Power distribution network multi-level management and control method and device based on mechanism modeling and data driving cooperation

Abstract

The application discloses a kind of mechanism modeling and data-driven collaborative multi-level control method and device of distribution network, method includes: in medium voltage distribution network, with total operating cost and voltage deviation minimum as target to build long time scale first regulation and control model;Solving first regulation and control model obtains the first regulation and control instruction of passive equipment in medium voltage distribution network;After implementing first regulation and control instruction, each public node power is collected;In each low voltage distribution network, with minimum network loss, voltage deviation and operation and maintenance cost as target, based on multi-agent principle, short time scale second regulation and control model is built;Second regulation and control model is trained based on multi-agent deep deterministic policy gradient algorithm, and third regulation and control model is obtained;According to each public node power and each third regulation and control model, the second regulation and control instruction of photovoltaic and energy storage in each low voltage distribution network is obtained, second regulation and control instruction is implemented, and regulation and control are completed.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a multi-level control method and device for power distribution networks that combines mechanism modeling and data-driven collaboration. Background Technology

[0002] In recent years, with the construction of new power systems, distributed resources, represented by photovoltaics, have been widely integrated into distribution networks. While promoting the low-carbon transformation of the energy structure, this has also profoundly changed the operating characteristics of distribution networks. For medium-voltage distribution networks, the integration of distributed resources significantly increases the complexity of system power flow distribution, leading to problems such as increased node voltage fluctuations and network losses, posing challenges to the safe and economical operation of the system. For low-voltage distribution networks, due to the highly dispersed and dynamic nature of distributed resources on the user side, problems such as voltage exceeding limits and local overloads may occur. Therefore, it is urgent to develop collaborative optimization and control technologies for medium- and low-voltage distribution networks that are adapted to distributed resources, in order to achieve a balance between system operational safety and economy.

[0003] Existing methods for optimizing and controlling distribution networks mostly focus on a single voltage level. For medium-voltage distribution networks with known architectures, optimization is primarily achieved using mathematical programming methods based on mechanistic models. However, medium-voltage control cannot precisely regulate low-voltage distribution networks, nor can it directly address objectives such as voltage quality, equipment load rate, and distributed energy consumption in low-voltage networks. Furthermore, for low-voltage distribution networks, most low-voltage control optimization objectives are often limited to local levels, lacking coordination with the medium-voltage level and failing to achieve global optimization. Especially when the physical architecture of the low-voltage side is unknown, traditional methods based on precise modeling are difficult to apply, posing a greater challenge to constructing effective control strategies. In addition, while there are some research results on the coordinated optimization of medium and low voltage systems, they are often limited by the difficulty of coordinated modeling and employ only a single time-scale control framework, making it difficult to simultaneously meet the needs of short-term optimization at the medium-voltage level and rapid real-time response at the low-voltage level, resulting in poor coordinated performance.

[0004] Therefore, a new technical solution is urgently needed to address the technical problem of how to optimize and control medium and low voltage distribution networks across multiple time scales. Summary of the Invention

[0005] This invention provides a multi-level control method and device for distribution networks that combines mechanism modeling and data-driven collaboration, in order to solve the technical problem of how to perform multi-timescale optimization and control of medium and low voltage distribution networks.

[0006] To achieve the above objectives, this invention provides a multi-level control method for distribution networks that combines mechanism modeling and data-driven collaboration. In this distribution network, the low-voltage distribution network architecture is unknown, while the medium-voltage distribution network architecture is known. The method includes:

[0007] In a medium-voltage distribution network, a first control model with a long time scale is constructed with the objectives of minimizing total operating cost and voltage deviation. The first control model is solved to obtain the first control command for passive equipment in the medium-voltage distribution network. After implementing the first control command, the power of each common node is collected.

[0008] In each low-voltage distribution network, with the goal of minimizing network loss, voltage deviation, and operation and maintenance costs, a second control model with a short time scale is constructed based on the multi-agent principle. The second control model is trained based on the multi-agent deep deterministic policy gradient algorithm to obtain a third control model. The second control commands for photovoltaic and energy storage in each low-voltage distribution network are obtained based on the power of each common node and each third control model. The second control commands are then implemented to complete the control.

[0009] Preferably, the first regulation model constructed on a long-term timescale with the goal of minimizing total operating cost and voltage deviation includes:

[0010] In medium-voltage distribution networks, an objective function is constructed with the goal of minimizing total operating cost and voltage deviation. Total operating cost includes network loss cost and electricity purchase cost. Based on the objective function and preset constraints, a first control model for a long time scale is constructed. Preset constraints include passive equipment constraints and operational constraints. Passive equipment constraints include on-load tap-changing transformer constraints and capacitor bank constraints. Operational constraints include network operation safety constraints and network power flow constraints.

[0011] Preferably, the first control command obtained by solving the first control model for passive equipment in the medium-voltage distribution network includes:

[0012] The first parameter set of each node in the medium-voltage distribution network is collected; the first parameter set includes active load, reactive load and impedance information of each branch; the first control model is solved according to the first parameter set of each node with a long time scale to obtain the first control command; the first control command includes the command to change the number of taps of the on-load tap-changing transformer and the command to switch the capacitor bank.

[0013] Preferably, a second control model with a short timescale, based on the multi-agent principle, is constructed with the objectives of minimizing network loss, voltage deviation, and operation and maintenance costs. This model includes:

[0014] In each low-voltage distribution network, a reward function is constructed based on network loss, voltage deviation, and operation and maintenance costs. The operation and maintenance costs include those of photovoltaic and energy storage systems. Each photovoltaic system and energy storage system in the low-voltage distribution network is treated as an intelligent agent. Based on the reward function and the intelligent agent, combined with the multi-agent principle, a second control model with a short time scale is obtained.

[0015] In the second regulation model, the photovoltaic agent is used to control the photovoltaic output of the node; the energy storage agent is used to select the node charging and discharging strategy according to the grid load demand, battery status and voltage changes.

[0016] Preferably, the third regulation model is obtained by training the second regulation model based on the multi-agent deep deterministic policy gradient algorithm, including:

[0017] The state space of the agent is constructed using a second set of parameters; the second set of parameters includes the node voltage state, the node active and reactive power, the output power of the photovoltaic unit, and the charging state of the energy storage unit.

[0018] The initial values ​​of each parameter in the second parameter set are collected; the initial values ​​are configured into the second control model, and the agent in the second control model interacts with the environment according to the state space, selects an action from the preset action space, and obtains a reward and the next state; the reward is obtained according to the reward function; the environment includes the current state;

[0019] The current state, action, reward, and next state are stored as a set of data in the experience pool. Each agent includes an independent algorithm network, which includes a target network and a value network. A set of data is sampled from the experience pool to update the parameters of the algorithm network. The agents in the second regulation model interact with the environment repeatedly, and the parameters of the algorithm network are updated cyclically until the preset number of iterations is reached, at which point the loop ends, and the third regulation model is obtained.

[0020] Preferably, selecting actions from a preset action space includes:

[0021] When the intelligent agent is a photovoltaic intelligent agent, the photovoltaic output adjustment target is selected from the preset photovoltaic intelligent agent action space; the photovoltaic intelligent agent action space includes the adjustable range of photovoltaic power generation.

[0022] When the intelligent agent is an energy storage intelligent agent, the energy storage charging or discharging adjustment target is selected from the preset energy storage intelligent agent action space; the energy storage intelligent agent action space includes the adjustable range of the energy storage system's charging and discharging power.

[0023] Preferably, the second control commands for photovoltaic and energy storage in each low-voltage distribution network, obtained based on the power of each common node and each third control model, include:

[0024] In each low-voltage distribution network, the power of the corresponding common node is configured as the total demand for the operation of the low-voltage distribution network in the corresponding third control model with a short time scale. The third model outputs the second control command for photovoltaic and energy storage.

[0025] The present invention also provides a multi-level control device for power distribution networks that combines mechanism modeling and data-driven collaboration, for use with the method of the present invention. The device includes a first module, a second module, a third module and a fourth module.

[0026] The first module is used to construct a first control model on a long time scale in a medium-voltage distribution network with the goal of minimizing total operating cost and voltage deviation; and to solve the first control model to obtain the first control command for passive equipment in the medium-voltage distribution network.

[0027] The second module is used to collect the power of each common node after the first control command is implemented;

[0028] The third module is used to construct a second control model with a short time scale in each low-voltage distribution network with the goal of minimizing network loss, voltage deviation and operation and maintenance costs based on the multi-agent principle; the second control model is trained based on the multi-agent deep deterministic policy gradient algorithm to obtain the third control model.

[0029] The fourth module is used to obtain the second control commands for photovoltaic and energy storage in each low-voltage distribution network based on the power of each common node and each third control model, implement the second control commands, and complete the control.

[0030] The present invention has the following beneficial effects:

[0031] The mechanism modeling and data-driven collaborative multi-level control method for distribution networks of this invention can solve the problems of difficult collaborative modeling of medium and low voltage distribution networks and unknown low-voltage layer architecture. It can achieve collaborative optimization of voltage quality in medium and low voltage distribution networks, reduce system voltage deviation, prevent voltage overruns, improve the absorption capacity of renewable energy, and reduce the total operating cost of the system. It provides a multi-timescale optimized control scheme that balances economy and stability for medium and low voltage distribution networks with high proportion of renewable energy. Based on the method of this invention, it can solve the problem of joint control of medium and low voltage under unknown low-voltage architecture and the problem of slow convergence speed and difficulty in coordinating conflicting objectives of multiple medium and low voltage devices in complex distribution networks using single deep reinforcement learning, thus eliminating node voltage overruns.

[0032] The mechanism modeling and data-driven collaborative multi-level control device for power distribution networks of the present invention, when used in the method of the present invention, has the same beneficial effects as the method of the present invention.

[0033] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0034] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0035] Figure 1 This is a schematic diagram of a medium- and low-voltage distribution network topology according to a preferred embodiment of the present invention.

[0036] Figure 2 This is a schematic diagram of the method flow of a preferred embodiment of the present invention.

[0037] Figure 3 This is a schematic diagram of the network topology of a preferred embodiment of the present invention.

[0038] Figure 4 This is a schematic diagram of the reward change curve for the intelligent agent training according to a preferred embodiment of the present invention.

[0039] Figure 5 This is a schematic diagram comparing the photovoltaic output under two schemes of a preferred embodiment of the present invention.

[0040] Figure 6 This is a schematic diagram comparing the energy storage output under two schemes of a preferred embodiment of the present invention.

[0041] Figure 7 This is a schematic diagram of the voltage distribution under this preferred embodiment of the present invention.

[0042] Figure 8 This is a schematic diagram of the voltage distribution under a comparative embodiment of a preferred embodiment of the present invention. Detailed Implementation

[0043] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0044] See Figures 1 to 2 In a preferred embodiment of the present invention, a multi-level control method for distribution networks based on mechanistic modeling and data-driven collaboration is provided. The distribution network includes a low-voltage distribution network and a medium-voltage distribution network. In the low-voltage and medium-voltage distribution networks, the architecture of the low-voltage distribution network is unknown, but the nodes of the installed distributed photovoltaic and energy storage systems have measurement capabilities. The architecture of the medium-voltage distribution network is known, and the installed passive devices include OLTC (on-load tap changer) and CB (capacitor bank). The method includes:

[0045] S1. In medium-voltage distribution networks, a first-time-scale control model is constructed with the objectives of minimizing total operating cost and voltage deviation. S1 specifically includes:

[0046] In medium-voltage distribution networks, an objective function is constructed with the goal of minimizing total operating cost and voltage deviation. Total operating cost includes network loss cost and electricity purchase cost. Based on the objective function and preset constraints, a first control model for a long time scale is constructed. Preset constraints include passive equipment constraints and operational constraints. Passive equipment constraints include on-load tap-changing transformer constraints and capacitor bank constraints. Operational constraints include network operation safety constraints and network power flow constraints.

[0047] In a preferred embodiment of the present invention, the objective function is specifically expressed as:

[0048] ;

[0049] ;

[0050] in, For total cost, For network loss costs, For medium voltage distribution network Network loss power of nodes For electricity purchase costs, To purchase electricity, It is the total number of distribution network nodes. Representative node Voltage deviation, Represents the set of all nodes in the distribution network. Representative node exist The node voltage at time t, and Represents the lower and upper limits of voltage; definition ,when hour, .

[0051] In a preferred embodiment of the present invention, the preset constraints specifically include:

[0052] (1) On-load tap-changing transformer constraints: In order to describe the effect of the turns ratio on voltage regulation, virtual nodes are introduced. To represent the impact of OLTC equipment. During regulation, the OLTC controls the voltage change at the node by changing its turns ratio. The effect of the turns ratio can be expressed by the following relationship:

[0053] ;

[0054] in, It is a node voltage, The rated voltage of the system. It is the first The ratio of each virtual node It is the reference ratio, and the adjustment node is changed by altering the ratio. To stabilize the grid voltage, the following constraints are imposed in practice: (e.g., limits on the number of adjustments and discretized adjustment step size). Further constraints can be imposed as follows:

[0055] ;

[0056] in, Indicates the node where OLTC is located; and Represented as a binary variable for OLTC gear shifting, when This means the OLTC gear increases by 1. This means the OLTC gear position is reduced by 1. A collection representing time; Represents the adjustment variable for adjacent gears; This represents the maximum number of times the OLTC gear can be adjusted within the total time period; Let be the ratio of OLTC at time t; This represents the lower limit of the OLTC ratio at time t; Represents variables with identifiers of 0-1; This represents the total gear adjustment amount at time t; This represents the maximum range of gear adjustment.

[0057] (2) Capacitor Bank Constraints: The capacitor bank (CB) adjusts the grid voltage by switching capacitor banks to change the reactive power in the grid. When the grid voltage is low, the CB increases reactive power and raises the voltage by switching capacitor banks; when the grid voltage is high, the CB reduces reactive power and suppresses excessive voltage by disconnecting capacitor banks. Compensation Power Represented as:

[0058] ;

[0059] in, Represents the set of nodes containing CB; Represents the number of CB units in operation; This represents the compensation power for each CB group.

[0060] Considering constraints such as the limited number of connection groups at each node and the limit on the total number of adjustments in practice, auxiliary variables are added. This represents the change of CB within adjacent time periods, and can be further constrained as follows:

[0061] ;

[0062] in, For connecting nodes The maximum number of CB groups on the top; This represents the maximum number of operations that can be performed on a CB.

[0063] (3) Operational constraints: Operational constraints include network operation security constraints and network power flow constraints, among which network operation security constraints can be expressed as:

[0064] ;

[0065] in, Voltage amplitude; and These are the lower and upper bounds of the node voltage; Power amplitude; and These are the lower and upper bounds of the power. For nodes With nodes Current amplitude between; This is the upper bound of the node current; This represents the set of all nodes.

[0066] Considering that the power flow constraints of the distribution network are non-convex, this paper employs a two-step relaxation process—phase angle relaxation and second-order cone relaxation—to improve solution efficiency and optimality. Let... , The relaxed power flow constraint is obtained as follows:

[0067] ;

[0068] in, , For nodes Injected active and reactive power, Represents the node to downstream nodes Total active power Represents the node to downstream nodes Total reactive power Represents the node Finally injected into the node Total active power Represents the node Finally injected into the node The total reactive power , For nodes To the node The resistance and reactance information are collected in S2. , For nodes and nodes The node voltage; , For nodes The active and reactive power of the load, i.e., the active and reactive loads collected in S2. , For nodes The active and reactive power generated by the generator; Indicates from node Flow to node The active power; Indicates from node Flow to node The reactive power.

[0069] S2. Solve the first control model to obtain the first control command for passive equipment in the medium-voltage distribution network; after implementing the first control command, collect the power of each common node.

[0070] In a preferred embodiment of the present invention, the first control command obtained by solving the first control model for passive equipment in a medium-voltage distribution network includes:

[0071] The system collects the first parameter set of each node in the medium-voltage distribution network. This first parameter set includes active load, reactive load, and impedance information of each branch. Using a long time scale (e.g., 1 hour), the system solves the first control model using the Cplex commercial solver based on the first parameter set of each node to obtain the first control command. The first parameter set is then substituted with relaxed power flow constraints. The first control command includes commands to change the number of taps on-load tap-changing transformers and commands to switch capacitor banks. If the capacitor bank switching command is 0, the capacitor bank (CB) does not operate; if it is 1, the CB operates.

[0072] S3. In each low-voltage distribution network, with the goal of minimizing network loss, voltage deviation, and operation and maintenance costs, a second control model with a short time scale is constructed based on the multi-agent principle. S3 specifically includes:

[0073] In each low-voltage distribution network, a reward function is constructed based on network loss, voltage deviation, and operation and maintenance costs. Operation and maintenance costs include those of photovoltaic (PV) systems and energy storage systems. Each PV system and energy storage system in the low-voltage distribution network is treated as an agent, and the set of agents is represented as follows: A second regulatory model with a short timescale is obtained by combining the reward function and the agent with the multi-agent principle.

[0074] In the second regulation model, the photovoltaic agent is used to control the photovoltaic output of the node; the energy storage agent is used to select the node charging and discharging strategy according to the grid load demand, battery status and voltage changes.

[0075] In a preferred embodiment of the present invention, in a low-voltage distribution network, photovoltaic and energy storage systems are the main active resources. Their dynamic changes and uncertainties make traditional control methods unable to cope with them. Therefore, a multi-agent deep deterministic policy gradient algorithm is considered for rapid adjustment.

[0076] In a preferred embodiment of the present invention, the reward function defines the agent's behavior and is the reward the agent receives under the current state and action, used to drive the agent to adjust its decisions under the optimization objective. The reward function is expressed as:

[0077] ;

[0078] ;

[0079] ;

[0080] in, Indicates that the intelligent agent is in The reward value at any given moment; It is the weighting coefficient of the network loss item, and the larger the weighting coefficient, the better. The value indicates that network loss is a key area for optimization, and the intelligent system will place greater emphasis on reducing network losses. To cover the operation and maintenance costs of photovoltaic systems, The cost of operation and maintenance for energy storage. It is the weighting coefficient of the voltage deviation term, ensuring that the agent adjusts the voltage; This refers to the power loss of the entire power distribution network; For nodes exist Total active power input at any given time; For nodes exist Total active power output at any given time; Indicates photovoltaic intelligent entity contribution; Represents energy storage intelligent agents of effort.

[0081] S4. Train the second regulation model based on the multi-agent deep deterministic policy gradient algorithm to obtain the third regulation model. S4 specifically includes:

[0082] The state space of the agent is constructed using a second set of parameters, which contains the information that each agent needs to perceive. The second set of parameters includes the node voltage state, the node's active and reactive power, the output power of the photovoltaic unit, and the charging state of the energy storage unit.

[0083] The initial values ​​of each parameter in the second parameter set are collected; these initial values ​​are configured into the second control model, and the agent in the second control model interacts with the environment according to the state space, selecting an action from a preset action space to obtain a reward and the next state; the reward is obtained according to a reward function; the environment includes the current state; specifically, in the state... The agent then selects actions based on the policy. Receive reward and the next state .

[0084] The current state, action, reward, and next state are stored as a set of data in the experience pool. Each agent includes an independent action network and evaluation network, each of which includes a target network and a valuation network. A set of data is sampled from the experience pool to update the parameters of the algorithm network. The agents in the second regulation model interact with the environment repeatedly, and the parameters of the algorithm network are updated cyclically until the preset number of iterations is reached, at which point the loop ends, resulting in the third regulation model.

[0085] In a preferred embodiment of the present invention, selecting an action from a preset action space includes:

[0086] When the intelligent agent is a photovoltaic intelligent agent, the photovoltaic output adjustment target is selected from the preset photovoltaic intelligent agent action space; the photovoltaic intelligent agent action space includes the adjustable range of photovoltaic power generation.

[0087] When the intelligent agent is an energy storage intelligent agent, the energy storage charging or discharging adjustment target is selected from the preset energy storage intelligent agent action space; the energy storage intelligent agent action space includes the adjustable range of the energy storage system's charging and discharging power.

[0088] In a preferred embodiment of the present invention, updating the parameters of the algorithm network includes:

[0089] Evaluation networks are used to assess the quality of actions generated by action networks. This is done by calculating state-action... Values ​​are used to measure the expected reward resulting from a combination of the current state and actions. The value reflects the cumulative reward that an agent can obtain after taking a specific action in a certain state. Its calculation formula is as follows:

[0090] ;

[0091] in, This is the current reward; It is the next local state; It is a discount factor; As a reward for the next moment; The action network provides the parameters for generating the next local state. The following action; Indicate target value.

[0092] The target network is a copy of the action and evaluation networks, and its update rate is relatively slow, typically achieved through soft updates. This mitigates instability during training and ensures... A smoothed estimate of the value. The update expression for the target network is:

[0093] ;

[0094] in, These are the parameters of the current network. These are the parameters of the target network. These are parameters for soft updates; the main role of the target network is in computation. When setting values ​​and update strategies, provide a stable reference to prevent instability caused by excessive updates.

[0095] The role of the valuation network is to provide a value assessment for each state, representing the agent's expected reward in that state. Valuation networks typically compute the value function of a state, helping the agent evaluate the long-term reward of the current state and thus make better decisions. For the valuation network in an action network, the parameters need to be updated via gradient ascent; the parameter update method is as follows:

[0096] ;

[0097] in, The current value of the action to the action Find the gradient; The gradient parameters of the value function are represented. Indicates the state The actions taken will affect the target network parameters under the current strategy. Find the gradient.

[0098] For the valuation network in the evaluation network, minimizing the prediction... Values ​​and Objectives The parameter is updated based on the difference between values. The parameter update method is as follows:

[0099] ;

[0100] in, It is the first The parameters of the network for estimating the actions of an agent; It is the performance objective function; Indicates in Next, take action The output after that; Indicate target value.

[0101] S5. Based on the power of each common node and each third control model, obtain the second control commands for photovoltaic and energy storage in each low-voltage distribution network, implement the second control commands, and complete the control.

[0102] In a preferred embodiment of the present invention, the second control commands for photovoltaic and energy storage in each low-voltage distribution network, obtained based on the power of each common node and each third control model, include:

[0103] In each low-voltage distribution network, with a short time scale as the period (preferably 15 minutes in the preferred embodiment of the present invention), the power of the corresponding common node is configured as the total demand for the operation of the low-voltage distribution network and then incorporated into the corresponding third control model. The third model outputs the second control commands for photovoltaic and energy storage.

[0104] The mechanism modeling and data-driven collaborative multi-level control method for distribution networks of this invention can solve the problems of difficult collaborative modeling of medium and low voltage distribution networks and unknown low-voltage layer architecture. It can achieve collaborative optimization of voltage quality in medium and low voltage distribution networks, reduce system voltage deviation, prevent voltage overruns, improve the absorption capacity of renewable energy, and reduce the total operating cost of the system. It provides a multi-timescale optimized control scheme that balances economy and stability for medium and low voltage distribution networks with high proportion of renewable energy. Based on the method of this invention, it can solve the problem of joint control of medium and low voltage under unknown low-voltage architecture and the problem of slow convergence speed and difficulty in coordinating conflicting objectives of multiple medium and low voltage devices in complex distribution networks using single deep reinforcement learning, thus eliminating node voltage overruns.

[0105] In a preferred embodiment of the present invention, a multi-level control device for distribution networks that combines mechanism modeling and data-driven collaboration is also provided for use with the method of the present invention. The device includes a first module, a second module, a third module, and a fourth module.

[0106] The first module is used to construct a first control model on a long time scale in a medium-voltage distribution network with the goal of minimizing total operating cost and voltage deviation; and to solve the first control model to obtain the first control command for passive equipment in the medium-voltage distribution network.

[0107] The second module is used to collect the power of each common node after the first control command is implemented;

[0108] The third module is used to construct a second control model with a short time scale in each low-voltage distribution network with the goal of minimizing network loss, voltage deviation and operation and maintenance costs based on the multi-agent principle; the second control model is trained based on the multi-agent deep deterministic policy gradient algorithm to obtain the third control model.

[0109] The fourth module is used to obtain the second control commands for photovoltaic and energy storage in each low-voltage distribution network based on the power of each common node and each third control model, implement the second control commands, and complete the control.

[0110] The mechanism modeling and data-driven collaborative multi-level control device for power distribution networks of the present invention, when used in the method of the present invention, has the same beneficial effects as the method of the present invention.

[0111] Verification section:

[0112] In a preferred embodiment of the present invention, see [reference needed]. Figure 3Simulation analysis was conducted using an IEEE 33-node medium-voltage distribution network with a reference voltage of 12.66kV, connected to low-voltage distribution networks with a reference voltage of 0.38kV (15-node and 8-node networks respectively, but with unknown topologies) as the test system. Figure 3 In the diagram, L represents the low-voltage distribution network, C represents the control and compensation capacitor bank, and nodes 1 to 33 are the medium-voltage distribution network nodes. Simulation parameters are shown in Table 1.

[0113] Table 1 Simulation Parameter Table

[0114] ;

[0115] Simulation Result Analysis:

[0116] Figure 4 The training reward curves of the agent are given in the example. It can be seen that as the number of training rounds increases, the agent gradually learns the optimal scheduling strategy, and the sum of the rewards of the photovoltaic energy storage agent begins to stabilize, indicating that a stable optimized system scheduling strategy has been formed.

[0117] To verify the superiority of the method of the present invention, two schemes were set up for comparison, and the schemes were set as follows:

[0118] In this implementation scheme (Scheme 1), the medium-voltage distribution network is optimized using a mixed integer second-order cone programming model, while the low-voltage distribution network is dynamically adjusted using a deep deterministic gradient strategy to optimize the output of photovoltaic and energy storage.

[0119] Comparative implementation method (Scheme 2): The medium-voltage distribution network is optimized using a mixed integer second-order cone programming model, and the photovoltaic and energy storage output of the low-voltage distribution network is instructed by equally distributing the purchased electricity.

[0120] (1) Economic analysis

[0121] The system cost results for the two schemes described above are compared in Table 2.

[0122] Table 2 System Cost Comparison

[0123] ;

[0124] Figure 5 A comparison chart of photovoltaic power output under the two schemes is provided. Figure 6A comparison chart of energy storage output under the two schemes is provided. It can be seen that due to the more precise control strategy in this implementation, photovoltaic and energy storage will have more frequent optimization and adjustment actions. As shown in Table 1, although the more frequent optimization and adjustment actions in this implementation scheme lead to an increase in the operation and maintenance costs of photovoltaic and energy storage, this proactive adjustment also brings significant systemic benefits: First, by dynamically tracking peak and valley electricity prices, the cost of electricity purchase is significantly reduced by 673.6 yuan / day (a decrease of 21.5%); second, the power flow optimization based on multi-agent collaboration reduces the network loss cost by 239.2 yuan / day (a decrease of 38.4%). Therefore, this increased cost is offset by the excess savings in electricity purchase and network losses, confirming the economic viability of the scheme.

[0125] Regarding photovoltaic (PV) grid integration, Scheme 1 achieved a grid integration rate of 97.4%, an improvement of 13.8 percentage points compared to Scheme 2's 83.6%. In high-penetration scenarios, Scheme 1, through real-time dynamic adjustments, achieved a curtailment rate of only 2.6%, while Scheme 2 reached as high as 16.4%. This improvement stems from Scheme 1's two-tiered control architecture: the medium-voltage layer provides guidance through optimization, while the low-voltage layer achieves precise control of PV output through multi-agent reinforcement learning. In contrast, Scheme 2's fixed allocation mode cannot adapt to random fluctuations in PV output. This implementation demonstrates superior overall performance and confirms the feasibility and advantages of the scheme.

[0126] (2) Validity analysis

[0127] The effectiveness of the proposed method is considered from the perspectives of voltage control and photovoltaic power consumption. The effectiveness comparison results are shown in Table 3:

[0128] Table 3. Comparison of Effectiveness

[0129] ;

[0130] Figure 7 The voltage distribution under this embodiment is shown. Figure 8 The voltage distribution under the comparative implementation methods is given, where Indicates the maximum voltage. The value represents the minimum voltage. It can be seen that the voltage compliance rate reaches 100% in this implementation method, while in the comparative implementation method, some node voltages have exceeded the limit. Referring to Table 2, in terms of voltage control, this scheme improves by 12.7 percentage points compared to the comparative scheme. Simultaneously, Scheme 1, through mechanistic modeling and deep reinforcement learning collaborative control, controls the voltage fluctuation standard deviation to 0.009 pu, a 71% reduction compared to Scheme 2's 0.031 pu, significantly improving voltage stability.

[0131] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of multi-level management and control of a power distribution network in which a low-voltage power distribution network architecture is unknown and a medium-voltage power distribution network architecture is known, the method being characterized in that, The method comprises: In the medium-voltage distribution network, a first regulation model of a long time scale is constructed with the minimum total operation cost and voltage deviation as the target; a first regulation instruction of a passive device in the medium-voltage distribution network is obtained by solving the first regulation model; and the first regulation instruction is implemented, and the power of each public node is collected; In each low-voltage distribution network, a second regulation model of a short time scale is constructed with the minimum network loss, voltage deviation and operation and maintenance cost as the target based on the multi-agent principle; the second regulation model is trained based on the multi-agent deep deterministic policy gradient algorithm to obtain a third regulation model; and a second regulation instruction of photovoltaic and energy storage in each low-voltage distribution network is obtained according to the power of each public node and each third regulation model, the second regulation instruction is implemented, and the regulation is completed.

2. The mechanism modeling and data driving collaborative power distribution network multi-level management and control method according to claim 1, characterized in that, The first regulation model of a long time scale is constructed with the minimum total operation cost and voltage deviation as the target, which comprises: In the medium-voltage distribution network, an objective function is constructed with the minimum total operation cost and voltage deviation as the target; the total operation cost comprises a network loss cost and a power purchase cost; the first regulation model of a long time scale is constructed according to the objective function and a preset constraint condition; the preset constraint condition comprises a passive device constraint and an operation constraint; the passive device constraint comprises an on-load voltage regulating transformer constraint and a capacitor bank constraint; and the operation constraint comprises a network operation safety constraint and a network power flow constraint.

3. The mechanism modeling and data driving collaborative power distribution network multi-level management and control method according to claim 2, characterized in that, The first regulation instruction of a passive device in the medium-voltage distribution network is obtained by solving the first regulation model, which comprises: A first parameter set of each node in the medium-voltage distribution network is collected; the first parameter set comprises an active load, a reactive load and impedance information of each branch; the first regulation model is solved according to the first parameter set of each node with a long time scale as a period to obtain the first regulation instruction; the first regulation instruction comprises a gear number change instruction of an on-load voltage regulating transformer and a switching instruction of a capacitor bank.

4. The mechanism modeling and data driving collaborative power distribution network multi-level management and control method according to claim 3, characterized in that, The second regulation model of a short time scale is constructed with the minimum network loss, voltage deviation and operation and maintenance cost as the target based on the multi-agent principle, which comprises: In each low-voltage distribution network, a reward function is constructed with the minimum network loss, voltage deviation and operation and maintenance cost as the target; the operation and maintenance cost comprises an operation and maintenance cost of photovoltaic and energy storage; each photovoltaic system and energy storage system in the low-voltage distribution network is regarded as an agent, and the second regulation model of a short time scale is obtained according to the reward function and the agent based on the multi-agent principle; In the second regulation model, the photovoltaic agent is used to control the photovoltaic output of a node; and the energy storage agent is used to select a node charging and discharging strategy according to the load demand of the power grid, the battery state and the voltage change.

5. The mechanism modeling and data driving collaborative power distribution network multi-level management and control method according to claim 4, characterized in that, The second regulation model is trained based on the multi-agent deep deterministic policy gradient algorithm to obtain a third regulation model, which comprises: A state space of the agent is constructed with a second parameter set; the second parameter set comprises a node voltage state, a node active and reactive power, a photovoltaic unit output power and a charging state of an energy storage unit; Collect initial values of each parameter in the second parameter set; configure the initial values into the second regulation model, and interact the agent in the second regulation model with the environment according to the state space, select an action from a preset action space, obtain a reward and a next state; the reward is obtained according to the reward function; the environment includes the current state; Store the current state, action, reward and next state as a group of data into an experience pool; each agent includes an independent algorithm network, the algorithm network includes a target network and a value network, sample a group of data from the experience pool to update the parameters of the algorithm network; repeat the interaction of the agent in the second regulation model with the environment, and cyclically update the parameters of the algorithm network until a preset iteration number is reached, end the cycle, and obtain the third regulation model.

6. The mechanism modeling and data driving collaborative power distribution network multi-level management and control method according to claim 5, characterized in that, The selecting an action from a preset action space includes: When the agent is a photovoltaic agent, selecting a photovoltaic output adjustment target from a preset photovoltaic agent action space; the photovoltaic agent action space includes an adjustable range of photovoltaic power generation power; When the agent is an energy storage agent, selecting an energy storage charging or discharging adjustment target from a preset energy storage agent action space; the energy storage agent action space includes an adjustable range of energy storage system charging and discharging power.

7. The mechanism modeling and data driving collaborative power distribution network multi-level management and control method according to claim 6, characterized in that, According to the each common node power and each third regulation model, a second regulation instruction of photovoltaic and energy storage in each low-voltage distribution network includes: In each low-voltage distribution network, taking a short time scale as a cycle, the corresponding common node power is configured into the corresponding third regulation model as the total demand of the low-voltage distribution network operation, and the third model outputs the second regulation instruction of the photovoltaic and the energy storage.

8. A mechanism modeling and data driving collaborative power distribution network multi-level management and control device, used for the method of any one of claims 1 to 7, characterized in that, The device includes a first module, a second module, a third module and a fourth module; The first module is used for constructing a long time scale first regulation model in a medium-voltage distribution network, taking the minimum total operation cost and voltage deviation as the target; and solving the first regulation model to obtain a first regulation instruction of passive equipment in the medium-voltage distribution network; The second module is used for collecting each common node power after implementing the first regulation instruction; The third module is used for constructing a short time scale second regulation model in each low-voltage distribution network, taking the minimum network loss, voltage deviation and operation and maintenance cost as the target, based on the multi-agent principle; and training the second regulation model based on the multi-agent deep deterministic policy gradient algorithm to obtain a third regulation model; The fourth module is used for obtaining a second regulation instruction of photovoltaic and energy storage in each low-voltage distribution network according to the each common node power and each third regulation model, implementing the second regulation instruction, and completing regulation.

Images