A digital power energy storage system dynamic network reconfiguration method

Abstract

The application relates to a kind of digital power energy storage system dynamic network reconfiguration methods in the technical field of power energy storage, S1. Build energy storage system;It is composed of intelligent layer, control layer and data layer;S2. Energy storage network modeling;Adopt graph capsule neural network to learn the optimal control strategy of energy storage network;S3. Dynamic network reconfiguration;Based on Markov decision process, generate fault emergency response model;By systematic optimization of the topological structure and optimal control strategy of energy storage network, the quick isolation of fault area is realized, and the dynamic adjustment and reconfiguration of the working mode of energy storage equipment and load equipment are realized, to ensure that the energy storage network can still operate stably under fault situation, improve the flexibility and elasticity (fault tolerance) of energy storage system in dynamic environment;By reducing the energy loss in the transmission process, balancing network load and preventing local failure, the technical problems existing in the prior art are solved, and the sustainable development of new renewable energy system is promoted.

Description

Technical Field

[0001] This invention belongs to the field of power energy storage technology, and specifically provides a method for dynamic network reconfiguration of a digital power energy storage system. Background Technology

[0002] Energy storage units, as a crucial component of power systems, can convert and store energy in the form of electricity for unforeseen needs. They play a key role in the flexibility and stability of the power grid and are also an important means of promoting the widespread application of renewable energy and improving energy efficiency. Currently, new renewable energy generation technologies are rapidly developing as an important part of new power systems. Based on their characteristics, energy storage technologies are mainly classified into energy-type and power-type energy storage technologies. Common energy-type energy storage technologies include compressed air energy storage, pumped hydro storage, and battery energy storage, while common power-type energy storage technologies include flywheel energy storage, supercapacitor energy storage, and superconducting energy storage. Traditional energy storage units constantly face economic constraints and safety concerns, while digital energy storage can discretize energy and information flows, improving the economic efficiency and safety of energy storage systems.

[0003] Energy storage systems possess a certain degree of resilience in maintaining their functionality and performance when faced with external disturbances or changes (such as fluctuations in electricity demand, instability in renewable energy generation, equipment failures, and natural disasters). However, as energy storage system networks expand their coverage and the connections between nodes become more complex, the interactions and interdependencies between subnetworks or nodes are gradually increasing. This brings challenges such as network interference migration and the possibility of cascading failures, which weaken the resilience of energy storage systems to some extent. Dynamically reconstructing the network topology of energy storage systems and rationally deploying network connections can effectively isolate faulty areas, thereby improving the utilization efficiency of energy storage resources, reducing energy waste and losses, and promoting the integration of more renewable energy sources, thus reducing dependence on traditional fossil fuels. Therefore, combining energy storage with digital technologies to achieve dynamic reconstruction of the energy storage network topology, fault isolation, optimization of transmission paths, and optimization of energy storage status can provide important technical support for promoting energy transition and the development of renewable energy.

[0004] When energy storage networks face topology changes such as faults, existing solutions mainly focus on two approaches: one is heuristic routing optimization algorithms, which optimize link weight settings or splitting ratios under different traffic demands or network topologies. However, when faced with network topology changes, the previous routing configuration cannot adapt to the new environment, requiring recalculation of routing strategies, which is time-consuming and computationally expensive. The other approach focuses on deep reinforcement learning based on neural networks, suitable for handling complex dynamic environments. However, when the network topology changes, the trained neural network model cannot generalize well to other network topologies, and retraining the model still consumes significant time and computational resources. Existing research has significant limitations in energy storage system optimization and fault management under dynamic environments. Most methods still focus on optimization under static conditions, ignoring the dynamic changes in complex power grid environments and the dependencies between multiple nodes. Especially in complex energy storage networks with multiple nodes and multiple energy types, how to achieve real-time adaptive reconfiguration of the network topology to cope with uncertainty and sudden faults remains a key problem that has not been effectively solved. Therefore, how to design a general fault management method that can adaptively face various network fault situations is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] To address the above issues, this invention provides a dynamic network reconfiguration method for digital power storage systems. Based on the rapid development of graph neural networks and reinforcement learning, this method captures the complex dependencies between system nodes and adjusts the system configuration in real time to achieve adaptive optimization and fault isolation. This overcomes the limitations of existing methods (such as the inability to dynamically reconfigure the network and the lack of robustness in emergency response), thereby improving the flexibility and resilience (fault tolerance) of energy storage systems in dynamic environments. By reducing energy loss during transmission, balancing network load, and preventing local faults, this invention solves the technical problems existing in the prior art and promotes the sustainable development of new renewable energy systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A dynamic network reconfiguration method for a digital power storage system includes the following steps:

[0008] S1. Energy Storage System Construction; The energy storage system consists of an intelligent layer, a control layer, and a data layer; The intelligent layer acquires the current state of the energy storage system and extracts the node feature information of the energy storage network based on a graph neural network. It makes decisions through a reinforcement learning agent, and the decision-making action information is transmitted to the control layer; The control layer receives the action information from the intelligent layer, generates control commands according to rules, and transmits the control commands to the data layer; The data layer includes the energy storage network, which includes power generation equipment, energy storage equipment, load equipment, and a control center. The energy storage network receives control commands from the control layer, makes adjustments, updates the energy storage network, sends the update information to the control center, and simultaneously feeds the update information back to the control layer, which then transmits it to the intelligent layer;

[0009] S2. Energy Storage Network Modeling: Graph Capsule Neural Networks (GCNs) are used to learn the optimal control strategy in energy storage networks. Compared with other GCNs, they are more effective at capturing the structural information of energy storage networks, and the enhanced state representation helps to calculate better actions than traditional feature abstraction networks (such as MLPs). Global information of the energy storage system is gradually extracted and optimized through GCN construction, feature extraction, capsule encoding, dynamic routing, and GCN updates. First, GCNs are used to extract node embeddings to form primary capsules. Then, an attention module is applied to scale the node embeddings, followed by dynamic routing to generate graph capsules. Finally, dynamic routing is applied again to perform graph classification.

[0010] A graph capsule neural network is constructed using the devices and connections in the energy storage network as nodes and edges, respectively. It can be represented as a graph structure G=(N,E), where nodes represent energy storage devices, substations, and loads, and edges represent physical or logical connections between devices, such as transmission lines or communication links.

[0011] Based on graph capsule neural networks, corresponding features are extracted for each node and edge, such as the capacity, charging and discharging status, and location of the energy storage device, as well as the transmission capacity and distance of the edge.

[0012] The degenerate capsule neural network encodes the features of each node and edge. Each node and edge is represented as a capsule, which consists of a set of feature vectors. These feature vectors not only describe the properties of the nodes and edges, but also capture the spatial and functional relationships between them.

[0013] By using dynamic routing algorithms, information is transmitted between capsules, and features are aggregated and combined layer by layer to capture high-order structural information of the energy storage network.

[0014] Finally, by combining the message passing mechanism of the graph capsule neural network, the feature vectors of the capsules are used to iteratively update the graph structure, gradually extracting global information of the energy storage system, and achieving a deep understanding and optimization of the entire network.

[0015] S3. Dynamic Network Reconfiguration: In order to quickly implement effective emergency response measures when a fault occurs, a combined control method based on the concept of energy discretization and digitization is proposed, which covers two modules: energy storage network topology reconfiguration and energy storage equipment and load equipment reconfiguration.

[0016] This method achieves rapid isolation of faulty areas, dynamically adjusts the working mode of energy storage devices, and reconfigures the operating status of energy storage devices and load devices by systematically optimizing the topology and operation strategy of the energy storage network. This ensures that the energy storage network can maintain efficient and stable operation under various fault scenarios.

[0017] Specifically, the core of this method lies in discretizing and digitizing energy flow to more precisely manage and control the energy distribution and transmission of each node and edge.

[0018] This method follows Kirchhoff's laws to change the line impedance and rearranges the network structure through switching operations, thereby mitigating faults. It enables the system to respond quickly when a fault occurs, flexibly adjust the network topology and operating strategies, minimize the impact of faults on the overall system, and improve fault recovery efficiency.

[0019] After isolating the faulty area and reconstructing the network topology, the system can further optimize the network's operating efficiency and stability by actively adjusting the charging and discharging status of the energy storage devices.

[0020] When the network topology changes, the system will not only redistribute energy transmission paths, but also dynamically adjust the working mode of energy storage devices based on current load demand, fault conditions, and the status of each energy storage node.

[0021] Energy storage devices with high energy storage capacity can be regarded as power sources. When a faulty node cannot connect to a power generation node, the energy storage device discharges to ensure that the isolated area that is not connected to the power generation node can still work for a period of time.

[0022] Fault emergency response in energy storage networks can be represented as a Markov decision process (MDP) in the graph domain, denoted as This tuple represents the state, action, transition probability, and reward, as defined below:

[0023] (1) State (S):

[0024] The state consists of relevant observations within the energy storage network, representing the network's current operational status. It includes node variables, edge variables, network topology, and other system variables, represented as follows:

[0025]

[0026] in, Represents the estimated or predicted values ​​of the active and reactive power demand of a node; Represents the active and reactive power generation of a node; This represents the three-phase voltage measured on the bus, that is, at the graph node; This represents the voltage constraint violation value in the network; edge variables include the power flow through network branches, denoted as... The network's operating topology is as follows: The total energy supply in the network is represented as State variables This includes fault scenarios, namely multi-line faults in the network, including switch faults;

[0027] The inoperability of fault switches is handled through a shielding mechanism, using state variables. This means that if the switch operates normally, then... If the switch cannot be operated due to a circuit fault, then The system will disable the switch and will no longer perform any operations on it.

[0028] (2) Action (A):

[0029] Emergency response control actions include switching operations, energy storage device state of charge (SOC) operations, and load operating state operations. Therefore, the action space is represented as follows:

[0030]

[0031] in, This indicates the number of switchable lines, including sectionalizing switches and tie switches. The number of operable energy storage devices in the network is expressed as... .

[0032] The circuit switch uses binary variables The value 0 indicates that the switch is open, and 1 indicates that the switch is closed. The State of Charge (SOC) of an energy storage device is determined by variables. The values ​​indicate that 1 represents the energy storage device charging, -1 represents the energy storage device discharging, and 0 represents the energy storage device being idle.

[0033] The working status of the load is determined by The value 1 indicates that the load is in a working state, and 0 indicates that the load is in an idle state.

[0034] (3) Transition probability :

[0035] The transition probability captures the dynamic characteristics of the network under emergency response conditions, and is expressed as...

[0036] .

[0037] This means at time step Perform actions After that, the network status changes from the time step. status Shift to time step status The transition probabilities are learned through the interaction between the agent and the environment.

[0038] (4) Rewards (R):

[0039] The reinforcement learning algorithm on the reward-guided graph adopts optimal control actions to mitigate the impact of faults in the energy storage network, and its formula is as follows:

[0040] (1)

[0041] The reward reflects the achievement of maximizing the supply of energy. Violation of minimizing voltage constraints The goal is to improve the resilience of energy storage networks.

[0042] To account for network imbalances under specific fault conditions and switching operations, operational constraints on the energy storage network are introduced into the rewards. Actively avoid misalignment and constraint violations caused by malfunctions or switching operations.

[0043] Due to topology changes caused by faults and switching operations, energy storage networks may consist of multiple independent network parts, each with different active components (such as transformers, regulators, generators, loads, energy storage devices, etc.) and their corresponding state variables.

[0044] In some scenarios, the isolation of these components from a strong backup power source (substation) causes network imbalance, making it difficult to achieve node power balance within a preset error range.

[0045] This misalignment causes the power flow to fail to converge, which is identified by a flag in the solver. At this point, the reward value is zero because the solver cannot accurately reflect the effect of the switch on the network state.

[0046] On the other hand, network operation with high voltage violations is not feasible because it would cause the network to collapse immediately.

[0047] To prevent agents from taking actions that would result in invalid states, a penalty item has been added to the rewards. The goal is to maintain voltage levels within acceptable ranges to ensure the sustainability of network operation.

[0048] For exceeding its upper limit and lower limit Each bus The voltage violation, after power flow estimation, is evaluated as follows:

[0049] (2)

[0050] (3)

[0051] in, This represents the set of connections on the bus. Voltage measurements and supplied energy are estimated in units (pu) and relative to the reference voltage of the corresponding network. and reference power Perform the calculation.

[0052] Per-unit calculations in power systems eliminate the unit problem, essentially normalizing them using a reference value:

[0053] (4)

[0054] in, It is the cardinality of the network bus set. and These represent violations exceeding the maximum and minimum voltage limits, respectively.

[0055] Therefore, when a fault occurs, the agent can not only quickly identify the fault area and reconfigure the switch status, but also reconfigure the charging and discharging status of the energy storage devices and the load configuration according to the actual network topology, thereby effectively isolating the fault and ensuring the normal operation of other parts of the network. In a dynamic environment, through repeated exploration and experience accumulation, the agent can find an optimal path to minimize the impact of the fault on network operation.

[0056] Furthermore, the dynamic network reconstruction method includes the construction of a learning architecture, which combines graph capsule neural networks and policy gradient methods to build a policy network for learning energy storage network control and output corresponding actions to ensure the efficiency and safety of system operation.

[0057] Each node in the energy storage network diagram Having active power demand Reactive power demand These node attributes are first projected onto a high-dimensional space through a linear transformation to generate initial node features.

[0058]

[0059] in, The weight matrix is ​​a learnable matrix. For learnable bias vectors, This represents the projected dimension. These features are then processed through a multi-layer graph capsule neural network, where each layer extracts node embeddings layer by layer using a polynomial-form graph convolutional filter.

[0060] (5)

[0061] in, For the graph Laplace matrix, Let k be the order of the statistical moments, and k be the order of the convolution filter. For the first The output of the layer, The weight matrix is ​​learnable. Node embeddings not only contain information about the node itself, but also integrate information aggregated from multiple neighboring nodes.

[0062] As the number of graph convolutional layers increases and the order of convolutional filters is improved, the model is able to capture more complex graph structure information.

[0063] The output of each layer is concatenated with the convolution results of all orders to form the final node embedding:

[0064] (6)

[0065] After linear transformation, the final node embedding matrix is ​​generated. Subsequently, the node embeddings are processed through a series of linear transformations to generate a global graph embedding:

[0066] (7)

[0067] This embedding is obtained by averaging the embedding values ​​of all nodes, thus providing a global state representation for the entire power distribution network.

[0068] Besides graph structure information, some state variables in energy storage networks cannot be directly represented by nodes in the graph, such as energy supply, voltage violations, and power flow at edges. This information is crucial for the network's operational safety and operability; therefore, it is necessary to generate context embeddings as additional state information using feedforward neural networks.

[0069] (8)

[0070] This contextual information, together with the graph embedding, constitutes the state representation of the entire system. This is then further processed by a multilayer perceptron to generate the final state embedding.

[0071] (9)

[0072] Furthermore, the dynamic network reconstruction method also includes a model training process, as detailed below:

[0073] (1) Data preparation and scenario generation:

[0074] Faults in energy storage systems often originate from localized failures, which can lead to cascading effects.

[0075] Specifically, the faults in the network are mainly caused by line faults, which are simulated using a graph capsule neural network-based method.

[0076] To simulate this behavior, a subgraph method with random edge removal was employed.

[0077] This method involves randomly selecting nodes from a graph capsule neural network of an energy storage system. and create nodes with different radii around these nodes. The subgraph with the maximum radius.

[0078] set up ,in It is the diameter of the diagram.

[0079] Within each selected subgraph, randomly remove a portion of the edges. To simulate the effects of local accidents.

[0080] The proportion of edge failures gradually increased from 0 to 50%.

[0081] By changing the node Subgraph radius and remove edges Scenarios with multiple line faults are generated for model training.

[0082] In addition, in each scenario, the load multiplier and generator point are varied by randomly selecting the multiplier and generator point from the annual configuration file in the OpenDSS package with hourly resolution.

[0083] (2) Training process:

[0084] The training process generates samples simulating various fault scenarios on the energy storage network by introducing line faults, adjusting loads and energy storage device capacities, and considering different fault conditions.

[0085] The network faults are mainly caused by power distribution line faults, which are simulated using a graph capsule neural network-based method. Load demand and energy storage capacity, two key operational parameters for the energy storage network, are randomly selected from annual load curves provided by OpenDSS.

[0086] Considering that the state and action spaces expand rapidly with increasing network size, fault scenarios, and the number of devices, methods such as Q-learning cannot provide sufficient storage and computational resources in practice. Therefore, the Proximal Policy Optimization (PPO) algorithm is used to train the policy network.

[0087] The training process involves collecting experience in the form of tuples containing state, action, reward, and next state. PPO is run on a rollout basis, with each operation consisting of a fixed number of N steps. Weight updates are performed in batches of N (≤N steps) after the rollout, minimizing the cost function containing policy gradient loss and state value loss through backpropagation. The policy network is trained for a total of Ntotal steps.

[0088] The present invention also includes other components that enable its normal use, all of which are conventional means in the art. In addition, any devices or components not limited in the present invention adopt the prior art in the art.

[0089] The beneficial effects of this invention are as follows:

[0090] 1. To address the issue that traditional numerical or image data are insufficient to effectively represent the interconnections between nodes and edges in large and complex networks, thus failing to clearly define the dynamic characteristics and operational status of energy storage networks, this paper utilizes the concept of graph neural networks to represent energy storage networks as graphs and uses them as state representations for reinforcement learning, thereby improving the real-time performance and accuracy of energy storage network management.

[0091] 2. The switching control method based on the concept of energy discretization and digitization can reconstruct the network through switching operations when a fault occurs, and isolate the fault area. This can significantly enhance the robustness and fault tolerance of the energy storage system, thereby ensuring the safety of the system. It overcomes the phenomenon that traditional energy storage networks cannot automatically reconstruct the network when a local fault occurs, which leads to fault propagation and waste of resources, thereby improving the economic efficiency of the system.

[0092] 3. A smart control mechanism for energy storage devices based on dynamic topology reconfiguration is proposed. This mechanism actively adjusts the charging and discharging state of energy storage devices, enabling flexible responses to load demands and fault conditions when the network topology changes. When necessary, it can also act as a power source to discharge, ensuring continuous power supply in isolated areas, thereby optimizing network operating efficiency and stability. Attached Figure Description

[0093] Figure 1 This is a schematic diagram of the energy storage system structure in Example 1.

[0094] Figure 2 This is a schematic diagram of capsule neural network modeling in Example 1.

[0095] Figure 3 This is a schematic diagram of the reinforcement learning process in Example 2.

[0096] Figure 4 This is the convergence graph of the reward function during policy model training in Example 3; where, Figure 4 a is a comparison chart of DRO and MLP for the IEEE 34-bus distribution test network. Figure 4 b is a comparison diagram of DRO and MLP of the IEEE 123 bus distribution test network.

[0097] Figure 5 These are the states of the decision variables obtained from different models in three scenarios in Example 4; where, Figure 5 'a' represents the switch decision result diagram of the IEEE 34 bus distribution test network, with dark color indicating switch closed and light color indicating switch open. Figure 5 b represents the decision result diagram of energy storage devices in the IEEE 34 bus distribution test network. Yellow indicates that the energy storage device is idle, green indicates that the energy storage device is discharging, and red indicates that the energy storage device is charging. Figure 5 c represents the load status decision result of the IEEE 34 bus distribution test network. Dark color indicates that the load is working, and light color indicates that the load is idle.

[0098] Figure 6 This is a schematic diagram of the DRO optimization results based on fault scenario 1 on the IEEE 34 bus distribution test network in Example 4.

[0099] Figure 7 These are the states of the decision variables obtained from different models in three scenarios in Example 4; where, Figure 7 'a' represents the switch decision result diagram of the IEEE 123 bus distribution test network, with dark color indicating switch closed and light color indicating switch open. Figure 7 b represents the decision result diagram of energy storage devices in the IEEE 123 bus distribution test network. Yellow indicates that the energy storage device is idle, green indicates that the energy storage device is discharging, and red indicates that the energy storage device is charging. Figure 7 c represents the load status decision result of the IEEE 123 bus distribution test network. Dark color indicates that the load is working, and light color indicates that the load is idle.

[0100] Figure 8 This is a schematic diagram of the DRO optimization results based on fault scenario 1 on the IEEE 123 bus distribution test network in Example 4.

[0101] Figure 9 This is a comparison diagram of the energy provided by the IEEE 34-bus distribution test network and the IEEE 123-bus distribution test network under different model decisions in Example 4; wherein, Figure 9A comparison chart showing the power provided by an IEEE 34-bus distribution test network under three different fault conditions; Figure 9 b. Comparison of power supply provided by the IEEE 123 bus distribution test network under three different fault conditions.

[0102] Figure 10 This example compares the voltage constraint violations of the IEEE 34-bus distribution test network and the IEEE 123-bus distribution test network under different model decisions in Example 4; whereby, Figure 10 A comparison of voltage constraint violations in an IEEE 34-bus distribution test network under three different fault conditions; Figure 10 b. Comparison of voltage constraint violations in the IEEE 123 bus distribution test network under three different fault conditions. Detailed Implementation

[0103] The technical solution of the present invention will be clearly and completely described below with reference to specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0104] Example 1

[0105] A dynamic network reconfiguration method for a digital power storage system, comprising the following steps:

[0106] S1. Energy storage system construction; such as Figure 1 As shown, the energy storage system consists of an intelligent layer, a control layer, and a data layer. The intelligent layer acquires the current state of the energy storage system and extracts the node feature information of the energy storage network based on a graph neural network. It makes decisions through a reinforcement learning agent, and the decision-making action information is transmitted to the control layer. The control layer receives the action information from the intelligent layer, generates control commands according to rules, and transmits the control commands to the data layer. The data layer includes the energy storage network, which includes power generation equipment, energy storage equipment, load equipment, and a control center. The energy storage network receives control commands from the control layer, makes adjustments, updates the energy storage network, sends the update information to the control center, and simultaneously feeds the update information back to the control layer, which then transmits it to the intelligent layer.

[0107] S2. Energy Storage Network Modeling: Graph capsule neural networks are used to learn the optimal control strategy in the energy storage network. Compared with other graph capsule neural networks, they can more effectively capture the structural information of the energy storage network. The enhanced state representation helps to calculate better actions than traditional feature abstraction networks (such as MLP). Through graph capsule neural network construction, feature extraction, capsule encoding, dynamic routing, and graph capsule neural network updates, the global information of the energy storage system is gradually extracted and optimized.

[0108] like Figure 2As shown, firstly, a graph capsule neural network is used to extract node embeddings to form primary capsules; then, an attention module is applied to scale the node embeddings, followed by dynamic routing to generate graph capsules; finally, dynamic routing is applied again to perform graph classification.

[0109] A graph capsule neural network is constructed using the devices and connections in the energy storage network as nodes and edges, respectively. It can be represented as a graph structure G=(N,E), where nodes represent energy storage devices, substations, and loads, and edges represent physical or logical connections between devices, such as transmission lines or communication links.

[0110] Based on graph capsule neural networks, corresponding features are extracted for each node and edge, such as the capacity, charging and discharging status, and location of the energy storage device, as well as the transmission capacity and distance of the edge.

[0111] The degenerate capsule neural network encodes the features of each node and edge. Each node and edge is represented as a capsule, which consists of a set of feature vectors. These feature vectors not only describe the properties of the nodes and edges, but also capture the spatial and functional relationships between them.

[0112] By using dynamic routing algorithms, information is transmitted between capsules, and features are aggregated and combined layer by layer to capture high-order structural information of the energy storage network.

[0113] Finally, by combining the message passing mechanism of the graph capsule neural network, the feature vectors of the capsules are used to iteratively update the graph structure, gradually extracting global information of the energy storage system, and achieving a deep understanding and optimization of the entire network.

[0114] S3. Dynamic Network Reconfiguration: In order to quickly implement effective emergency response measures when a fault occurs, a combined control method based on the concept of energy discretization and digitization is proposed, which covers two modules: energy storage network topology reconfiguration and energy storage equipment and load equipment reconfiguration.

[0115] This method achieves rapid isolation of faulty areas, dynamically adjusts the working mode of energy storage devices, and reconfigures the operating status of energy storage devices and load devices by systematically optimizing the topology and operation strategy of the energy storage network. This ensures that the energy storage network can maintain efficient and stable operation under various fault scenarios.

[0116] Specifically, the core of this method lies in discretizing and digitizing energy flow to more precisely manage and control the energy distribution and transmission of each node and edge.

[0117] This method follows Kirchhoff's laws to change the line impedance and rearranges the network structure through switching operations, thereby mitigating faults. It enables the system to respond quickly when a fault occurs, flexibly adjust the network topology and operating strategies, minimize the impact of faults on the overall system, and improve fault recovery efficiency.

[0118] After isolating the faulty area and reconstructing the network topology, the system can further optimize the network's operating efficiency and stability by actively adjusting the charging and discharging status of the energy storage devices.

[0119] When the network topology changes, the system will not only redistribute energy transmission paths, but also dynamically adjust the working mode of energy storage devices based on current load demand, fault conditions, and the status of each energy storage node.

[0120] Energy storage devices with high energy storage capacity can be regarded as power sources. When a faulty node cannot connect to a power generation node, the energy storage device discharges to ensure that the isolated area that is not connected to the power generation node can still work for a period of time.

[0121] Example 2

[0122] Based on the dynamic network reconfiguration method for digital power storage systems described in Example 1, the fault emergency response in the energy storage network is represented as a Markov decision process (MDP) in the graph domain, denoted as... ;

[0123] like Figure 3 As shown, this tuple represents the state, action, transition probability, and reward, as defined below:

[0124] (1) State (S):

[0125] The state consists of relevant observations within the energy storage network, representing the network's current operational status. It includes node variables, edge variables, network topology, and other system variables, represented as follows:

[0126]

[0127] in, Represents the estimated or predicted values ​​of the active and reactive power demand of a node; Represents the active and reactive power generation of a node; This represents the three-phase voltage measured on the bus, that is, at the graph node; This represents the voltage constraint violation value in the network; edge variables include the power flow through network branches, denoted as... The network's operating topology is as follows: The total energy supply in the network is represented as State variables This includes fault scenarios, namely multi-line faults in the network, including switch faults;

[0128] The inoperability of fault switches is handled through a shielding mechanism, using state variables. This means that if the switch operates normally, then... If the switch cannot be operated due to a circuit fault, then The system will disable the switch and will no longer perform any operations on it.

[0129] (2) Action (A):

[0130] Emergency response control actions include switching operations, energy storage device state of charge (SOC) operations, and load operating state operations. Therefore, the action space is represented as follows:

[0131]

[0132] in, This indicates the number of switchable lines, including sectionalizing switches and tie switches. The number of operable energy storage devices in the network is expressed as... .

[0133] The circuit switch uses binary variables The value is represented as 0 (0 indicates the switch is open) and 1 (1 indicates the switch is closed). The State of Charge (SOC) of an energy storage device is determined by variables. The values ​​indicate that 1 represents the energy storage device charging, -1 represents the energy storage device discharging, and 0 represents the energy storage device being idle.

[0134] The working status of the load is determined by The value 1 indicates that the load is in a working state, and 0 indicates that the load is in an idle state.

[0135] (3) Transition probability :

[0136] The transition probability captures the dynamic characteristics of the network under emergency response conditions, and is expressed as...

[0137] .

[0138] This means at time step Perform actions After that, the network status changes from the time step. status Shift to time step status The transition probabilities are learned through the interaction between the agent and the environment.

[0139] (4) Rewards (R):

[0140] The reinforcement learning algorithm on the reward-guided graph adopts optimal control actions to mitigate the impact of faults in the energy storage network, and its formula is as follows:

[0141] (1)

[0142] The reward reflects the achievement of maximizing the supply of energy. Violation of minimizing voltage constraints The goal is to improve the resilience of energy storage networks.

[0143] To account for network imbalances under specific fault conditions and switching operations, operational constraints on the energy storage network are introduced into the rewards. Actively avoid misalignment and constraint violations caused by malfunctions or switching operations.

[0144] Due to topology changes caused by faults and switching operations, energy storage networks may consist of multiple independent network parts, each with different active components (such as transformers, regulators, generators, loads, energy storage devices, etc.) and their corresponding state variables.

[0145] In some scenarios, the isolation of these components from a strong backup power source (substation) causes network imbalance, making it difficult to achieve node power balance within a preset error range.

[0146] This misalignment causes the power flow to fail to converge, which is identified by a flag in the solver. At this point, the reward value is zero because the solver cannot accurately reflect the effect of the switch on the network state.

[0147] On the other hand, network operation with high voltage violations is not feasible because it would cause the network to collapse immediately.

[0148] To prevent agents from taking actions that would result in invalid states, a penalty item has been added to the rewards. The goal is to maintain voltage levels within acceptable ranges to ensure the sustainability of network operation.

[0149] For exceeding its upper limit and lower limit Each bus The voltage violation, after power flow estimation, is evaluated as follows:

[0150] (2)

[0151] (3)

[0152] in, This represents the set of connections on the bus. Voltage measurements and supplied energy are estimated in units (pu) and relative to the reference voltage of the corresponding network. and reference power Perform the calculation.

[0153] Per-unit calculations in power systems eliminate the unit problem, essentially normalizing them using a reference value:

[0154] (4)

[0155] in, It is the cardinality of the network bus set. and These represent violations exceeding the maximum and minimum voltage limits, respectively.

[0156] Therefore, when a fault occurs, the agent can not only quickly identify the fault area and reconfigure the switch status, but also reconfigure the charging and discharging status of the energy storage devices and the load configuration according to the actual network topology, thereby effectively isolating the fault and ensuring the normal operation of other parts of the network. In a dynamic environment, through repeated exploration and experience accumulation, the agent can find an optimal path to minimize the impact of the fault on network operation.

[0157] Example 3

[0158] Based on the dynamic network reconfiguration method for digital power storage systems described in Example 2, the dynamic network reconfiguration method also includes the construction of a learning architecture. It combines graph capsule neural networks and policy gradient methods to construct a policy network for learning energy storage network control and outputs corresponding actions to ensure the efficiency and safety of system operation.

[0159] Each node in the energy storage network diagram Having active power demand Reactive power demand These node attributes are first projected onto a high-dimensional space through a linear transformation to generate initial node features.

[0160]

[0161] in, The weight matrix is ​​a learnable matrix. For learnable bias vectors, This represents the projected dimension. These features are then processed through a multi-layer graph capsule neural network, where each layer extracts node embeddings layer by layer using a polynomial-form graph convolutional filter.

[0162] (5)

[0163] in, For the graph Laplace matrix, Let be the order of the statistical moments. Let be the order of the convolution filter. For the first The output of the layer, The weight matrix is ​​learnable. Node embeddings not only contain information about the node itself, but also integrate information aggregated from multiple neighboring nodes.

[0164] As the number of graph convolutional layers increases and the order of convolutional filters is improved, the model is able to capture more complex graph structure information.

[0165] The output of each layer is concatenated with the convolution results of all orders to form the final node embedding:

[0166] (6)

[0167] After linear transformation, the final node embedding matrix is ​​generated. Subsequently, the node embeddings are processed through a series of linear transformations to generate a global graph embedding:

[0168] (7)

[0169] This embedding is obtained by averaging the embedding values ​​of all nodes, thus providing a global state representation for the entire power distribution network.

[0170] Besides graph structure information, some state variables in energy storage networks cannot be directly represented by nodes in the graph, such as energy supply, voltage violations, and power flow at edges. This information is crucial for the network's operational safety and operability; therefore, it is necessary to generate context embeddings as additional state information using feedforward neural networks.

[0171] (8)

[0172] This contextual information, together with the graph embedding, constitutes the state representation of the entire system. This is then further processed by a multilayer perceptron to generate the final state embedding.

[0173] (9)

[0174] Example 4

[0175] Based on the dynamic network reconfiguration method for digital power storage systems described in Example 3, the dynamic network reconfiguration method also includes a model training process, as detailed below:

[0176] (1) Data preparation and scenario generation:

[0177] Faults in energy storage systems often originate from localized failures, which can lead to cascading effects.

[0178] Specifically, the faults in the network are mainly caused by line faults, which are simulated using a graph capsule neural network-based method.

[0179] To simulate this behavior, a subgraph method with random edge removal was employed.

[0180] This method involves randomly selecting nodes from a graph capsule neural network of an energy storage system. and create nodes with different radii around these nodes. The subgraph with the maximum radius.

[0181] set up ,in It is the diameter of the diagram.

[0182] Within each selected subgraph, randomly remove a portion of the edges. To simulate the effects of local accidents.

[0183] The proportion of edge failures gradually increased from 0 to 50%.

[0184] By changing , and Scenarios with multiple line faults are generated for model training.

[0185] In addition, in each scenario, the load multiplier and generator point are varied by randomly selecting the multiplier and generator point from the annual configuration file in the OpenDSS package with hourly resolution.

[0186] (2) Training process:

[0187] The training process generates samples simulating various fault scenarios on the energy storage network by introducing line faults, adjusting loads and energy storage device capacities, and considering different fault conditions.

[0188] The network faults are mainly caused by power distribution line faults, which are simulated using a graph capsule neural network-based method. Load demand and energy storage capacity, two key operational parameters for the energy storage network, are randomly selected from annual load curves provided by OpenDSS.

[0189] Considering that the state and action spaces expand rapidly with increasing network size, fault scenarios, and the number of devices, methods such as Q-learning cannot provide sufficient storage and computational resources in practice. Therefore, the Proximal Policy Optimization (PPO) algorithm is used to train the policy network, implemented using the stable-baselines3 Python library.

[0190] The training process involves collecting experience in the form of tuples containing state, action, reward, and next state. PPO is run on a rollout basis, with each operation consisting of a fixed number of N steps. Weight updates are performed in batches of N (≤N steps) after the rollout, minimizing the cost function containing policy gradient loss and state value loss through backpropagation. The policy network is trained for a total of Ntotal steps.

[0191] To evaluate the performance of the proposed model and assess the impact of local and global structural information during encoding, comparative experiments were conducted with another learning-based framework, MLP. This framework also uses the PPO algorithm, with the policy network based on the MLP architecture. To ensure a fair comparison, MLP and DRO were trained under the same settings.

[0192] It is worth noting that for any given or existing network, this process only needs to be performed offline once or a few times. The trained network model does not need to be retrained, and current computing resources are fully capable of supporting this operation. Therefore, for training a real-time fault management decision support system, this is a reasonable offline investment, because once the system is trained, it can be used for real-time decision support for a long time.

[0193] (3) Training details:

[0194] Taking the IEEE 34-bus distribution test network and IEEE 123-bus distribution test network from the typical IEEE models as examples:

[0195] In the IEEE 34-bus distribution test network, the MLP can be trained up to 1.5 million steps. In the IEEE 123-bus distribution test network, the MLP can be trained up to 500,000 steps.

[0196] To ensure fairness, the same settings were applied to DRO.

[0197] Specifically, DRO was trained for 1.5 million steps in the IEEE 34-bus distribution test network and 500,000 steps in the IEEE 123-bus distribution test network.

[0198] In addition, the Adam optimization algorithm is used to enhance the training process.

[0199] like Figure 4 As shown, the real-time reward convergence value of the trained DRO model is consistently better than the current benchmark scheme, and the convergence speed of the DRO model is faster. This observation demonstrates the superior performance of DRO in effectively managing network faults and optimizing the operating status of energy storage networks.

[0200] This invention also utilizes the open-source power distribution system simulator OpenDSS to implement and simulate the energy storage network. The energy storage network is modeled using general-purpose generators, solar photovoltaic elements, and energy storage components from OpenDSS.

[0201] Switches are defined on lines with associated switch controls, while load disable or enable attributes are used to unload or pick up loads. Energy storage devices adjust their charging, discharging, or idle states based on current capacity and system status. The implementation of switching operations may result in multiple segments within the network, allowing for isolation of faulty sections while maintaining normal operation in other areas. Energy storage devices in a discharging state can be considered power sources, providing energy to nodes isolated from power generation equipment due to faults.

[0202] An improved version of the IEEE 34-bus distribution test network was used as the case study energy storage network to evaluate the proposed framework. Nine switches were considered for the line switches in the IEEE 34-bus distribution test network. All buses, except the substation bus, were considered either load buses or energy storage device buses. The total grid-connected load was 2.04 MW. A 146 kW grid-formed distributed generation source was connected to bus 890, a 96 kW grid-feeder distributed generation source was connected to bus 820, and five energy storage devices with capacities of 20%, 80%, 90%, 50%, and 10% were connected at bus 860, 844, 816, 836, and 822, respectively. Loads were connected to eight buses: 808, 818, 816, 824, 828, 854, and 832.

[0203] The experiment included three different fault scenarios, each consisting of one of three different line faults. For example... Figure 5 As shown, the states of decision variables obtained from different models in three scenarios. Figure 5 'a' represents the switch decision result diagram of the IEEE 34 bus distribution test network. Dark color indicates the switch is closed, and light color indicates the switch is open.

[0204] Figure 5 b represents the decision result diagram of energy storage devices in the IEEE 34 bus distribution test network. Yellow indicates that the energy storage device is idle, green indicates that the energy storage device is discharging, and red indicates that the energy storage device is charging. Figure 5 c represents the load status decision result of the IEEE 34 bus distribution test network. Dark color indicates that the load is working, and light color indicates that the load is idle.

[0205] On the IEEE 34-bus distribution test network, based on the network reconstruction and transmission optimization results of fault scenario 1, such as... Figure 6As shown in the diagram, fault scenario 1 involved four connection faults: 814-828, 828-830, 834-858, and 888-890. The switches at 814-828 and 828-830 failed to close. Connection 834-858, being a highly centralized connection, also failed. Following the fault, the grid-formed distributed generation at bus 890 became disconnected from the overall network. The energy storage devices at 816 and 844, with their higher capacities, acted as power sources for discharging. By performing switching operations, the network structure was restructured, the energy flow direction was changed, and the energy storage network was able to continue operating.

[0206] In this embodiment, an experiment was also conducted on the IEEE 123 bus distribution test network. Similar to the experiment on the IEEE 34 bus distribution test network, an improved version of the IEEE 123 bus distribution test network was used as the energy storage network for the research case, and 22 switches were considered to be configured.

[0207] Connect one 259kW grid-formed distributed power source to each of the 39, 46, 71, 75, and 79 buses. Connect one 80kW grid-feeder distributed power source to each of the 11, 82, 91, and 104 buses. Connect seven energy storage devices with storage capacities of 90%, 80%, 70%, 50%, 10%, 80%, and 90% respectively at the 33, 56, 82, 84, 92, 96, and 108 bus locations. Connect the load to 16 bus locations: 11, 16, 20, 22, 30, 48, 49, 53, 64, 65, 66, 69, 88, 107, 109, and 113.

[0208] Three different fault scenarios were set up in the experiment, each consisting of three different line faults. For example... Figure 7 As shown, the states of decision variables obtained in different models for three scenarios include Figure 7 a shows the status of the circuit breaker. Figure 7 b shows the status of the energy storage device. Figure 7 c indicates the state of the load.

[0209] On the IEEE 123 bus distribution test network, the schematic diagram of network reconstruction and transmission optimization results based on fault scenario 1 is shown below. Figure 8 As shown in the diagram, fault scenario 1 involved three connection faults: 13-18, 51-151, and 65-66. After the fault occurred, the energy storage devices at buses 33, 56, 96, and 100, with their higher capacities, acted as power sources to discharge. By performing a switching operation, the network structure was restructured, the energy flow direction was changed, and the energy storage network was able to continue operating.

[0210] In the experiments, this invention modified the IEEE 34-bus distribution test network and the IEEE 123-bus distribution test network to construct corresponding energy storage networks for these networks. Using a traditional model—a reinforcement learning model employing a multilayer perceptron (MLP) as the policy network—and a baseline model, the performance of the proposed DRO model and the MLP model in terms of real-time decision-making and network resilience improvement was compared.

[0211] The energy supplied during an interruption can serve as a measure of resilience, such as... Figure 9 As shown in the figure. The results show that the DRO model proposed in this invention provides high energy under different network and fault scenarios, ensuring the economic efficiency of the energy storage system.

[0212] Furthermore, compared to the MLP baseline model, the DRO model better captures the topology of the distribution network during the decision-making process, exhibits fewer voltage violation constraints, demonstrates superior performance, and better adheres to constraints, thus ensuring the safety of the energy storage system. Figure 10 As shown.

[0213] Three different fault scenarios were proposed for the two test networks, and all three could correctly decide to maintain the continuous and stable operation of the energy storage network without retraining the model. This demonstrates the excellent generalization ability of DRO; in networks of the same size, it only needs to be trained once to handle all fault scenarios, saving training time and computational resources, and providing a guarantee for subsequent practical applications.

[0214] The real-time energy storage network fault management model DRO proposed in this invention successfully integrates the topology information of the energy storage network into the reinforcement learning decision framework, effectively improving the model's adaptability and fault management efficiency in dynamic environments.

[0215] By comparing experimental results, the DRO model not only outperforms traditional methods in terms of energy supply, but also demonstrates good decision-making accuracy and network reconstruction capabilities in complex scenarios with different faults and network changes. Furthermore, by integrating graph capsule neural networks, the model fully mines the local and global structural information of the energy storage network, providing strong support for dealing with complex energy storage device management and charge / discharge scheduling.

[0216] The technical solutions of the present invention are not limited to the specific embodiments described above. Without departing from the scope and spirit of the described embodiments, many modifications and changes will be obvious to those skilled in the art. Any technical modifications made within the spirit and principles of the present invention will fall within the protection scope of the present invention.

Claims

1. A method for dynamic network reconfiguration of a digitized power energy storage system, characterized in that, The steps are as follows: S1. Energy storage system construction; the energy storage system is composed of an intelligent layer, a control layer and a data layer; the intelligent layer obtains the current state of the energy storage system, extracts the node feature information of the energy storage network based on the graph neural network, and transmits the action information decided by the reinforcement learning agent to the control layer; the control layer receives the action information and generates control instructions according to the rules and transmits them to the data layer; the data layer includes an energy storage network, which adjusts according to the control instructions and updates the energy storage network to the control center; S2. Energy storage network modeling; the graph capsule neural network is used to learn the optimal control strategy of the energy storage network, and the devices and connection relationships in the energy storage network are used as nodes and edges to construct the graph capsule neural network. On the basis of the graph capsule neural network, the corresponding features of each node and edge are extracted, and the features of each node and edge are encoded. Each node and edge is represented as a capsule composed of a group of feature vectors. Through the dynamic routing algorithm, information is transmitted between capsules, and features are aggregated and combined layer by layer to capture high-order structural information of the energy storage network. Then, combined with the message passing mechanism of the graph capsule neural network, the feature vectors of the capsules are iteratively updated to gradually extract and optimize the global information of the energy storage system; S3. Dynamic network reconstruction; based on the combination control method of energy discretization and digitization, the topological structure reconstruction of the energy storage network and the reconfiguration of the energy storage devices and load devices are covered. Through systematic optimization of the topological structure and optimal control strategy of the energy storage network, rapid isolation of the fault area and dynamic adjustment and reconfiguration of the working mode of the energy storage devices and load devices are realized. Based on the Markov decision process, a fault emergency response model is generated to ensure that the energy storage network can still operate stably under fault conditions. The Markov decision process of the fault emergency response in the energy storage network is denoted as where S, A, , R represent state, action, transition probability and reward, respectively. The action A includes switch operation, energy storage device charging state operation and load working state operation; the action space is denoted as wherein, represents the number of switchable lines, including sectionalizing switches and tie switches; the number of operable energy storage devices in the energy storage network is represented as ; line switches are represented by binary variables , with 0 indicating that the switch is open and 1 indicating that the switch is closed; binary variables represent the energy storage charging status, with 1 indicating that the energy storage device is charging, -1 indicating that the energy storage device is discharging, and 0 indicating that the energy storage device is idle; binary variables represent the load operating status, with 1 indicating that the load is in an operating state and 0 indicating that the load is in an idle state.

2. The method of claim 1, wherein: Step S3 includes the construction of the learning architecture; by combining the graph capsule neural network with the policy gradient method, a policy network for learning the control of the energy storage network is constructed and the corresponding action is output to ensure the efficiency and safety of the system operation.

3. The method of claim 2, wherein: Step S3 also includes the model training process; by introducing line faults, adjusting the capacity of load devices and energy storage devices, and considering different fault situations, samples simulating various fault situations are generated.

4. The method of claim 1, wherein: The state S is composed of relevant observation values in the energy storage network, representing the current operating condition of the energy storage network; Including node variables, edge variables and network topology, denoted as wherein, represents an estimated or predicted value of the active power demand of a node; represents an estimated or predicted value of the reactive power demand; represents the generation of active and reactive power of a node; represents the three-phase voltage measured at a bus node; represents a voltage violation constraint value in the energy storage network; represents the power flow through an energy storage network branch; the operating topology of the network is represented as ; the total energy supply in the energy storage network is represented as ; the state variable represents a fault scenario; represents the handling of the inoperability of a faulty switch by a shielding mechanism; If the switch is operating properly, ; if the switch is inoperable due to a fault, , the switch is masked.

5. The method of claim 1, wherein: Transition probabilities To capture the dynamic characteristics of the energy storage network under fault emergency response, denoted as Indicates at time step Perform actions Afterwards, the energy storage network status changes from time step status Shift to time step status The transition probability is learned through the interaction between the agent and the environment.

6. The method of claim 1, wherein: The reward R is obtained by taking the optimal control action through the reinforcement learning algorithm to alleviate the impact of faults in the energy storage network, and its formula is By maximizing supply energy and minimizing voltage constraint violations , the objective of improving the resilience of the energy storage network, in addition to introducing operational constraints of the energy storage network to actively avoid the problem of imbalance and constraint violations caused by faults or switch operations.

7. The method of claim 2, wherein: During the construction of the learning architecture, the attribute of each node of the energy storage network is projected to a high-dimensional space through a linear transformation to generate an initial node feature; the initial node feature is processed through a multi-layer graph capsule neural network, a graph convolution filter is used to extract node embedding layer by layer, the output of each layer is spliced with the convolution results of all stages to form the final node embedding, and a linear transformation is performed to generate a final node embedding matrix, and a series of linear transformation processing is performed to generate a global graph embedding, thereby providing a global state representation for the energy storage network.

8. The method of claim 3, wherein: Data preparation and scenario generation are also required in the early stage of model training. Based on the graph capsule neural network, a subgraph method with random edge removal is used to simulate the fault scenario in the energy storage network. By randomly selecting a node from the graph capsule neural network of the energy storage system and creating a subgraph with different radii around the selected node, a part of the edges in each selected subgraph is randomly removed to simulate the impact of local faults, and by changing the selected node, the subgraph radius and the removed edges, scenarios with multiple line faults are generated for model training.

Images