ； ； ； ； ； ； ； ； ； In the formula, Representation Node of electric vehicles in The energy state at any moment, and Representation Node of electric vehicles in The upper and lower limits of energy state at each moment; Representation Node The maximum energy state technically allowed for electric vehicles, and Representation Node The energy status of electric vehicles during the time of grid connection and off-grid connection; and They represent the grid-connected time and off-grid time of electric vehicles respectively; Representation Node The maximum charging power technically allowed for electric vehicles; Representation Node The technically permissible charging efficiency of electric vehicles; Representation Node of electric vehicles in The actual charging power at the moment, and Respectively represent nodes of electric vehicles in Maximum charging power and minimum charging power at the moment; and Respectively represent nodes of electric vehicles in The upper and lower limits of energy state at each moment; Representation Node of electric vehicles in The energy state at the moment; and Respectively represent nodes The remaining energy of electric vehicles when connected to the grid and the minimum energy required when off-grid; The operating characteristics of temperature control load are: ； ； ； ； In the formula, Representation Node The temperature control load is The state of charge at the moment, Representation Node The temperature control load is The state of charge at the moment; Representation Node The temperature control load is The cooling power at the time, Representation Node The temperature control load is Refrigeration power at the time; Representation Node exist The outdoor ambient temperature at the moment; , , , All represent characteristic parameters of the equivalent energy storage model of the temperature control load; and They represent equivalent thermal resistance and equivalent heat capacity respectively; It indicates the refrigeration coefficient; and Respectively represent nodes the maximum and minimum temperatures; and Respectively represent nodes The minimum cooling power and maximum cooling power of the temperature control load; Representation Node exist The indoor temperature at the moment.

3. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: The network topology constraints include: power flow balance constraints, voltage safety constraints and line capacity constraints; The power flow balance constraint is: ； ； ； ； In the formula, and Respectively represent nodes exist Active power and reactive power generated at all times; and Representation Node exist Active load and reactive load at the moment; and Indicated in Timeline Active power flow and reactive power flow on the and Node and nodes The resistance and reactance between Indicated in Timeline The square of the upper current; and Indicated in Timeline Active power flow and reactive power flow on the and Respectively represent nodes The collection of parent nodes and child nodes; Indicated in Time Node The voltage at Indicated in Time Node The voltage at represents the L2 norm; The voltage safety constraint is: ； In the formula, and Representation Node exist The upper and lower limits of voltage at all times; The line capacity constraint is: ； ； In the formula, and Respectively The active power and reactive power of the virtual power plant at the common coupling point at each moment; Indicates line The line capacity limit on Represents the line capacity limit at the virtual power plant common coupling point.

4. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: The state space is represented as: ； In the formula, represents the state space; represents the projection variable, , and They represent the active power and reactive power of the virtual power plant at the common coupling point respectively; Represents internal decision variables and state variables; , All represent coefficient matrices; represents a constant vector; Represents 2-dimensional real number space, referring to is a two-dimensional vector; express dimensional real number space, referring to is a dimensional vector.

5. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: The breadth-first search method is used to reduce the dimension of the historical state space and obtain the feasible domain of the virtual power plant output power under the distribution network reconstruction strategy received this time, including: The optimization problem is determined as: ; In the formula, Indicates along The width of the movement when searching for vertices; represents the unit outward normal vector of the tangent plane searching for new vertices, , and Respectively represent the direction vectors on the P axis and Q axis; express Nodes in the feasible region at time , at this time node Income during vertex search The vertex of represents a vertex set, , and Respectively In the feasible domain of virtual power plant Active and reactive power at the point of common coupling; represents the feasible domain of the output power of the virtual power plant; represents the projection variable; The termination condition is established as: Less than threshold ;in, Indicates the relative motion width; According to the optimization problem and the termination condition, a breadth-first search method is used to perform dimension reduction projection on the historical state space to obtain an optimal vertex set; The convex hull of the optimal vertex set is calculated and determined as the feasible domain of the virtual power plant output power under the distribution network reconstruction strategy received this time.

6. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: The calculation formula of the feasible region area is: ； In the formula, represents the feasible region area; and Respectively Moment and In the feasible domain of virtual power plant Active power at the point of common coupling, express Nodes in the feasible region at time , at this time node Income during vertex search The vertex of represents a vertex set, ; and Respectively Moment and In the feasible domain of virtual power plant Reactive power at the point of common coupling; det represents the determinant of the matrix.

7. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: According to the feasible area of ​​this time, the output power of the common coupling point in the received distribution network reconstruction strategy is used as the decision variable to calculate the target value that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity, including: The objective function of the evaluation layer is determined as: ; In the formula, It represents the target value when the distribution network reconstruction cost is the minimum and the adjustable capacity is the maximum in the evaluation layer; and are weight coefficients, and ; represents the reconstruction cost, , represents the cost of a single line switch operation, represents the set of common nodes in the distribution network, and Respectively indicate lines exist Moment and A binary variable indicating the switch state at the moment, where Indicates line The switch on is closed, Indicates the switch is on; represents the feasible region area of ​​this time; and They represent the minimum cost and maximum cost of network reconstruction under the initial network topology respectively; and They represent the minimum feasible area and the maximum feasible area of ​​the virtual power plant under the initial network topology respectively; and They represent the feasible domain of the virtual power plant. Active and reactive power at the point of common coupling, express Nodes in the feasible region at time , at this time node Income during vertex search The vertex of represents a vertex set, ; According to the feasible region area, the active power and reactive power of the common coupling point in the received distribution network reconstruction strategy are taken as decision variables, and the objective function of the evaluation layer is used to calculate the target value that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity.

8. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: According to the feasible region area corresponding to the larger target value, the line connection state of the distribution network is used as the decision variable to obtain the distribution network reconstruction strategy that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity, which specifically includes: The objective function of the reconstruction layer is determined as: ; In the formula, It represents the target value when the distribution network reconstruction cost is minimum and the adjustable capacity is maximum in the reconstruction layer; and are weight coefficients, and ; represents the reconstruction cost, , represents the cost of a single line switch operation, represents the set of common nodes in the distribution network, and Respectively indicate lines exist Moment and A binary variable indicating the switch state at the moment, where Indicates line The switch on is closed, Indicates the switch is on; represents the feasible region area of ​​this time; and They represent the minimum cost and maximum cost of network reconstruction under the initial network topology respectively; and They represent the minimum feasible area and the maximum feasible area of ​​the virtual power plant under the initial network topology respectively; Indicates the feasible region area corresponding to the larger target value; The constraints for establishing a radial network structure are: ； ； ； ； In the formula, represents a binary variable, Representation Node Is a node The parent node of Representation Node Not a node The parent node of represents the set of generator nodes; According to the feasible region area corresponding to the larger target value, the line connection status of the distribution network is taken as the decision variable, and the objective function of the reconstruction layer and the radial network structure constraints are utilized to obtain the distribution network reconstruction strategy that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity.

9. The method for rapidly evaluating the optimal adjustable capacity of a virtual power plant according to claim 1, characterized in that: The integrated network includes a spatial convolution module, a sequence learning module and a fully connected layer connected in sequence; The spatial convolution module includes a plurality of graph convolution network layers connected in sequence; the spatial convolution module is used to obtain the spatial dependency relationship between the nodes of the distribution network and output the spatial dependency features; The sequence learning module includes a plurality of bidirectional long short-term memory network layers connected in sequence; the sequence learning module is used to capture the long-term dependency relationship of the spatial dependency features output by the spatial convolution module, and output the spatial and long-term dependency features; The fully connected layer is used to calculate the spatial and long-term dependency features using the formula , output the optimal distribution network reconstruction strategy; where, represents the optimal distribution network reconstruction strategy, represents a nonlinear activation function, Represents spatial and long-term dependency features, and They represent the trainable weight matrix and bias matrix of the fully connected layer respectively.

10. A device for rapidly evaluating the optimal adjustable capacity of a virtual power plant, characterized in that: The virtual power plant optimal adjustable capacity rapid assessment device comprises: Establishing a module for establishing a two-layer evaluation framework including a reconstruction layer and an evaluation layer; A strategy generation module is used to generate a distribution network reconstruction strategy for the distribution network under each historical operating state in the reconstruction layer and transmit it to the evaluation layer; The state space construction module is used to construct a historical state space by combining the node characteristic constraints and network topology constraints of the distributed resources in the virtual power plant under the distribution network reconstruction strategy received this time at the evaluation layer; The dimension reduction projection module is used to perform dimension reduction projection on the historical state space using a breadth-first search method, obtain the feasible domain of the virtual power plant output power under the distribution network reconstruction strategy received this time, and calculate the area of ​​the feasible domain this time; The area calculation module is used to calculate the target value when the distribution network reconstruction cost is minimized and the adjustable capacity is maximized according to the feasible domain area of ​​this time and the output power of the common coupling point in the received distribution network reconstruction strategy as the decision variable, and transmit the target value and the feasible domain area of ​​this time to the reconstruction layer; The strategy updating module is used to select the larger target value between the target value received this time and the target value received last time in the reconstruction layer, and according to the feasible domain area corresponding to the larger target value, take the line connection state of the distribution network as the decision variable, obtain the distribution network reconstruction strategy that minimizes the distribution network reconstruction cost and maximizes the adjustable capacity, and transmit it to the evaluation layer, call the state space composition module, until the target values ​​received twice adjacently are equal, and determine the latest distribution network reconstruction strategy in the reconstruction layer as the optimal distribution network reconstruction strategy under each historical operating state; A training module is used to take each historical operating state as input and the optimal distribution network reconstruction strategy under each historical operating state as a label, train an integrated network of a graph convolutional network and a bidirectional long short-term memory network, and obtain a strategy solution model; A model application module is used to obtain the optimal distribution network reconstruction strategy of the target distribution network by using the strategy solving model according to the current operating state of the target distribution network in the reconstruction layer, and transmit it to the evaluation layer; A constraint integration module is used to combine the node characteristic constraints and network topology constraints of distributed resources in the virtual power plant under the optimal distribution network reconstruction strategy of the target distribution network to form the current state space; The optimal adjustable capacity determination module is used to use the breadth-first search method in the evaluation layer to perform dimensionality reduction projection on the current state space, obtain the feasible domain of the virtual power plant output power under the optimal distribution network reconstruction strategy of the target distribution network, and calculate the feasible domain area of ​​the target distribution network, and transmit it to the reconstruction layer; the feasible domain area of ​​the target distribution network represents the optimal adjustable capacity of the virtual power plant.

Images