Distributed multi-agent adaptive optimization incremental power distribution network power voltage balance method

Abstract

The invention discloses a distributed multi-agent self-adaptive optimization incremental power distribution network power voltage balancing method, which comprises the following specific steps: carrying out data monitoring by a voltage and current sensor, acquiring real-time voltage and current data of each feeder line of a system, calculating real-time power on each sub-network node by agents through the data, and calculating the real-time power of each sub-network node according to the real-time power of each sub-network node; calculating reference power by using real-time power data of the current converter and an adjacent agent, and realizing decoupling control of voltage by the current converter through the data; a radiation type system is considered, a virtual voltage source-sub-network decomposition model is constructed, the system is decomposed into a plurality of sub-networks, a network construction type power supply is decomposed into two virtual voltage sources, and each sub-network comprises the virtual voltage sources on the two sides, an internal load and a network following type power supply; and constructing a cost optimization function of each sub-network, carrying out iteration based on a spherical search analysis method to continuously obtain optimal voltage and power points, and carrying out multiple iterations to realize the optimal solution of the voltage and power of the system.

Description

Technical Field

[0001] This invention relates to the field of power system control technology, and specifically to a distributed multi-agent adaptive optimization method for incremental power and voltage balance in distribution networks. Background Technology

[0002] The high proportion of distributed renewable energy integrated into the distribution system has given rise to a new business model: incremental distribution networks. As a distribution network system connected to the public grid with a defined power supply range, it typically contains diverse distributed resources: it can include grid-based power sources, such as battery energy storage systems, which possess autonomous networking and flexible control capabilities, as well as grid-connected power sources, such as photovoltaic systems, whose output is intermittent and random. While this structure improves energy efficiency and power supply reliability, it also brings unprecedented challenges to system operation and control. On the one hand, the distribution network is transforming from a traditional "passive" radial network to an "active" interconnected network, resulting in increasingly complex power flow distribution; on the other hand, the system's operational objectives have shifted from a single voltage qualification rate control to simultaneously considering multiple objectives, including power quality within the grid, economic allocation of reactive resources, and protocol constraints with the main grid (such as prohibiting reverse flow and limiting maximum demand).

[0003] Existing distributed optimization algorithms for microgrids offer valuable insights into solving internal resource coordination problems. Distributed schemes, such as those based on the alternating direction multiplier method or the augmented Lagrange alternating direction incomplete Newton method, have been proposed to achieve accurate reactive power sharing and average voltage recovery. However, these methods have significant limitations in their applicability when applied to incremental distribution networks tightly coupled with the main grid. First, most of these methods are designed for islanded microgrids, focusing on internal frequency stability and voltage regulation, lacking an interface model for interaction with the upper-level grid, and unable to inherently handle crucial boundary conditions such as power constraints at the point of common coupling. Simply applying these methods can easily lead to optimization results that violate grid connection protocols, causing safety incidents or economic penalties.

[0004] Secondly, at the control architecture level, existing distributed algorithms typically divide the network into complex subnets containing multiple network-type units, resulting in a still large sub-optimization problem scale. Solving these problems still relies on high-performance computing platforms, making it difficult to meet the requirements of incremental distribution networks for low-cost and high-real-time deployment of control strategies. Furthermore, some distributed control schemes based on consensus theory, while having lower communication requirements, often exhibit poor dynamic response performance, performing poorly when facing nonlinear loads or severe power fluctuations. They also typically rely on droop control as the underlying support, making it difficult to fundamentally solve the reactive power distribution deviation problem caused by line impedance.

[0005] Furthermore, from the perspective of optimization models, most studies treat active power as an optimization variable or equality constraint, which is reasonable in islanded systems. However, in grid-connected modes, the system frequency is dominated by the main grid, and the active power balance within the incremental distribution network is more affected by the power exchange constraints at the point of common coupling and the underlying primary frequency regulation characteristics. How to redefine the active-reactive coupling relationship in the optimization model, and how to efficiently solve the reactive voltage optimization problem while ensuring that active power exchange does not exceed the limits, has become a blind spot that has not been fully addressed by existing technologies.

[0006] Finally, regarding communication reliability, incremental distribution networks have much higher requirements for communication systems than islanded microgrids. In islanded mode, communication interruptions may lead to a decrease in control performance, but the system can still maintain operation; however, in grid-connected mode, communication failures may result in the inability to perceive the status of the point of common coupling in real time, leading to serious violations such as power backfeeding, and lacking fault-tolerant control mechanisms in the event of communication limitations or interruptions.

[0007] Therefore, a novel distributed optimization control method is urgently needed. This method must: first, deeply embed grid connection protocol constraints to ensure absolute safety and compliance of all optimization results; second, employ a highly decoupled distributed architecture to significantly reduce computational and communication complexity and adapt to low-cost hardware deployment; third, possess excellent real-time performance and robustness, enabling rapid tracking of random fluctuations in internal photovoltaic output and load; and fourth, clearly distinguish between control strategies for frequency / active power and voltage / reactive power, focusing on solving the core problems of incremental distribution networks—voltage management and reactive power optimization. To address these issues, we propose a distributed multi-agent adaptive optimization method for incremental distribution network power and voltage balance. Summary of the Invention

[0008] To address this issue, the present invention provides a distributed multi-agent adaptive optimization method for incremental power and voltage balance in distribution networks. This method solves the problem that existing distributed optimization algorithms for microgrids lack an interface model for interaction with the upper-level power grid, which can easily lead to optimization results violating grid connection protocols and causing safety accidents or economic penalties.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A distributed multi-agent adaptive optimization method for incremental power and voltage balance in distribution networks, comprising the following steps:

[0011] Step 1: Data monitoring is performed by voltage and current sensors to collect real-time voltage and current data of each feeder in the system. The intelligent agent calculates the real-time power on each sub-network node using this data, and calculates the reference power using its own and neighboring intelligent agents' real-time power data. The converter achieves decoupled voltage control using this data.

[0012] Step 2: Considering the radial system, construct a "virtual voltage source-sub-network" decomposition model, decompose the system into multiple sub-networks, and decompose the grid-type power supply into two virtual voltage sources. Each sub-network includes virtual voltage sources on both sides, as well as internal loads and grid-type power supplies. Each sub-network acts as an intelligent agent and participates in the operation of the distributed system.

[0013] Step 3: Construct the cost optimization function for each sub-network. Considering the differences in the common access points, the cost optimization function will be different. Based on the spherical search analysis method, the optimal voltage and power points are obtained iteratively, and the reference voltage value that meets the judgment threshold is output to each actual network source. After multiple iterations, the optimal solution of system voltage and power is achieved.

[0014] Preferably, in step one, the real-time voltage and current of the grid-type source and load feeder are collected, the agent calculates the real-time power, and collects data from itself and neighboring agents. Through the difference equation, the global power is obtained in iterations, thereby making the reference power more and more accurate, and finally realizing the reasonable allocation of the system's global power.

[0015] Preferably, the difference equation is:

[0016]

[0017] in Represents a node All active power operations and, , , The estimated power operation is derived from adjacent nodes, and the calculation method for reactive power is similar.

[0018] Preferably, in step two, considering the radial system, a "virtual voltage source-sub-network" decomposition model is constructed. Each sub-network contains one branch of the virtual voltage source in the decomposed network source. Each sub-network includes the load between virtual voltage sources and the network source, and participates in the system voltage power balance calculation as a sub-network.

[0019] Preferably, in the "virtual voltage source-sub-network" decomposition model, the grid-type source mainly supplements the active power deficit in the system and does not provide reactive power support. In addition to the regular work of the energy storage power station, the grid-type source also needs to perform voltage balancing and reactive power output.

[0020] The general cost optimization function of the system is:

[0021]

[0022] in It is the first Individual networks, For the first The number of load feeders in each subnetwork For the first The weight of each load feeder For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources It is the first A network-type source has power-constrained Lagrange multipliers. The reference reactive power for grid-connected power sources. For the first The actual reactive power of a grid-type power supply. The reference active power for grid-connected power sources. For the first The actual active power of a grid-type power supply.

[0023] The present invention has the following advantages:

[0024] 1) This method decomposes a complex large system into simple sub-problems through a virtual decoupling mechanism, and each intelligent agent solves the problem in parallel, resulting in extremely high computational efficiency. Communication only requires neighbor information, making it very suitable for implementation with low-cost hardware (such as digital signal processors), thus meeting the stringent requirements of incremental distribution networks for economy and real-time performance.

[0025] 2) This method innovatively integrates global hard constraints of common connection points into the control system through economic penalties and upper-level supervision, fundamentally preventing illegal operations such as power backfeeding and ensuring the safety and compliance of system operation.

[0026] 3) The spherical decoupling search algorithm of this method has good convergence characteristics and low sensitivity to parameter changes. It can quickly respond to violent disturbances such as sudden changes in photovoltaic output and load switching in the grid, and maintain voltage stability and reactive power distribution balance.

[0027] 4) Under the premise of satisfying all safety constraints, this method automatically realizes the optimal economic allocation of reactive power resources in the network, effectively improves voltage quality, reduces system network losses, and thus improves the overall operating efficiency and economic benefits of incremental distribution networks. Attached Figure Description

[0028] Figure 1 This invention provides an optimized control structure scheme for incremental distribution networks.

[0029] Figure 2 This is the iterative convergence graph of the spherical search analysis method of this invention. Detailed Implementation

[0030] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1

[0032] To address the need for incremental distribution networks to accommodate a large number of distributed renewable energy sources and diverse loads, this patent aims to provide a computationally efficient distributed voltage and reactive power coordination control method suitable for incremental distribution networks and strictly adhering to grid connection protocols. The core of this invention lies in constructing a hierarchical-distributed hybrid control architecture and proposing an innovative distributed spherical decoupling search algorithm, thereby achieving optimized allocation of reactive power resources within the network and precise restoration of voltage levels.

[0033] A distributed multi-agent adaptive optimization method for incremental power and voltage balance in distribution networks, comprising the following steps:

[0034] Step 1: Data monitoring is performed by voltage and current sensors to collect real-time voltage and current data of each feeder in the system. The intelligent agent calculates the real-time power on each sub-network node using this data, and calculates the reference power using its own and neighboring intelligent agents' real-time power data. The converter achieves decoupled voltage control using this data.

[0035] Step 2: Considering the radial system, construct a "virtual voltage source-sub-network" decomposition model, decompose the system into multiple sub-networks, and decompose the grid-type power supply into two virtual voltage sources. Each sub-network includes virtual voltage sources on both sides, as well as internal loads and grid-type power supplies. Each sub-network acts as an intelligent agent and participates in the operation of the distributed system.

[0036] Step 3: Construct the cost optimization function for each sub-network. Considering the differences in the common access points, the cost optimization function will be different. Based on the spherical search analysis method, the optimal voltage and power points are obtained iteratively, and the reference voltage value that meets the judgment threshold is output to each actual network source. After multiple iterations, the optimal solution of system voltage and power is achieved.

[0037] The above method can effectively address the inherent uncertainties in incremental distribution networks. The method makes decisions based on real-time sensor data (step one) and employs data-driven techniques such as multi-agent reinforcement learning. This means it does not rely on precise mathematical models of the power grid and can autonomously learn and adapt to uncertainties such as fluctuations in renewable energy output and load changes through continuous interaction with the environment.

[0038] Using the "virtual voltage source-subnetwork" decomposition model (step two), the system is divided into multiple autonomous agents. Each agent makes decisions primarily based on local and neighboring information (steps one and three). This architecture reduces reliance on a central controller. Even if one agent fails or experiences a local communication interruption, other agents can still operate independently and coordinate through limited communication, effectively avoiding the risk of single points of failure and thus improving the overall system reliability.

[0039] Each sub-network agent can compute the optimization objective for its region in parallel (step three), and combine it with the "centralized training, distributed execution" framework of multi-agent reinforcement learning. During the deployment phase, the agents can make rapid decisions based on local information, achieving millisecond-level rapid response to the power grid status, which is very suitable for dealing with rapid voltage fluctuations in the distribution network.

[0040] The distributed architecture allows most computations to be performed locally, eliminating the need to transmit all data to a central processing center and reducing the requirements for communication bandwidth. Furthermore, decomposing a large system into multiple sub-problems for parallel optimization avoids the "curse of dimensionality" problem that arises in centralized optimization algorithms as the power grid expands, resulting in higher computational efficiency.

[0041] By constructing a "virtual voltage source-sub-network" model (step two) and designing a cost function for each sub-network in step three, this method ensures that each sub-network is optimized locally (local autonomy) and enables agents to perceive each other's states and coordinate their actions through reference power calculation and iterative update mechanisms, ultimately leading the system to tend toward the globally optimal power and voltage state (global coordination).

[0042] The iterative optimization process based on spherical search analysis (step three) ensures that the system can effectively converge to a satisfactory solution. Related research shows that similar distributed collaborative optimization methods can approach or even reach the global optimum in reducing network losses and voltage deviations.

[0043] The method, through virtual decomposition and coordinated control of grid-connected power sources (step two), can proactively adjust reactive power, effectively suppress voltage exceedance issues caused by renewable energy fluctuations, provide voltage support for the grid, and thus improve the distribution network's ability to absorb distributed renewable energy. This approach can be extended to multi-timescale optimization; for example, optimizing traditional equipment on a slow timescale (hourly level) and optimizing fast-response equipment such as photovoltaic inverters on a fast timescale (minute level), achieving coordinated operation across different timescales and comprehensively improving voltage control quality.

[0044] Example 2

[0045] A distributed multi-agent adaptive optimization incremental power and voltage balance method for distribution networks includes all the contents of Example 1. In addition, in step one, the real-time voltage and current of the source and load feeders of the network are collected, the agents calculate the real-time power, and collect data of themselves and neighboring agents. Through the difference equation, the global power is obtained in the iteration, so that the reference power becomes more and more accurate, and finally the reasonable allocation of the global power of the system is achieved.

[0046] Difference equations:

[0047]

[0048] in Represents a node All active power operations and, , , The estimated power operation is derived from adjacent nodes, and the calculation method for reactive power is similar.

[0049] The decoupled voltage control implemented by the intelligent agent is performed on the grid-type source. Since each grid-type source is decomposed into two virtual voltage sources in step two, this process also occurs on each virtual voltage source. The decomposition of positive and negative sequence components is primarily based on a dual-synchronous reference frame phase-locked loop. This algorithm combines the sequence component method with a cost optimization function to ensure reliable power support from the converter even under transient conditions.

[0050] In converter control, a control strategy that uses positive sequence voltage as reference voltage and suppresses negative sequence voltage to zero is also required. In addition, a low-voltage ride-through strategy needs to be set according to national regulations to maintain voltage stability as much as possible, meet the grid connection requirements of the system, and prevent the system from disconnecting from the grid.

[0051] Example 3

[0052] A distributed multi-agent adaptive optimization method for incremental power and voltage balance of distribution networks includes all the contents of Example 1. In addition, in step two, considering the radial system, a "virtual voltage source-sub-network" decomposition model is constructed. Each sub-network contains one branch of the virtual voltage source in the decomposed network source. Each sub-network includes the load between virtual voltage sources and the network source, and participates in the system voltage and power balance calculation as a sub-network.

[0053] In the "virtual voltage source-sub-network" decomposition model, the grid-type source mainly supplements the active power deficit in the system and does not provide reactive power support. In addition to the regular work of the energy storage power station, the grid-type source also needs to perform voltage balancing and reactive power output.

[0054] The general cost optimization function of the system is:

[0055]

[0056] in It is the first Individual networks, For the first The number of load feeders in each subnetwork For the first The weight of each load feeder For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources It is the first A network-type source has power-constrained Lagrange multipliers. The reference reactive power for grid-connected power sources. For the first The actual reactive power of a grid-type power supply. The reference active power for grid-connected power sources. For the first The actual active power of a grid-type power supply.

[0057] Example 4

[0058] A distributed multi-agent adaptive optimization method for incremental distribution network power and voltage balance, including all the contents of Example 1, further, in step three, considering that the grid side does not need voltage constraints, and according to national regulations, the incremental distribution network should not supply power to the grid in reverse, the optimization cost function of the sub-network near the common access point is:

[0059]

[0060] in It is the number of iterations. For the first The number of load feeders in each subnetwork For the first The weight of each load feeder For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources For the first Voltage-constrained Lagrange multipliers in a network-type source For the first The network-type source in the first Step size in the next iteration It is the cost factor of active power transmitted from the power grid to the system. It is the active power transmitted from the power grid to the system. It is a Lagrange multiplier for PCC point power constraints. It is the amount of power that the system allows the power grid to transmit. For the active power constrained Lagrange multipliers in the first subnetwork;

[0061] , , , The first The voltage magnitudes of the virtual voltage sources on the left and right sides of the energy storage power station.

[0062] Considering the actual physical limitations of grid-connected power sources, the reactive power output of these sources cannot exceed the actual equipment capacity during voltage restoration. Therefore, a penalty term is needed to constrain the reactive power of the grid-connected sources. The optimization cost function for sub-networks far from the common access point is:

[0063]

[0064] in For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources For the first Voltage-constrained Lagrange multipliers in a network-type source For the first The network-type source in the first Step size in the next iteration For the first Lagrange multipliers with active power constraints in a subnetwork;

[0065] , , , The first The magnitudes of the virtual voltage sources on the left and right sides of the energy storage power station;

[0066] It is the reactive power penalty factor of the subnetwork, mainly to prevent the network configuration from exceeding the allowable reactive power support capacity. It is the first The maximum reactive power output capacity that a network-type source can tolerate.

[0067] Considering that the system is an active network, the voltage far from the common access point is relatively weak and has a weak anti-interference ability during the voltage balancing process. Therefore, in the cost optimization function, the weight of the far-end voltage will be higher than that of the voltage near the common access point, so as to take into account the overall voltage stability.

[0068] The voltage weight of each feeder is obtained by calculating the electrical distance, and the calculation method is as follows:

[0069]

[0070] in For the first The scaling factor of the feeder line, For the first The equivalent impedance magnitude of the feeder to the common access point is calculated; then, when the voltage values ​​of the virtual voltage sources of the left and right arms are close in each iteration, the average value of the two voltages is output to the corresponding grid-type source as the optimal parameter to enter the next iteration process, and finally the optimal solution of system voltage and power is achieved.

[0071] The step size of the spherical search analysis algorithm needs to consider both speed and stability; therefore, a dynamic step size scheme is adopted. The formula for the dynamic step size is as follows:

[0072]

[0073] Among them, coefficient This is the scaling factor. This represents the total number of iterations. For the number of iterations, , The first , The voltage of the next iteration.

[0074] When calculating the reference power, the agent needs to perform distributed power averaging estimation, estimating active and reactive power using its own nodes and neighboring nodes. Through a differential algorithm, the global power status is gradually calculated during the iteration process.

[0075] like Figures 1 to 2As shown, the multi-agent adaptive optimization method for incremental distribution network power and voltage balance involved in this invention first proposes a novel optimization scheme that decomposes the complex incremental distribution network into multiple sub-networks, enabling each sub-network to exchange data with neighboring sub-networks. This achieves precise reactive power sharing and voltage recovery, reducing the computational and communication complexity of the incremental distribution network. The system can utilize inexpensive digital control chips, effectively reducing system costs. Second, it improves the transient stability of the system by decoupling the positive and negative sequence control of the system voltage, enhancing the system's anti-interference capability after a fault. Finally, a novel spherical search analysis method guides the search towards the global optimum by employing a search direction that satisfies all network constraints and appropriate step size control. The optimization method proposed in this invention can improve the power and voltage balance capability of incremental distribution networks.

[0076] In this method, we consider the incremental distribution network voltage and power balance problem of multi-agent adaptive optimization. Figure 1 It is an optimization model for incremental distribution networks, which restricts the grid-type power source of each node to two virtual voltage sources. The left and right arms belong to different sub-networks. Each sub-network consists of the right arm of the previous grid-type power source, the left arm of the next energy storage power station, the load connected to the intermediate line, and the grid-type power source.

[0077] When constructing the cost optimization function, each sub-network is used as a unit to construct a cost optimization function that includes the energy storage power stations on both sides and the load. This invention proposes a spherical search analysis method for decoupling the two virtual voltage sources. During the optimization process, the network tends towards the optimal result by making the voltages of the two virtual voltage sources close together. Figure 2 It describes the change of the decoupled left and right arm voltages in the cost optimization function during the iteration process. Figure 2 As can be seen from the process, the voltages of the left and right arms attract each other during the iteration process, and the global optimal solution is obtained in the form of distributed decoupling.

[0078] This method reduces the computational complexity of the system, allowing agents to use low-cost digital signal processors. Local computation improves the sensitivity of the subnetwork in local perception, enabling the system to sense and adjust the voltage of its own nodes more quickly, thus maintaining the normal operation of the incremental distribution network.

Claims

1. A distributed multi-agent adaptive optimization incremental power-voltage balancing method for power distribution networks, characterized in that, The specific steps are as follows: Step 1: Data monitoring is performed by voltage and current sensors to collect real-time voltage and current data of each feeder in the system. The intelligent agent calculates the real-time power on each sub-network node using this data, and calculates the reference power using its own and neighboring intelligent agents' real-time power data. The converter achieves decoupled voltage control using this data. Step 2: Considering the radial system, construct a "virtual voltage source-sub-network" decomposition model, decompose the system into multiple sub-networks, and decompose the grid-type power supply into two virtual voltage sources. Each sub-network includes virtual voltage sources on both sides, as well as internal loads and grid-type power supplies. Each sub-network acts as an intelligent agent and participates in the operation of the distributed system. Step 3: Construct the cost optimization function for each sub-network. Considering the differences in the common access points, the cost optimization function will be different. Based on the spherical search analysis method, the optimal voltage and power points are obtained iteratively, and the reference voltage value that meets the judgment threshold is output to each actual network source. After multiple iterations, the optimal solution of system voltage and power is achieved.

2. The incremental power and voltage balance method for distributed multi-agent adaptive optimization of distribution networks according to claim 1, characterized in that, In step one, the real-time voltage and current of the source and load feeders in the network are collected, the agent calculates the real-time power, and collects data from itself and neighboring agents. Through the difference equation, the global power is obtained in repeated iterations, so that the reference power becomes more and more accurate, and finally the reasonable allocation of the global power of the system is achieved.

3. The incremental power and voltage balance method for distributed multi-agent adaptive optimization of distribution networks according to claim 2, characterized in that, The difference equation: ； in Represents a node All active power operations and, , , The estimated power operation is derived from adjacent nodes, and the calculation method for reactive power is similar.

4. The incremental power and voltage balance method for distributed multi-agent adaptive optimization according to claim 1, characterized in that, In step two, considering the radial system, a "virtual voltage source-sub-network" decomposition model is constructed. Each sub-network contains one branch of the virtual voltage source in the decomposed network source. Each sub-network includes the load between virtual voltage sources and the network source, and participates in the system voltage and power balance calculation as a sub-network.

5. The incremental power and voltage balance method for distributed multi-agent adaptive optimization of distribution networks according to claim 4, characterized in that, In the "virtual voltage source-sub-network" decomposition model, the grid-type source mainly supplements the active power deficit in the system and does not provide reactive power support. In addition to the regular work of the energy storage power station, the grid-type source also needs to perform voltage balancing and reactive power output. The general cost optimization function of the system is: ； in It is the first Individual networks, For the first The number of load feeders in each subnetwork For the first The weight of each load feeder For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources It is the first A network-type source has power-constrained Lagrange multipliers. The reference reactive power for grid-connected power sources. For the first The actual reactive power of a grid-type power supply. The reference active power for grid-connected power sources. For the first The actual active power of a grid-type power supply.

6. The incremental power and voltage balance method for distributed multi-agent adaptive optimization of distribution networks according to claim 1, characterized in that, Furthermore, in step three, considering that voltage constraints are not required on the grid side, the incremental distribution network should not supply power to the grid in reverse. Therefore, the optimal cost function of the sub-network near the common access point is: ； in It is the number of iterations. For the first The number of load feeders in each subnetwork For the first The weight of each load feeder For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources For the first Voltage-constrained Lagrange multipliers in a network-type source For the first The network-type source in the first Step size in the next iteration It is the cost factor of active power transmitted from the power grid to the system. It is the active power transmitted from the power grid to the system. It is a Lagrange multiplier for PCC point power constraints. It is the amount of power that the system allows the power grid to transmit. For the active power constrained Lagrange multipliers in the first subnetwork; , , , The first The voltage magnitudes of the virtual voltage sources on the left and right sides of the energy storage power station.

7. The incremental power and voltage balance method for distributed multi-agent adaptive optimization according to claim 6, characterized in that, Considering the actual physical limitations of grid-connected power sources, the reactive power output of these sources cannot exceed the actual equipment capacity during voltage restoration. Therefore, a penalty term is needed to constrain the reactive power of the grid-connected sources. The optimization cost function for sub-networks far from the common access point is: ； in For the first Positive sequence voltage of the load feeder. For the first Reference voltage of the load feeder It is the first Reactive power constraint weights for individual grid-type sources For the first Voltage-constrained Lagrange multipliers in a network-type source For the first The network-type source in the first Step size in the next iteration For the first Lagrange multipliers with active power constraints in a subnetwork; , , , The first The magnitudes of the virtual voltage sources on the left and right sides of the energy storage power station; It is the reactive power penalty factor of the subnetwork, mainly to prevent the network configuration from exceeding the allowable reactive power support capacity. It is the first The maximum reactive power output capacity that a network-type source can tolerate.

8. The incremental power and voltage balance method for distributed multi-agent adaptive optimization of distribution networks according to claim 7, characterized in that, Considering that the system is an active network, the voltage far from the common access point is relatively weak and has a weak anti-interference ability during the voltage balancing process. Therefore, in the cost optimization function, the weight of the far-end voltage will be higher than that of the voltage near the common access point, so as to take into account the overall voltage stability. The voltage weight of each feeder is obtained by calculating the electrical distance, and the calculation method is as follows: ； in For the first The scaling factor of the feeder line, For the first The equivalent impedance modulus of the feeder to the common access point is calculated; then, when the voltage values ​​of the virtual voltage sources of the left and right arms are close in each iteration, the average value of the two voltages is output to the corresponding grid-type source as the optimal parameter to enter the next iteration process, and finally the optimal solution of system voltage and power is achieved.

9. A distributed multi-agent adaptive optimization method for incremental power and voltage balance in distribution networks according to claim 8, characterized in that, The step size of the spherical search analysis algorithm needs to consider both speed and stability; therefore, a dynamic step size scheme is adopted. The formula for the dynamic step size is as follows: ； Among them, coefficient This is the scaling factor. This represents the total number of iterations. For the number of iterations, , The first , The voltage of the next iteration.

10. A distributed multi-agent adaptive optimization method for incremental power and voltage balance in distribution networks according to claim 9, characterized in that, When calculating the reference power, the agent needs to perform distributed power averaging estimation, estimating active and reactive power through its own nodes and neighboring nodes, and gradually calculating the global power status through a differential algorithm during the iteration process.

Images