Voltage control method for ac-dc microgrid group collaborative power supply

Abstract

The application discloses a voltage control method for AC / DC micro-grid group collaborative power supply, comprising the following steps: establishing a voltage stability control optimization model according to the priority of multiple voltage control methods, wherein the control optimization target of the model comprises: keeping the voltage V i Stable at the micro-grid access feeder point, the power loss of ILC power transmission is the lowest, and the voltage stability control cost is the lowest; the voltage stability control optimization model is solved by using a deep reinforcement learning algorithm, an agent capable of autonomously deciding the voltage stability control and optimizing the control strategy is trained, the agent is applied to a local controller and a coordination controller, the monitored grid state quantity is input into the agent, the agent outputs the voltage stability control optimization strategy, and the voltage is controlled according to the voltage stability control optimization strategy. The application realizes the optimization of economy while controlling the voltage stability, and therefore, the stability and economy of the AC / DC micro-grid group interconnection system are improved.

Description

Technical Field

[0001] The present invention belongs to the field of power grid operation technology, and particularly relates to AC / DC microgrid group operation control technology. Background Art

[0002] The proportion of renewable energy in the power grid is gradually increasing, and distributed power generation is also developing rapidly. Today, microgrid systems that integrate distributed power generation, energy storage units, power loads, and advanced power electronics have become the main method for absorbing renewable energy.

[0003] Microgrids can be categorized as AC microgrids powered primarily by wind turbines and DC microgrids powered primarily by photovoltaics. Hybrid AC / DC microgrids, interconnecting these two types of microgrids, are a common scenario. The interconnection of multiple adjacent AC / DC microgrid clusters is now commonplace, necessitating the research on the structural design and control methods for secure and stable control systems for these clusters.

[0004] The widespread use of advanced power electronics in smart grids has led to the increasing application of data-driven artificial intelligence algorithms. Deep reinforcement learning algorithms, a combination of deep learning and reinforcement learning, empower intelligent agents with both enhanced environmental perception and powerful intelligent decision-making capabilities, showing broad application prospects in the field of smart grid control. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a voltage control method for the coordinated mutual supply of AC / DC microgrid groups, which effectively maintains the voltage stability of the AC / DC microgrid group and realizes the coordinated mutual supply of power between the AC / DC microgrids.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A voltage control method for collaboratively supplying power to an AC / DC microgrid group includes the following steps:

[0008] A voltage stability control optimization model is established based on the priority of multiple voltage control methods. The control optimization objectives of this model include: the voltage V at the microgrid access feeder point i Stable, ILC power transmission has the lowest power loss and the lowest voltage stability control cost;

[0009] Use deep reinforcement learning algorithms to solve the voltage stability control optimization model and train an intelligent agent that can make autonomous decisions to control voltage stability and optimize control strategies;

[0010] The intelligent agent is applied to the local controller and the coordinated controller, the monitored grid state quantity is input to the intelligent agent, the intelligent agent outputs the voltage stability control optimization strategy, and controls the voltage according to the voltage stability control optimization strategy.

[0011] Preferably, the priorities of the multiple voltage control methods are as follows:

[0012] (1) Prioritize the control of the reactive output voltage regulation of distributed power sources: adjust the microgrid photovoltaic reactive power Q pv and wind turbine reactive Q wt ;

[0013] (2) Then control the voltage regulation of the energy storage unit: adjust the active power P of the energy storage ESS and reactive Q ESS ;

[0014] (3) Then control the active output voltage regulation of the distributed power supply: adjust the active output of the microgrid photovoltaic power P pv and wind turbine active output P wt ;

[0015] (4) Finally, control the AC and DC load reduction and voltage regulation: Partial load p load Load shedding to cope with large, difficult-to-regulate voltage fluctuations.

[0016] Preferably, the objective function of the voltage stability control optimization model is:

[0017]

[0018] Where, Indicates the total deviation between the voltage of each microgrid connection point and the feeder reference voltage. The smaller it is, the more stable the voltage is at the reference value V rv Department;

[0019] It represents the power loss of the converter ILC. The conversion power loss calculation formula is:

[0020]

[0021] Where η is the conversion efficiency, P ref is the transformed power value;

[0022] The voltage stability control cost function is:

[0023] α3(β1C Qpv +β2C Qwt +β3C PESS +β4C QESS +β5C Pwt +β6C Ppv +β7C load ) (3)

[0024] Among them, C Qpv is the photovoltaic reactive power regulation cost, C Qwtis the reactive power regulation cost of the wind turbine, C PESS is the active regulation cost of energy storage, C QESS is the reactive regulation cost of energy storage, C Ppv is the photovoltaic active power regulation cost, C Pwt is the active power regulation cost of the fan, C load is the load regulation cost;

[0025] α1, α2, and α3 represent the weight coefficients of the three optimization objectives of grid connection point voltage stability, ILC power loss, and voltage stability cost, respectively, α1>α2>α3;

[0026] β1, β2, β3, β4, β5, β6, and β7 are the weight coefficients of the cost function of PV reactive power, wind turbine reactive power, energy storage active power, energy storage reactive power, wind turbine active power, PV active power, and load participating in voltage regulation, respectively. The weight coefficients satisfy β1 = β2 < β3 = β4 < β5 = β6 < β7;

[0027] The reactive power regulation cost model of photovoltaics is:

[0028] C Qpv =γ pv ΔQ pv (4)

[0029] where Q pv is the photovoltaic reactive output, γ pv is the photovoltaic management and operation cost coefficient;

[0030] The active power regulation cost model of photovoltaics is:

[0031] C Ppv =γ pv ΔP pv +λ pv ΔP pv (5)

[0032] Where P pv is the photovoltaic active output, γ pv is the photovoltaic management operation cost coefficient, λ pv is the photovoltaic curtailment penalty factor;

[0033] The reactive power regulation cost model of the wind turbine is:

[0034] C Qwt =γ wt ΔQ wt (6)

[0035] where Q wt is the wind turbine reactive output, γ wt is the wind turbine management and operation cost coefficient;

[0036] The active power regulation cost model of the wind turbine is:

[0037] C Pwt =γ wt ΔP wt +λ wt ΔP wt (7)

[0038] Among them, P wt is the fan power output, γ wt is the fan management operation cost coefficient, λ wt is the wind turbine’s wind curtailment penalty factor; the cost model for active power regulation of the energy storage unit is:

[0039] C PESS =(γ ES.om +γ z )ΔP ESS (8)

[0040] Where P ESS is the active power change of the energy storage unit, γ ES.om is the management and maintenance cost coefficient, γ z is the depreciation cost coefficient; the cost model of reactive power regulation of energy storage unit is:

[0041] C QESS =(γ ES.om +γ z )ΔQ ESS (9)

[0042] Where Q ESS is the active power change of the energy storage unit, γ ES.om is the management and maintenance cost coefficient, γ z Depreciation cost coefficient; The constraints of the above voltage stability control optimization model are:

[0043]

[0044] In the above constraints:

[0045] (10) is the range constraint of the wind turbine active power (reactive power) output;

[0046] (11) is the range constraint of PV active power (reactive power) output;

[0047] (12) is the charge and discharge power range constraint of the energy storage system;

[0048] (13) is the reactive power regulation capability constraint of the energy storage system;

[0049] (14) and (15) are the state of charge constraints of the energy storage system, where δ is the energy storage conversion efficiency, R ES is the total capacity of the energy storage unit;

[0050] (16) is the load reduction constraint, P load.s.max is the maximum allowable load reduction;

[0051] (17) is the power range constraint of the ILC converter.

[0052] Preferably, a multi-agent deep deterministic policy gradient algorithm is used to solve the voltage stability control optimization model, and the algorithm model includes: multiple agent action networks Actors, and multiple evaluation networks Critic of Actors;

[0053] Among them, the input of the Actor network is the state S of the environment, and the output is the action a of the agent; the input of the Critic network includes the agent's state before action S, the state after action S', the action set a of all agents, and the reward R, and the output is the Q value of the agent.

[0054] Preferably, θ=[θ1,…,θ n ] represents the strategy parameters of n agents, π=[π1,…,π n ] represents the strategy of n agents;

[0055] The cumulative expected reward of the i-th agent is:

[0056]

[0057] The deterministic policy gradient can be calculated from the cumulative expected reward:

[0058]

[0059] In the formula, o i represents the observation of the i-th agent, represents the centralized state-action function of the i-th agent;

[0060] The update method of the centralized critic is:

[0061]

[0062] Where, Indicates the target network.

[0063] Preferably, define the state space S of the Actor:

[0064]

[0065] Where i is the number of AC microgrids, j is the number of DC microgrids, and the state quantities are: microgrid grid connection point voltage at time t, AC / DC bus voltage, ILC converter power, wind turbine reactive power, wind turbine active power, PV reactive power, PV active power, energy storage unit SOC, and load state.

[0066] Define the Actor's action space a:

[0067]

[0068] Where i is the number of AC microgrids, j is the number of DC microgrids, and the actions are as follows at time t: AC microgrid wind turbine reactive power, DC microgrid photovoltaic reactive power, energy storage unit active power, energy storage unit reactive power, wind turbine active power, photovoltaic active power, and load shedding;

[0069] Define the action reward R:

[0070]

[0071] Where r1 is the voltage instability penalty, and the voltage stability range is 0.95 to 1.05 times the standard value; r2 is the converter loss penalty; and r3 is the voltage stability control cost. A negative reward indicates a higher control cost and a smaller reward. Considering the importance of voltage instability penalty, converter loss penalty, and control cost optimization, α1>>α2>α3.

[0072] To maximize the absorption of renewable energy generation and ensure the stable operation of AC / DC multi-microgrid systems, this paper prioritizes various voltage control methods and constructs an optimization model for voltage stability control in AC / DC microgrid clusters. This approach, using the MADDPG method for solving this optimization model, effectively addresses the issues of power supply and voltage fluctuation stability control in interconnected AC / DC microgrid clusters, particularly when multiple microgrid currents interact. While maintaining voltage stability, this paper analyzes the costs of various voltage control methods, establishes an economic optimization model, and fully utilizes these methods in control decisions, achieving optimal economic performance while maintaining voltage stability. This approach improves the stability and economic efficiency of interconnected AC / DC microgrid clusters.

[0073] The specific technical solutions and beneficial effects of the present invention will be described in detail in the following specific embodiments with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0075] Figure 1 This is a flow chart of the voltage control method for the AC / DC microgrid group proposed in the present invention.

[0076] Figure 2 This is the topological structure diagram of the AC / DC microgrid group's coordinated mutual supply and voltage stability control system.

[0077] Figure 3 This is the voltage control stabilization priority proposed by the present invention. DETAILED DESCRIPTION

[0078] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention, its application, or use. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0079] The present invention proposes a voltage control method for AC / DC microgrid group cooperative mutual supply based on MADDPG. The topology of the voltage stability control system for AC / DC microgrid group cooperative mutual supply is as follows: Figure 2 ,The system is divided into two parts: power lines and communication lines.

[0080] In the power transmission line, the buses of multiple AC microgrids are connected to the feeder via transformers and grid-connected switches (PCCs). Multiple DC microgrids are connected to the feeder via bidirectional power converters (ILCs) and grid-connected switches (PCCs). In the AC microgrid, distributed power sources (wind turbines) and energy storage systems (ESSs) are connected to the AC bus via transformers, and AC loads are directly connected to the AC bus. In the DC microgrid, distributed power sources (photovoltaic power generation) and energy storage systems (ESSs) are connected to the DC bus via bidirectional isolation DC transformers, and DC loads are directly connected to the DC bus.

[0081] Each microgrid has a local controller within the communication line, which monitors the microgrid's operating status and controls power electronics. All microgrid local controllers are connected to a microgrid coordination controller, which coordinates the microgrids to achieve control objectives such as inter-microgrid power supply and voltage stability.

[0082] In order to maximize the absorption of renewable energy power generation in AC / DC microgrids and ensure system voltage stability, the present invention proposes the following technical solutions:

[0083] A voltage control method for coordinated mutual supply of AC / DC microgrid groups based on MADDPG is used to solve complex optimization problems in multi-microgrid coordination scenarios, such as Figure 1 As shown, the following steps are included:

[0084] Step 1: Obtain the AC / DC microgrid operation data and equipment information through the local controller, including: the voltage V of each microgrid feeder node i , feeder rated voltage V rv , microgrid bus voltage U dc , energy storage system information, ILC inter-network transmission power P ref , the reactive power Q of each microgrid distributed power source PV and Q WT , Active power P of each microgrid distributed power source PV and P WT , load size P load , the conversion efficiency η of the converter ILC, etc.

[0085] Step 2: Aiming at the voltage stability control problem of multiple microgrids, the economic feasibility of various voltage control methods of microgrids is analyzed to determine the priority of various voltage control methods to guide the design of voltage stability optimization strategy.

[0086] Step 3: Based on the voltage control method priority described in step 2, a voltage stability control optimization model is established. The control optimization objectives of this model include: the voltage V at the microgrid access feeder point i Stable, ILC power transmission has the lowest power loss and the lowest voltage stability control cost;

[0087] Step 4: Use the deep reinforcement learning algorithm to solve the model established in step 3, and train an intelligent agent that can make autonomous decisions to control voltage stability and optimize control strategies.

[0088] Step 5: Apply the intelligent agent to the local controller and coordination controller of the AC / DC microgrid group collaborative control system, input the monitored grid state quantity, and obtain the voltage stability control optimization strategy.

[0089] Furthermore, in step 1, in order to ensure the maximum absorption of new energy, P PV and P WT It is usually set to the maximum output power P of photovoltaic and wind turbines. PVmax and P WTmax .

[0090] In step 2, through the economic and safety analysis of various voltage control methods, we can get Figure 3 The voltage stability control priority shown in Figure 2 can be used to guide the design of voltage stability optimization strategies. The priorities of multiple voltage control methods are as follows:

[0091] (1) Prioritize the control of the reactive output voltage regulation of distributed power sources: adjust the microgrid photovoltaic reactive power Q pvand wind turbine reactive Q wt .

[0092] (2) Then control the voltage regulation of the energy storage unit: adjust the active power P of the energy storage ESS and reactive Q ESS .

[0093] (3) Then control the active output voltage regulation of the distributed power supply: adjust the active output of the microgrid photovoltaic power P pv and wind turbine active output P wt .

[0094] (4) Finally, control the AC and DC load reduction and voltage regulation: Partial load p load Load shedding to cope with large, difficult-to-regulate voltage fluctuations.

[0095] In step 3, the voltage stability control optimization model objective function is established as:

[0096]

[0097] Where, Indicates the total deviation between the voltage of each microgrid connection point and the feeder reference voltage. The smaller it is, the more stable the voltage is at the reference value V rv Place.

[0098] In the above voltage stability control optimization model, Represents the power loss of the converter ILC. The conversion power loss calculation formula is:

[0099]

[0100] Where η is the conversion efficiency, P ref is the conversion power value.

[0101] In the above voltage stability control optimization model, the voltage stability control cost function is:

[0102] α3(β1C Qpv +β2C Qwt +β3C PESS +β4C QESS +β5C Pwt +β6C Ppv +β7C load ) (3)

[0103] Among them, C Qpv is the photovoltaic reactive power regulation cost, C Qwt is the reactive power regulation cost of the wind turbine, C PESS is the active regulation cost of energy storage, C QESS is the reactive regulation cost of energy storage, C Ppvis the photovoltaic active power regulation cost, C Pwt is the active power regulation cost of the fan, C load is the load regulation cost.

[0104] In the voltage stability control optimization model described above, α1, α2, and α3 represent the weighting coefficients for the three optimization objectives: grid connection point voltage stability, ILC power loss, and voltage stability cost. This is because voltage stability outweighs power loss reduction, which outweighs voltage stability cost optimization. Therefore, α1 > α2 > α3.

[0105] In the voltage stability control optimization model, β1, β2, β3, β4, β5, β6, and β7 are the weight coefficients of the cost function for PV reactive power, wind turbine reactive power, energy storage active power, energy storage reactive power, wind turbine active power, PV active power, and load, respectively, in voltage regulation. Based on the voltage stability control priority in step 2, these weight coefficients satisfy β1 = β2 < β3 = β4 < β5 = β6 < β7.

[0106] In the above voltage stability control optimization model, the photovoltaic reactive power regulation cost model is:

[0107] C Qpv =γ pv ΔQ pv (4)

[0108] where Q pv is the photovoltaic reactive output, γ pv is the photovoltaic management operation cost coefficient.

[0109] The active power regulation cost model of photovoltaics is:

[0110] C Ppv =γ pv ΔP pv +λ pv ΔP pv (5)

[0111] Where P pv is the photovoltaic active output, γ pv is the photovoltaic management operation cost coefficient, λ pv It is the penalty factor for photovoltaic curtailment.

[0112] The reactive power regulation cost model of the wind turbine is:

[0113] C Qwt =γ wt ΔQ wt (6)

[0114] where Q wt is the wind turbine reactive output, γ wt is the fan management operation cost coefficient.

[0115] The active power regulation cost model of the wind turbine is:

[0116] C Pwt =γ wt ΔP wt +λ wt ΔP wt (7)

[0117] Among them, P wt is the fan power output, γ wt is the fan management operation cost coefficient, λ wt is the wind turbine curtailment penalty factor.

[0118] The cost model for active power regulation of energy storage units is:

[0119] C PESS =(γ ES.om +γ z )ΔP ESS (8)

[0120] Where P ESS is the active power change of the energy storage unit, γ ES.om is the management and maintenance cost coefficient, γ z is the depreciation cost factor.

[0121] The cost model of reactive power regulation of energy storage units is:

[0122] C QESS =(γ ES.om +γ z )ΔQ ESS (9)

[0123] Where Q ESS is the active power change of the energy storage unit, γ ES.om is the management and maintenance cost coefficient, γ z Depreciation cost factor.

[0124] The constraints of the above voltage stability control model are:

[0125]

[0126] In the above constraints:

[0127] (10) is the range constraint of the wind turbine active power (reactive power) output;

[0128] (11) is the range constraint of PV active power (reactive power) output;

[0129] (12) is the charge and discharge power range constraint of the energy storage system;

[0130] (13) is the reactive power regulation capability constraint of the energy storage system;

[0131] (14) and (15) are the state of charge constraints of the energy storage system, where δ is the energy storage conversion efficiency, R ES is the total capacity of the energy storage unit;

[0132] (16) is the load reduction constraint, P load.s.max is the maximum allowable load reduction.

[0133] (17) is the power range constraint of the ILC converter.

[0134] In step 4, to address the coordinated control of multiple controllers in multiple microgrids, the present invention uses a multi-agent deep deterministic policy gradient algorithm (MADDPG) to solve the voltage stability control optimization model. This algorithm model includes: multiple agent action networks (Actors) and multiple evaluation networks (Critics).

[0135] The input of the Actor network is the state S of the environment, and the output is the action a of the agent. The input of the Critic network includes the agent's state before the action S, the state after the action S', the action set a of all agents, and the reward R, and the output is the Q value of the agent.

[0136] The basic training process of the MADDPG algorithm is:

[0137] 1. Initialize the neural network parameters, state value S, and training parameters.

[0138] 2. The Actor randomly selects an action, obtains the current state value S, reward R, the set of all agent actions a, and the next state value S', and stores them in the experience pool D. The agent round is also updated.

[0139] 3. The Critic randomly selects a batch of data from the experience pool, calculates the loss function, and updates the policy network according to the gradient update strategy. Centralized training establishes an action-value function for each agent.

[0140] 4. Critic outputs the Q value of each agent based on the training information of all agents in the experience pool.

[0141] 5. The Actor then updates the neural network parameters based on the Q value returned by the Critic, and updates to the target network parameters after a certain number of rounds.

[0142] 6. During the training process, the target network parameters are continuously optimized.

[0143] 7. Training is completed and the target network is obtained

[0144] In process 3, use θ=[θ1,…,θ n] represents the strategy parameters of n agents, π=[π1,…,π n ] represents the strategy of n agents.

[0145] In process 3, the cumulative expected reward of the i-th agent is:

[0146]

[0147] The deterministic policy gradient can be calculated from the cumulative expected reward:

[0148]

[0149] In the formula, o i represents the observation of the i-th agent, represents the centralized state-action function of the i-th agent.

[0150] In process 4, the update method of the centralized critic is:

[0151]

[0152] Where, Indicates the target network.

[0153] The key advantages of the MADDPG algorithm for voltage stability control in AC / DC microgrids lie in its centralized training and distributed execution. Specifically, the Critic uses centralized training, allowing each agent to understand the strategies of other agents. This means that each microgrid's control strategy considers the influence of other microgrids, improving the stability and robustness of the control strategy.

[0154] Combined with the voltage stability optimization model, the state space S of the Actor is defined as follows:

[0155]

[0156] Where i is the number of AC microgrids, j is the number of DC microgrids, and the state quantities are: microgrid grid connection point voltage at time t, AC / DC bus voltage, ILC converter power, wind turbine reactive power, wind turbine active power, PV reactive power, PV active power, energy storage unit SOC, and load status.

[0157] Combined with the voltage stability optimization model, the action space a of the Actor is defined as follows:

[0158]

[0159] Where i is the number of AC microgrids, j is the number of DC microgrids, and the actions at time t are: AC microgrid wind turbine reactive power, DC microgrid photovoltaic reactive power, energy storage unit active power, energy storage unit reactive power, wind turbine active power, photovoltaic active power, and load shedding.

[0160] Combined with the voltage stability optimization model, the action reward R is defined as:

[0161]

[0162] Where r1 is the voltage instability penalty, and the voltage stability range is 0.95 to 1.05 times the standard value; r2 is the converter loss penalty; and r3 is the voltage stability control cost. A negative reward indicates that the higher the control cost, the smaller the reward.

[0163] When setting the above rewards, the importance of voltage instability penalty, converter loss penalty, and control cost optimization are taken into consideration, that is: α1>>α2>α3.

[0164] The intelligent agent trained by the MADDPG algorithm is used in the local controller of the microgrid. The local controller obtains the grid status value through real-time monitoring of the grid, and obtains the control strategy of other intelligent agents through the coordination controller to achieve the training requirements of centralized criticism and distributed execution of actors.

[0165] Compared to traditional microgrid control, this design effectively addresses the complex issue of voltage stability control in interconnected AC / DC microgrid clusters, where multiple microgrid flows interact. While maintaining voltage stability, this approach analyzes the costs of various voltage control methods, establishes an economic optimization model, and leverages these models in control decisions. This approach achieves optimal economic performance while maintaining voltage stability.

[0166] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art will understand that the present invention includes, but is not limited to, the contents described in the above specific embodiment. Any modifications that do not deviate from the functional and structural principles of the present invention are intended to be included within the scope of the claims.

Claims

1. A voltage control method for coordinated mutual supply of AC / DC microgrids, characterized in that: The following steps are included: establishing a voltage stability control optimization model according to the priority of multiple voltage control methods, and the control optimization objectives of the model include: the voltage V at the microgrid access feeder point i Stable, ILC power transmission has the lowest power loss and the lowest voltage stability control cost; Use deep reinforcement learning algorithms to solve the voltage stability control optimization model and train an intelligent agent that can make autonomous decisions to control voltage stability and optimize control strategies; Apply the intelligent agent to the local controller and the coordinated controller, input the monitored grid state quantity to the intelligent agent, and the intelligent agent outputs the voltage stability control optimization strategy and controls the voltage according to the voltage stability control optimization strategy; The priorities of multiple voltage control methods are as follows: (1) Prioritize the control of the reactive output voltage regulation of distributed power sources: adjust the microgrid photovoltaic reactive power Q pv and wind turbine reactive Q wt ; (2) Then control the voltage regulation of the energy storage unit: adjust the active power P of the energy storage ESS and reactive Q ESS ; (3) Then control the active output voltage regulation of the distributed power supply: adjust the active output of the microgrid photovoltaic power P pv and wind turbine active output P wt ; (4) Finally, control the AC and DC load reduction and voltage regulation: Partial load p load Load shedding to cope with large, difficult-to-regulate voltage fluctuations; The objective function of the voltage stability control optimization model is: Where, Indicates the total deviation between the voltage of each microgrid connection point and the feeder reference voltage. The smaller it is, the more stable the voltage is at the reference value V rv Department; It represents the power loss of the converter ILC. The conversion power loss calculation formula is: Where η is the conversion efficiency, P ref is the transformed power value; The voltage stability control cost function is: α3(β1C Qpv +β2C Qwt +β3C PESS +β4C QESS +β5C Pwt +β6C Ppv +β7C load ) (3) Among them, C Qpv is the photovoltaic reactive power regulation cost, C Qwt is the reactive power regulation cost of the wind turbine, C PESS is the active regulation cost of energy storage, C QESS is the reactive regulation cost of energy storage, C Ppv is the photovoltaic active power regulation cost, C Pwt is the active power regulation cost of the fan, C load is the load regulation cost; α1, α2, and α3 represent the weight coefficients of the three optimization objectives of grid connection point voltage stability, ILC power loss, and voltage stability cost, respectively, α1>α2>α3; β1, β2, β3, β4, β5, β6, and β7 are the weight coefficients of the cost function of PV reactive power, wind turbine reactive power, energy storage active power, energy storage reactive power, wind turbine active power, PV active power, and load participating in voltage regulation, respectively. The weight coefficients satisfy β1 = β2 < β3 = β4 < β5 = β6 < β7; The reactive power regulation cost model of photovoltaic is: C Qpv =γ pv ΔQ pv (4) where Q pv is the photovoltaic reactive output, ΔQ pv is the change in photovoltaic reactive output, γ pv is the photovoltaic management and operation cost coefficient; The active power regulation cost model of photovoltaics is: C Ppv =c pv ΔP pv +λ pv ΔP pv (5) Where P pv is the photovoltaic active output, ΔP pv is the change in photovoltaic active output, γ pv is the photovoltaic management operation cost coefficient, λ pv is the photovoltaic curtailment penalty factor; The reactive power regulation cost model of the wind turbine is: C Qwt =γ wt ΔQ wt (6) where Q wt is the wind turbine reactive output, ΔQ wt is the change in wind turbine reactive output, γ wt is the wind turbine management and operation cost coefficient; The active power regulation cost model of the wind turbine is: C Pwt =c wt ΔP wt +λ wt ΔP wt (7) Among them, P wt is the active power output of the fan, ΔP wt is the change in the active power output of the wind turbine, γ wt is the fan management operation cost coefficient, λ wt is the wind turbine’s wind curtailment penalty factor; The cost model for active power regulation of energy storage units is: C PESS =(γ ES.om +g z )ΔP ESS (8) Where P ESS is the active output of the energy storage unit, ΔP ESS is the change in active output of the energy storage unit, γ ES.om is the management and maintenance cost coefficient, γ z is the depreciation cost factor; The cost model of reactive power regulation of energy storage units is: C QESS =(γ ES.om +g z )ΔQ ESS (9) Where Q ESS is the reactive output of the energy storage unit, ΔQ ESS is the change in reactive output of the energy storage unit, γ ES.om is the management and maintenance cost coefficient, γ z Depreciation cost factor; The constraints of the above voltage stability control optimization model are: In the above constraints: (10) is the range constraint of wind turbine active power / reactive power output; (11) is the range constraint of PV active power / reactive power output; (12) is the charge and discharge power range constraint of the energy storage system; (13) is the reactive power regulation capability constraint of the energy storage system; (14) and (15) are the state of charge constraints of the energy storage system, where δ is the energy storage conversion efficiency, R ES is the total capacity of the energy storage unit; (16) is the load reduction constraint, P load.s.max is the maximum allowable load reduction; (17) is the ILC converter power range constraint; A multi-agent deep deterministic policy gradient algorithm is used to solve the voltage stability control optimization model, which includes: multiple agent action networks (Actors) and multiple evaluation networks (Critics). The input of the Actor network is the state S of the environment, and the output is the action a of the agent. The input of the Critic network includes the agent's state before the action S, the state after the action S', the action set a of all agents, and the reward R, and the output is the Q value of the agent. Use θ=[θ1,…,θ n ] represents the strategy parameters of n agents, π=[π1,…,π n ] represents the strategy of n agents; the cumulative expected reward of the i-th agent is: The deterministic policy gradient can be calculated from the cumulative expected reward: In the formula, o i represents the observation of the i-th agent, represents the centralized state-action function of the i-th agent; The update method of the centralized critic is: Where, represents the target network; Define the state space S of the Actor: Where i is the number of AC microgrids, j is the number of DC microgrids, and the state quantities are: microgrid grid connection point voltage at time t, AC / DC bus voltage, ILC converter power, wind turbine reactive power, wind turbine active power, PV reactive power, PV active power, energy storage unit SOC, and load state. Define the Actor's action space a: Where i is the number of AC microgrids, j is the number of DC microgrids, and the actions are as follows at time t: AC microgrid wind turbine reactive power, DC microgrid photovoltaic reactive power, energy storage unit active power, energy storage unit reactive power, wind turbine active power, photovoltaic active power, and load shedding; Define the action reward R: Where r1 is the voltage instability penalty, and the voltage stability range is 0.95 to 1.05 times the standard value; r2 is the converter loss penalty; and r3 is the voltage stability control cost. A negative reward indicates that the higher the control cost, the smaller the reward. Considering the importance of voltage instability penalty, converter loss penalty, and control cost optimization, α1>>α2>α3.

Images