Power distribution network transient control method and device based on knowledge-data fusion driving

Abstract

The application discloses a power distribution network transient control method and device based on knowledge-data fusion driving, and relates to the technical field of power grid control. The method comprises the following steps: acquiring state data of power distribution network transient control; performing forward propagation processing on the state data based on a preset intelligent agent to obtain an action variable of the power distribution network, wherein the intelligent agent is obtained by embedding a control barrier function representing a safety boundary of the power distribution network into a reinforcement learning framework constructed based on a Markov decision process, and is obtained by dynamically adjusting a safety constraint by using a state-wise Lagrange multiplier network; the Markov decision process is obtained after modeling the power distribution network transient control problem; and a control reference value of primary frequency modulation in the action variable is sent to a corresponding target inverter, and the target inverter is driven to adjust based on the control reference value to complete the transient control of the power distribution network, so as to ensure stable operation of the power distribution network under a complex disturbance scenario.

Description

Technical Field

[0001] This application relates to the field of power grid control technology, and in particular to a method and apparatus for transient control of distribution networks based on knowledge-data fusion. Background Technology

[0002] As distributed photovoltaic power sources are connected to the distribution network via power electronic inverters, it also poses a severe challenge to the transient safety and stability control of the distribution network.

[0003] Because distribution networks primarily rely on the rotational inertia of synchronous generators to maintain transient stability, while inverter-dominated distribution networks lack natural physical inertia support, the control systems of these networks require improved response speed and coordination capabilities. Furthermore, due to the low inertia and weak damping characteristics of inverters, the voltage and frequency of the inverters are prone to significant oscillations when subjected to large disturbances such as short-circuit faults, sudden load changes, or drastic fluctuations in photovoltaic output. Currently, deep reinforcement learning methods are commonly used to derive control strategies from historical data or real-time interactive learning, and these strategies are then used to control the transient state of the distribution network. However, due to the lack of guidance from knowledge of the distribution network's transient operation and the reliance solely on reward functions to adjust the control strategy, the safety of the control process cannot be strictly guaranteed. Therefore, it is impossible to ensure the safe and stable operation of a distribution network connected to inverters under complex disturbances.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main purpose of this application is to provide a knowledge-data fusion-driven transient control method and device for distribution networks, which aims to solve the technical problem that distribution networks connected to inverters cannot guarantee safe and stable operation under complex disturbances.

[0006] To achieve the above objectives, this application proposes a knowledge-data fusion-driven transient control method for power distribution networks, the method comprising: Acquire status data for transient control of the power distribution network; Based on a pre-defined intelligent agent, the state data is forward-propagated to obtain the action variables of the distribution network. The intelligent agent is obtained by embedding the control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on Markov decision process, and then dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained by modeling the transient control problem of the distribution network. The control reference value for primary frequency regulation in the action variables is sent to the corresponding target inverter, and the target inverter is driven to adjust based on the control reference value to complete the transient control of the distribution network.

[0007] In one embodiment, before the step of acquiring the state data of the distribution network transient control, the method further includes: Introduce security constraints into a reinforcement learning framework based on Markov decision processes; The policy network in the reinforcement learning framework interacts with the simulation environment of the power distribution network to collect interaction samples; Based on the interaction samples and the security constraints, the policy network, value network, Lagrange multiplier network and control barrier network in the reinforcement learning framework are trained alternately. Iteratively execute the steps of collecting interactive samples and alternately training each network until the policy network converges, then stop the iteration; The converged policy network is used as the agent for transient control of the distribution network.

[0008] In one embodiment, the safety constraints include a safety penalty term determined based on a preset safety voltage and safety frequency, and a safety guidance term determined by the control barrier network and the Lagrange multiplier network based on the interaction sample output, respectively.

[0009] In one embodiment, the step of alternately training the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework based on the interaction samples and the security constraints includes: Using the safety penalty term as a constraint, and based on the interaction samples, train the first loss function of the control obstacle network to obtain the updated control obstacle network. Based on the interaction samples, the second loss function of the value network is trained to obtain the trained value network. Using the output of the control barrier network as a constraint, and based on the interaction samples, the third loss function of the Lagrange multiplier network is trained to obtain the updated Lagrange multiplier network. The first loss function is embedded into the fourth loss function of the policy network, and the output of the Lagrange multiplier network before the update is used as the guiding weight of the safety guidance term. Based on the interaction samples, the policy network is trained to obtain the updated policy network.

[0010] In one embodiment, the step of alternately training the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework based on the interaction samples and the security constraints further includes: Based on training samples randomly drawn from the interaction samples and the security constraints, the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework are trained alternately.

[0011] In one embodiment, prior to the step of introducing security constraints into the reinforcement learning framework based on Markov decision processes, the method further includes: The voltage amplitude, frequency, and phase angle of each inverter in the power distribution network are determined as state variables. The frequency reference value and voltage reference value of each inverter are determined as the operating variables; The negative value of the weighted sum of the squares of the voltage deviation and frequency deviation of each inverter is determined as the reward function; Based on voltage safety upper and lower limit constraints and frequency safety upper and lower limit constraints, a set of safe states is defined, which includes all system states that satisfy the condition that the voltage amplitude is between the voltage safety upper and lower limits and the frequency is between the frequency safety upper and lower limits. A Markov decision process is created based on the determined state variables, action variables, reward function, and the set of safe states.

[0012] In one embodiment, the step of creating a Markov decision process based on the determined state variables, action variables, reward function, and the set of safe states includes: The state space formed by the state variables is divided into a first subset and a second subset that do not overlap. The first subset includes voltage amplitude states that are between the upper and lower voltage safety limits and whose frequencies are between the upper and lower frequency safety limits, and is defined as the safe state set; the second subset includes voltage amplitude states that are not between the upper and lower voltage safety limits or the upper and lower frequency safety limits. The action space consisting of the action variables, the reward function, the first subset, and the second subset are collectively defined as components of the Markov decision process.

[0013] Furthermore, to achieve the above objectives, this application also proposes a knowledge-data fusion-driven distribution network transient control device, which includes: The acquisition module is used to acquire status data of transient control in the distribution network; The processing module is used to perform forward propagation processing on the state data based on a preset intelligent agent to obtain the action variables of the distribution network. The intelligent agent is obtained by embedding the control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on Markov decision process, and then dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained by modeling the transient control problem of the distribution network. The adjustment module is used to send the control reference value of the primary frequency regulation in the action variables to the corresponding target inverter, and drive the target inverter to adjust based on the control reference value to complete the transient control of the distribution network.

[0014] Furthermore, to achieve the above objectives, this application also proposes a knowledge-data fusion-driven distribution network transient control device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the knowledge-data fusion-driven distribution network transient control method described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the knowledge-data fusion-driven power distribution network transient control method described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: In the transient control process of the distribution network, the agent embeds the control barrier function, which represents the safety boundary of the distribution network, as physical prior knowledge into a reinforcement learning framework based on Markov decision processes. This ensures that the forward propagation processing of the acquired state data by the agent is constrained by the safety boundary, avoiding the risk of voltage or frequency exceeding the upper or lower limits due to deviations in policy exploration or reward design. Furthermore, by introducing a state-wise Lagrange multiplier network to dynamically adjust the safety constraints, the agent can adaptively balance control performance and safety margin according to the current system state. In addition, the Markov decision process is directly based on the transient control problem of the distribution network, and the state-action space of the reinforcement learning framework fits the actual physical dynamics. This ensures that the control parameter values ​​used for primary frequency regulation in the action variables obtained by the agent satisfy the safety constraints. Therefore, the voltage and frequency response of the inverter will not cause large oscillations due to improper control commands. This achieves a fast, coordinated, and safe transient response of the inverter-dominated distribution network under large disturbances such as short-circuit faults, load changes, or severe fluctuations in photovoltaic output, ensuring the stable operation of the low-inertia distribution network in complex disturbance scenarios. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating an embodiment of the knowledge-data fusion-driven transient control method for power distribution networks in this application. Figure 2 This is a flowchart illustrating Embodiment 2 of the knowledge-data fusion-driven transient control method for power distribution networks in this application; Figure 3 This is a flowchart illustrating Embodiment 3 of the knowledge-data fusion-driven transient control method for power distribution networks in this application. Figure 4 This is a schematic diagram of the module structure of the knowledge-data fusion-driven power distribution network transient control device according to an embodiment of this application; Figure 5 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the knowledge-data fusion-driven power distribution network transient control method in the embodiments of this application.

[0020] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or a power distribution network control system capable of performing the above functions. The following description uses a power distribution network control system as an example to illustrate this embodiment and the subsequent embodiments.

[0024] Based on this, embodiments of this application provide a knowledge-data fusion-driven transient control method for power distribution networks, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the knowledge-data fusion-driven transient control method for power distribution networks according to this application.

[0025] In this embodiment, the knowledge-data fusion-driven distribution network transient control method includes steps S10 to S30: Step S10: Obtain the status data of the power distribution network transient control; It should be noted that transient control is a set of technical measures that, when a distribution network encounters sudden large disturbances (such as short circuits, load changes, or fluctuations in distributed power output), rapidly intervene through automatic or manual means to prevent the distribution network from becoming unstable and collapsing. State data consists of observable measurements such as voltage amplitude, frequency, and phase angle at each inverter interface in the distribution network, reflecting the current operating state of the distribution network, and is used to form the input state vector of the reinforcement learning agent.

[0026] Understandably, by acquiring real-time status data such as voltage amplitude, frequency, and phase angle of each inverter in the distribution network, the instantaneous operating condition of the distribution network after a disturbance can be comprehensively and accurately reflected. This provides a high-fidelity input basis for intelligent agent decision-making, ensuring that the control strategy closely matches the real dynamics of the distribution network and avoiding control failures or malfunctions caused by incomplete or delayed state perception. Thus, precise and rapid transient stabilization of the distribution network can be achieved from the state perception level.

[0027] In practice, status data can be collected in real time through synchronous phasor measurement units deployed at key nodes of the distribution network (such as grid connection points of distributed photovoltaic inverters and energy storage converters) or smart meters that meet accuracy requirements.

[0028] Step S20: Based on the preset intelligent agent, the state data is forward-propagated to obtain the action variables of the distribution network. The intelligent agent is obtained by embedding the control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on Markov decision process, and then dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained by modeling the transient control problem of the distribution network. It should be noted that forward propagation processing involves inputting state data into each network of the pre-trained agent, sequentially calculating and outputting action variables, without involving parameter updates. Action variables are the frequency and voltage reference values ​​of the target inverter that the distribution network needs to control, used to drive the target inverter to adjust its output power. Control barrier functions are mathematical functions characterizing the voltage and frequency safety boundaries in the distribution network. Physical prior knowledge is formalizable constraint information extracted from the physical laws or operating specifications of the power system; for example, the set of safe states defined by the upper and lower voltage safety limits and the upper and lower frequency safety limits, and their corresponding control barrier functions.

[0029] A Markov decision process is a mathematical framework for modeling sequential decision problems, consisting of a state space, an action space, state transition probabilities, a reward function, and a set of safe states. The state space is composed of the inverter's voltage, frequency, and phase angle; the action space is the set of reference values; and the reward function is the negative of the weighted sum of squared voltage / frequency deviations.

[0030] Understandably, by using an agent that embeds the control barrier function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework, and can also use a state-wise Lagrange multiplier network to dynamically adjust the strength or weight of safety constraints, the action variables obtained through forward propagation of state data can not only pursue the optimization of control performance, but also fundamentally ensure that all control commands meet the preset safety constraints. This effectively avoids the safety risks that may arise from purely data-driven methods, and also avoids strong dependence on precise system models, thereby improving the safety of the control strategy.

[0031] Understandably, by explicitly embedding the control barrier function, which represents the safety boundary of the distribution network, as physical prior knowledge into the reinforcement learning framework, the agent's policy exploration can be strictly constrained within the feasible domain defined by the safety limits of electrical quantities such as voltage and frequency during its interaction and learning with the environment. This fundamentally solves the problem that purely data-driven deep reinforcement learning methods may produce dangerous actions that violate safety constraints due to a lack of understanding of physical laws.

[0032] In the specific implementation, a transient control problem model for the distribution network is first constructed using the transient model of the photovoltaic inverter, the objective function, and the safety constraints. Then, the transient control problem of the distribution network is modeled as a reinforcement learning framework of a Markov decision process. The Markov decision process includes state variables, action variables, and a reward function. The control obstacle function representing the safety boundary of the distribution network is used as physical prior knowledge and embedded into the reinforcement learning framework to obtain an initial agent including a policy network, a value network, a control obstacle network, and a Lagrange multiplier network. Then, the state-wise Lagrange relaxation method is used to perform safety reinforcement learning on the initial agent to obtain an optimized agent for the transient control of the distribution network. Finally, the state data of this agent is used for forward propagation processing to output the action variables required for the transient control of the distribution network.

[0033] In practical implementation, by introducing an independent state-wise Lagrange multiplier network into the agent, the agent can dynamically adjust the weight of safety constraints in the reinforcement learning optimization objective based on the current operating state of the distribution network (such as real-time values ​​of voltage and frequency and their rates of change). When the distribution network state is far from the safety boundary, the Lagrange multiplier network outputs a smaller multiplier value, causing the agent to focus on improving control performance during policy optimization. Conversely, when the distribution network state approaches or is in an emergency, the Lagrange multiplier network automatically outputs a larger multiplier value, forcing the policy network to place safety constraints with the highest priority, ensuring that all action commands can effectively guide the distribution network state back to the safe zone.

[0034] In practical implementation, the agent can be deployed on edge computing devices (control systems of power distribution networks) with GPU acceleration capabilities. Upon arrival of state data, it is first input into a lightweight convolutional neural network for feature extraction to capture the spatial coupling relationships between nodes. Subsequently, this feature vector is fed into the policy network. Simultaneously, the control barrier function network calculates the safety metric based on the current state, while the state-wise Lagrange multiplier network outputs the multiplier corresponding to the current state. The loss function of the policy network is designed as a weighted sum of the safety guidance term and the network security penalty term of the control barrier function, with the weights dynamically determined by the safety guidance and penalty terms. After one complete forward propagation, the policy network finally outputs the action variables for each target inverter, namely, the frequency reference increment and voltage reference value.

[0035] Step S30: Send the control reference value of the primary frequency regulation in the action variables to the corresponding target inverter, and drive the target inverter to adjust based on the control reference value to complete the transient control of the distribution network.

[0036] It should be noted that primary frequency regulation is a primary frequency response mechanism in which the inverter automatically adjusts the active power output based on preset droop characteristics or reference values ​​when the frequency of the distribution network deviates from the rated value, so as to quickly suppress frequency changes.

[0037] Understandably, by directly converting the action variables generated by the intelligent agent and embedded with safety guarantees into primary frequency regulation control reference values ​​that the inverter can execute, a seamless connection is achieved from intelligent decision-making to execution by the underlying physical equipment. Moreover, this conversion process makes full use of the fast response characteristics of modern power electronic equipment, which can complete transient adjustment within a time scale of hundreds of milliseconds, thus effectively suppressing frequency and voltage fluctuations that exceed limits.

[0038] Understandably, since the control reference value is strictly constrained within the safety boundary by the intelligent agent during the generation stage, no additional safety verification or intervention is required at the execution level. Therefore, the control logic of the distribution network control system for transient control of the distribution network is greatly simplified, the response speed and reliability of the distribution network are improved, and the safe and stable operation of the distribution network under the high proportion of renewable energy access is ultimately guaranteed.

[0039] In practical implementation, the calculated control reference values ​​(frequency reference increment and voltage reference value) for primary frequency regulation can be sent to the local controllers of each target inverter via a standard power communication protocol. After receiving the new reference values, the local controllers of the inverters can adjust the active power output commands of the corresponding inverters in real time according to the preset active power-frequency droop characteristic curve, thereby injecting or absorbing active power into the distribution network and quickly supporting the frequency of the entire distribution network.

[0040] This embodiment provides a knowledge-data fusion-driven transient control method for distribution networks. During the transient control process, the agent embeds a control barrier function representing the distribution network's safety boundary as physical prior knowledge into a reinforcement learning framework based on Markov decision processes. This ensures that the agent's forward propagation processing of acquired state data is constrained by the safety boundary, avoiding the risk of voltage or frequency exceeding upper or lower limits due to deviations in policy exploration or reward design. Furthermore, by introducing a state-wise Lagrange multiplier network to dynamically adjust the safety constraints, the agent can adjust the control based on the current system state. The system adaptively balances control performance with safety margin. Furthermore, the Markov decision process is directly constructed based on the transient control problem of the distribution network. The state-action space of the reinforcement learning framework fits the actual physical dynamics, ensuring that the control parameter values ​​used for primary frequency regulation in the action variables obtained by the agent meet safety constraints. Therefore, the voltage and frequency response of the inverter will not cause large oscillations due to improper control commands. This enables a fast, coordinated, and safe transient response to large disturbances such as short-circuit faults, load changes, or severe fluctuations in photovoltaic output in the inverter-dominated distribution network, thus ensuring the stable operation of the low-inertia distribution network in complex disturbance scenarios.

[0041] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 Before step S10, the knowledge-data fusion-driven distribution network transient control method further includes steps S01 to S05: Step S01: Introduce security constraints into a reinforcement learning framework based on Markov decision processes; Step S02: The policy network in the reinforcement learning framework interacts with the simulation environment of the power distribution network to collect interaction samples; Step S03: Based on the interaction samples and the security constraints, alternately train the policy network, value network, Lagrange multiplier network and control obstacle network in the reinforcement learning framework; Step S04: Iteratively execute the steps of collecting interactive samples and alternately training each network until the policy network converges, then stop the iteration; Step S05: The converged policy network is used as the intelligent agent for the transient control of the distribution network.

[0042] It should be noted that safety constraints are the physical and engineering limitations that the distribution network must meet during operation, specifically the allowed range of values ​​for the distribution network's state variables. The policy network represents the agent's policy; its input is the current distribution network state data, and its output is the action command (action variable) to be applied to the inverter. The value network evaluates the expected cumulative reward (state value) obtained by following the current policy in a given state and provides gradient information for updating the policy network based on this expected cumulative reward. The Lagrange multiplier network dynamically generates Lagrange multipliers related to the current distribution network state and uses these multipliers as weights in the optimization objective, achieving adaptive adjustment of the trade-off between safety and performance. The control barrier network can be understood as the carrier of the control barrier function; its input is the distribution network state data, and its output is the safety margin representing the distance of the current state from the safety boundary. Interaction samples are data tuples generated by the interaction between the agent and the distribution network simulation environment, containing information such as state, action, reward, and next state.

[0043] Understandably, using control barrier networks as carriers of physical prior knowledge can fundamentally constrain the exploration space of intelligent agents, thereby ensuring the theoretical security of the entire training process and the final deployment strategy.

[0044] Understandably, the introduction of Lagrange multiplier networks means that the trade-off between safety and performance is no longer fixed, but can be dynamically and adaptively adjusted according to the real-time status of the distribution network, thereby maximizing control performance while ensuring safety.

[0045] Understandably, introducing safety constraints into a reinforcement learning framework based on Markov decision processes can deeply integrate prior knowledge of the physical safety of the distribution network with the data-driven learning model, thus avoiding the limit-crossing problem that is prone to occur in pure data-driven control at the framework level.

[0046] It is understandable that training the policy network, value network, Lagrange multiplier network, and control barrier network alternately based on interactive samples and safety constraints can achieve synergistic optimization of safety constraints, value assessment, and policy decision-making, avoid the one-sidedness of training a single network, and enable each network to adapt to and correct each other, thereby improving the overall decision-making accuracy and safety control capability of the reinforcement learning framework.

[0047] In practical implementation, a transient control agent for the distribution network is constructed. This agent optimizes the Markov decision process, including a policy network, a value network, a control obstacle network, and a Lagrange multiplier network. Specifically: Building a policy network for intelligent agents and initialize the parameters of the policy network. The input to this policy network is the state variables of a Markov decision process. The output is the action variables of the Markov decision process. .

[0048] Building a value network for intelligent agents and initialize the parameters of the value network. The input to this value network is the state variables of a Markov decision process. and action variables The output is the cumulative discount reward of the fitted Markov decision process.

[0049] Constructing a target value network for intelligent agents Its target value network structure and value network Same, initialize its parameters Make it with same.

[0050] Constructing a Lagrange multiplier network for intelligent agents And initialize the parameters of the Lagrange multiplier network. The input to this Lagrange multiplier network is the state variables of a Markov decision process. The output is the Lagrange multiplier corresponding to this state; simultaneously, the activation function of the agent Lagrange multiplier network output layer is... This is to ensure that the output Lagrange multipliers are positive.

[0051] Constructing a control barrier network for intelligent agents and initialize the parameters of the control obstacle network. The input to this control barrier network is the state variables of a Markov decision process. The output is the control obstacle function value corresponding to that state. When the state is safe, the control obstacle network is required. Output a negative value; when the state is unsafe, the control obstacle network is required. Output a positive value; also require the state at the next time step. The control barrier function value is less than the current value. If the above safety requirements are met, the safety of the transient control process can be guaranteed.

[0052] Furthermore, the safety constraints include safety penalty terms determined based on preset safety voltage and safety frequency, and safety guidance terms determined by the control barrier network and the Lagrange multiplier network based on the interaction sample output, respectively.

[0053] It should be noted that the safe voltage is the permissible range of node voltage amplitude during steady-state and transient operation of the distribution network. It is determined by preset upper and lower limits of the rated voltage and is a rigid electrical constraint ensuring the insulation safety of distribution equipment and the stability of system voltage. The safe frequency is the permissible range of system frequency during steady-state and transient operation of the distribution network. It is determined by preset upper and lower limits of the rated frequency and is a core electrical constraint for maintaining the active power balance of the grid and preventing frequency collapse. The safety penalty term is a penalty calculation term built based on the safe voltage and safe frequency thresholds. When the operating parameters of the distribution network exceed the safety boundary, a positive penalty value is output to penalize over-limit control behavior during network training and force the strategy to return to the safe range. The control barrier network is a neural network built around the control barrier function, used to transform the physical prior knowledge of the distribution network safety boundary into quantified constraint signals. The safety guidance term is a guidance term jointly calculated based on interactive samples using the safety constraint signal output by the control barrier network and the dynamic weights output by the Lagrange multiplier network.

[0054] Understandably, breaking down safety constraints into safety penalty items and safety guidance items allows for a dual-constraint control strategy that addresses both exceeding limits and providing positive guidance. This makes the constraint logic more complete and the control direction more precise. Furthermore, determining safety penalty items based on safe voltage and safe frequency allows for the direct embedding of prior physical knowledge of the distribution network, forming a rigid safety baseline and preventing control strategies from causing voltage and frequency exceedances at the source.

[0055] Understandably, determining the safety guidance term based on the control barrier network, Lagrange multiplier network, and interactive samples can achieve state-by-state dynamic adjustment of safety constraints, adapting to the strong randomness and volatility of the distribution network transient process, and coping with the complex disturbance scenarios in which the distribution network is located.

[0056] Furthermore, step S03 also includes: Using the safety penalty term as a constraint, and based on the interaction samples, train the first loss function of the control obstacle network to obtain the updated control obstacle network. Based on the interaction samples, the second loss function of the value network is trained to obtain the trained value network. Using the output of the control barrier network as a constraint, and based on the interaction samples, the third loss function of the Lagrange multiplier network is trained to obtain the updated Lagrange multiplier network. The first loss function is embedded into the fourth loss function of the policy network, and the output of the Lagrange multiplier network before the update is used as the guiding weight of the safety guidance term. Based on the interaction samples, the policy network is trained to obtain the updated policy network.

[0057] It should be noted that the output of the control barrier network is the distribution network safety boundary constraint signal calculated by the control barrier network based on interactive samples.

[0058] Understandably, using the safety penalty term as a constraint to train the control barrier network's first loss function deeply embeds the prior knowledge of the distribution network's physical safety into the network learning process, enhancing the control barrier network's accurate characterization of the safety boundary. The second loss function, trained based on interactive samples, accurately optimizes the state-action value assessment capability, providing reliable value guidance for the policy network and improving the long-term optimality of the control strategy. Using the output of the control barrier network as a constraint to train the Lagrange multiplier network's third loss function enables state-by-state adaptive learning of the safety constraint weights, allowing the constraint strength to adapt to different transient operating conditions and balancing control safety and operational efficiency. Finally, embedding the first loss function into the policy network's fourth loss function and combining it with safety-guided weights to train the policy neural network directly integrates safety constraints into the policy decision-making optimization process, ensuring that the policy output simultaneously satisfies reward optimality and safety constraints, fundamentally preventing out-of-bounds control behavior.

[0059] In its implementation, a control obstacle network and a Lagrange multiplier network are integrated, and a state-wise Lagrange relaxation method is used to perform safety reinforcement learning on the agent, resulting in an optimized distribution network transient control agent. Specifically: The agent interacts with the power distribution network to obtain interaction samples at time t. Stored in the agent's experience pool P, where, Let be the state variables of the Markov decision process at time t. Let be the action variable at time t. Let be the reward function of the Markov decision process at time t. The state variables are predicted for time t+1.

[0060] A set of interaction samples D is randomly selected from the agent's experience pool P as training samples, and the policy network is computed. loss function .

[0061] ; in, The coefficient of the policy entropy. By embedding the control barrier function into the policy network loss function through state-wise Lagrange relaxation, the safety of the policy is improved and the agent is guided to converge quickly to a safe and stable policy.

[0062] Based on the calculated loss function of the policy network Update strategy network parameters : ;in, For parameters The learning rate.

[0063] A set of interaction samples D is randomly selected from the agent's experience pool P as training samples, and the agent's value network is calculated. loss function : ; in, for The target value is calculated as follows: ; in, Discount rate for future rewards .

[0064] Based on the calculated loss function of the intelligent agent value network Update agent value network parameters : ;in, For parameters The learning rate.

[0065] Based on the updated agent value network parameters Update agent target value network parameters : ;in, For parameters The update rate.

[0066] A set of interaction samples D is randomly selected from the agent's experience pool P as training samples, and the agent's Lagrange multiplier network is calculated. loss function : ; Based on the calculated loss function of the agent Lagrange multiplier network Update agent Lagrange multiplier network parameters : ;in, For parameters The learning rate.

[0067] A set of interaction samples D is randomly selected from the agent's experience pool P as training samples to compute the agent's control obstacle network. loss function : ; in, It is a linear rectified function, i.e. ; This defines the allowable error range for the control barrier function. The three penalty terms in the control barrier network loss function correspond to the three safety requirements of the control barrier function, resulting in a control barrier function that meets these requirements.

[0068] Based on the calculated loss function of the intelligent agent's control obstacle network Update the network parameters for controlling obstacles in the intelligent agent. : ;in, For parameters The learning rate.

[0069] Repeat the above steps until the agent's policy converges or the maximum number of iterations is reached, then store the trained policy network. This is to enable transient control of the power distribution network.

[0070] Optionally, step S03 may also be: Based on training samples randomly drawn from the interaction samples and the security constraints, the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework are trained alternately.

[0071] Understandably, each network is trained by randomly selecting training samples from the interaction samples, and by combining safety constraints with alternating training of the policy network, value network, Lagrange multiplier network, and control barrier network. This breaks the fixed sequence limitation of training samples, eliminates the directional induction of neural network weight updates by continuous similar samples, and allows safety constraints to be iteratively optimized under different random operating conditions. This makes the characterization of safety boundaries by the control barrier network and the adjustment of constraint weights by the Lagrange multiplier network more comprehensive and accurate, thus ensuring the safety of transient control strategies from the training level.

[0072] Based on the first and second embodiments of this application, the same or similar content as the above embodiments in the third embodiment of this application can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 3 Before step S01, the knowledge-data fusion-driven distribution network transient control method further includes steps S1 to S5: Step S1: Determine the voltage amplitude, frequency, and phase angle of each inverter in the power distribution network as state variables; Step S2: Determine the frequency reference value and voltage reference value of each inverter as the action variable; Step S3: The negative value of the weighted sum of squares of the voltage deviation and frequency deviation of each inverter is determined as the reward function; Step S4: Based on the upper and lower limits of voltage safety and the upper and lower limits of frequency safety, define a set of safe states. The set of safe states includes all system states that satisfy the condition that the voltage amplitude is between the upper and lower limits of voltage safety and the frequency is between the upper and lower limits of frequency safety. Step S5: Create a Markov decision process based on the determined state variables, action variables, reward function, and the set of safe states.

[0073] It should be noted that voltage deviation is the difference between the actual voltage amplitude of the inverter and the rated voltage amplitude. Frequency deviation is the difference between the actual operating frequency of the inverter and the rated frequency.

[0074] Understandably, defining inverter voltage amplitude, frequency, and phase angle as state variables allows for accurate characterization of the core electrical features of the distribution network's transient operation, providing a realistic and complete state input for the decision-making model. Furthermore, defining inverter frequency and voltage reference values ​​as action variables enables control commands to directly match the inverter's execution logic, reducing command conversion losses and improving the response speed and execution accuracy of transient control.

[0075] Understandably, using the negative of the weighted sum of squares of voltage and frequency deviations as the reward function can quantify control error and guide strategy optimization, enabling the control strategy to iterate in the direction of reducing voltage and frequency deviations, thereby improving the regulation accuracy of transient control.

[0076] In practical implementation, a transient control problem of the distribution network is modeled. This transient control problem includes a transient model of the photovoltaic inverter, an objective function, and security constraints. Specifically: The primary control process of a photovoltaic inverter is as follows: ; ; in, and The first One photovoltaic inverter A primary control signal that sets the time frequency and voltage amplitude; and The first One photovoltaic inverter The primary control reference values ​​for the instantaneous frequency and voltage amplitude; and The first The droop factor of the frequency and voltage amplitude of a photovoltaic inverter; and The first One photovoltaic inverter The active and reactive power output at all times.

[0077] No. A photovoltaic inverter measures its Output voltage at all times and output current After a time constant of The filter is used to measure the output active and reactive power. The method is as follows: ; ; in, and The first One photovoltaic inverter The derivative of active power output with respect to time and the derivative of reactive power with respect to time at any given moment; and The first One photovoltaic inverter Output voltage at all times of Axial components and Axial components; and The first One photovoltaic inverter Output current at all times of Axial components and Axial components.

[0078] No. One photovoltaic inverter Output current at all times Axial components and Axial components The derivative with respect to time is: ; ; in, and The first One photovoltaic inverter Output current at all times The derivatives of the axial components with respect to time and The derivative of the axial component with respect to time; and The first The resistance and reactance of the output interface of a photovoltaic inverter; For the first One photovoltaic inverter Frequency of time; and The first One photovoltaic inverter Voltage of the connected grid node at any time of Axial components and Axial components.

[0079] No. The time required for frequency and voltage regulation of a photovoltaic inverter is much less than that required for the same photovoltaic inverter. This can be considered to be completed instantaneously. Then the first... One photovoltaic inverter Time Frequency Output voltage Axial components and Axial components It can be represented as: ; ; At the same time, the One photovoltaic inverter Phase angle at time The derivative with respect to time is as follows: .

[0080] The objective function of the transient control problem of the distribution network is: ;

[0081] in, It is the transient control cycle of the power distribution network; It refers to the number of photovoltaic inverters in the power distribution network; and These are the weights for the voltage control objective and the frequency control objective, respectively. and These are the target values ​​for voltage and frequency, respectively.

[0082] The security constraints for the transient control problem of the distribution network are:

[0083] , , ;

[0084] , , ; in, and These are the lower and upper limits of the output voltage of the photovoltaic inverter, respectively. and These represent the lower and upper limits of the frequency for photovoltaic inverters, respectively.

[0085] In practical implementation, the transient control problem of the distribution network is modeled as a Markov decision process, specifically: State variables of Markov decision process in transient control problem of power distribution network as follows: ; The set of safe states for this Markov decision process is defined as follows: ; Action variables of Markov decision process in transient control problem of power distribution network for: ; Reward function of Markov decision process for transient control problem of distribution network for: .

[0086] Furthermore, step S5 may also include: The state space formed by the state variables is divided into a first subset and a second subset that do not overlap. The first subset includes voltage amplitude states that are between the upper and lower voltage safety limits and whose frequencies are between the upper and lower frequency safety limits, and is defined as the safe state set; the second subset includes voltage amplitude states that are not between the upper and lower voltage safety limits or the upper and lower frequency safety limits. The action space consisting of the action variables, the reward function, the first subset, and the second subset are collectively defined as components of the Markov decision process.

[0087] Understandably, defining states that simultaneously meet the upper and lower safety limits of voltage and frequency as the first subset (the set of safe states) allows for the precise embedding of the physical safety priors of the distribution network into the modeling process, clearly defining the safety target range for transient control. Furthermore, defining states that do not meet safety constraints as the second subset accurately identifies system risk states exceeding limits, facilitating the punishment of unsafe control behaviors during reinforcement learning training. This forces the strategy to revert to the safe range, thereby ensuring the safety of the control strategy from the modeling level, preventing voltage and frequency limit violations during transient control, and improving the operational stability of the distribution network.

[0088] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the knowledge-data fusion-driven distribution network transient control method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0089] This application also provides a knowledge-data fusion-driven power distribution network transient control device, please refer to... Figure 4 The knowledge-data fusion-driven power distribution network transient control device includes: Module 10 is used to acquire status data of transient control of the distribution network; The processing module 20 is used to perform forward propagation processing on the state data based on a preset intelligent agent to obtain the action variables of the distribution network. The intelligent agent is obtained by embedding the control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on Markov decision process, and then dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained by modeling the transient control problem of the distribution network. The adjustment module 30 is used to send the control reference value of the primary frequency regulation in the action variables to the corresponding target inverter, and drive the target inverter to adjust based on the control reference value to complete the transient control of the distribution network.

[0090] The knowledge-data fusion-driven distribution network transient control device provided in this application, employing the knowledge-data fusion-driven distribution network transient control method described in the above embodiments, can solve the technical problem of ensuring the safe and stable operation of distribution networks connected to inverters under complex disturbances. Compared with the prior art, the beneficial effects of the knowledge-data fusion-driven distribution network transient control device provided in this application are the same as those of the knowledge-data fusion-driven distribution network transient control method provided in the above embodiments, and other technical features in the knowledge-data fusion-driven distribution network transient control device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0091] This application provides a knowledge-data fusion-driven distribution network transient control device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the knowledge-data fusion-driven distribution network transient control method in the first embodiment described above.

[0092] The following is for reference. Figure 5This document illustrates a structural schematic diagram of a knowledge-data fusion-driven power distribution network transient control device suitable for implementing embodiments of this application. The knowledge-data fusion-driven power distribution network transient control device in this application embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 5 The knowledge-data fusion-driven distribution network transient control device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0093] like Figure 5 As shown, the knowledge-data fusion-driven power distribution network transient control device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1002 or the program loaded from the storage device 1003 into the random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the knowledge-data fusion-driven power distribution network transient control device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the knowledge-data fusion-driven power distribution network transient control equipment to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a knowledge-data fusion-driven power distribution network transient control equipment with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0094] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0095] The knowledge-data fusion-driven distribution network transient control device provided in this application, employing the knowledge-data fusion-driven distribution network transient control method described in the above embodiments, can solve the technical problem of ensuring the safe and stable operation of distribution networks connected to inverters under complex disturbances. Compared with the prior art, the beneficial effects of the knowledge-data fusion-driven distribution network transient control device provided in this application are the same as those of the knowledge-data fusion-driven distribution network transient control method provided in the above embodiments, and other technical features in this knowledge-data fusion-driven distribution network transient control device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0096] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0097] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0098] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the knowledge-data fusion-driven power distribution network transient control method in the above embodiments.

[0099] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0100] The aforementioned computer-readable storage medium may be included in a knowledge-data fusion-driven distribution network transient control device; or it may exist independently and not be assembled into a knowledge-data fusion-driven distribution network transient control device.

[0101] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a knowledge-data fusion-driven distribution network transient control device, the device performs the following actions: acquires state data for distribution network transient control; performs forward propagation processing on the state data based on a pre-defined intelligent agent to obtain action variables for the distribution network. The intelligent agent is obtained by embedding a control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on a Markov decision process, and dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained after modeling the distribution network transient control problem. The device then sends the control reference value used for primary frequency regulation in the action variables to the corresponding target inverter and drives the target inverter to adjust based on the control reference value to complete the transient control of the distribution network.

[0102] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0103] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0104] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0105] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the aforementioned knowledge-data fusion-driven distribution network transient control method. This solves the technical problem of ensuring the safe and stable operation of distribution networks connected to inverters under complex disturbances. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the knowledge-data fusion-driven distribution network transient control method provided in the above embodiments, and will not be repeated here.

[0106] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. A knowledge-data fusion-driven transient control method for power distribution networks, characterized in that, The method includes: Acquire status data for transient control of the power distribution network; Based on a pre-defined intelligent agent, the state data is forward-propagated to obtain the action variables of the distribution network. The intelligent agent is obtained by embedding the control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on Markov decision process, and then dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained by modeling the transient control problem of the distribution network. The control reference value for primary frequency regulation in the action variables is sent to the corresponding target inverter, and the target inverter is driven to adjust based on the control reference value to complete the transient control of the distribution network.

2. The knowledge-data fusion-driven transient control method for distribution networks as described in claim 1, characterized in that, Before the step of acquiring the status data of the distribution network transient control, the method further includes: Introduce security constraints into a reinforcement learning framework based on Markov decision processes; The policy network in the reinforcement learning framework interacts with the simulation environment of the power distribution network to collect interaction samples; Based on the interaction samples and the security constraints, the policy network, value network, Lagrange multiplier network and control barrier network in the reinforcement learning framework are trained alternately. Iteratively execute the steps of collecting interactive samples and alternately training each network until the policy network converges, then stop the iteration; The converged policy network is used as the agent for transient control of the distribution network.

3. The knowledge-data fusion-driven transient control method for distribution networks as described in claim 2, characterized in that, The safety constraints include safety penalty terms determined based on preset safety voltage and safety frequency, and safety guidance terms determined by the control barrier network and the Lagrange multiplier network based on the interaction sample output, respectively.

4. The knowledge-data fusion-driven transient control method for distribution networks as described in claim 3, characterized in that, The step of alternately training the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework based on the interaction samples and the security constraints includes: Using the safety penalty term as a constraint, and based on the interaction samples, train the first loss function of the control obstacle network to obtain the updated control obstacle network. Based on the interaction samples, the second loss function of the value network is trained to obtain the trained value network. Using the output of the control barrier network as a constraint, and based on the interaction samples, the third loss function of the Lagrange multiplier network is trained to obtain the updated Lagrange multiplier network. The first loss function is embedded into the fourth loss function of the policy network, and the output of the Lagrange multiplier network before the update is used as the guiding weight of the safety guidance term. Based on the interaction samples, the policy network is trained to obtain the updated policy network.

5. The knowledge-data fusion-driven transient control method for distribution networks as described in claim 2, characterized in that, The step of alternately training the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework based on the interaction samples and the security constraints further includes: Based on training samples randomly drawn from the interaction samples and the security constraints, the policy network, value network, Lagrange multiplier network, and control obstacle network in the reinforcement learning framework are trained alternately.

6. The knowledge-data fusion-driven transient control method for distribution networks as described in claim 2, characterized in that, Before the step of introducing security constraints into the reinforcement learning framework based on Markov decision processes, the method further includes: The voltage amplitude, frequency, and phase angle of each inverter in the power distribution network are determined as state variables. The frequency reference value and voltage reference value of each inverter are determined as the operating variables; The negative value of the weighted sum of the squares of the voltage deviation and frequency deviation of each inverter is determined as the reward function; Based on voltage safety upper and lower limit constraints and frequency safety upper and lower limit constraints, a set of safe states is defined, which includes all system states that satisfy the condition that the voltage amplitude is between the voltage safety upper and lower limits and the frequency is between the frequency safety upper and lower limits. A Markov decision process is created based on the determined state variables, action variables, reward function, and the set of safe states.

7. The knowledge-data fusion-driven transient control method for distribution networks as described in claim 6, characterized in that, The step of creating a Markov decision process based on the determined state variables, action variables, reward function, and the set of safe states includes: The state space formed by the state variables is divided into a first subset and a second subset that do not overlap. The first subset includes voltage amplitude states that are between the upper and lower voltage safety limits and whose frequencies are between the upper and lower frequency safety limits, and is defined as the safe state set; the second subset includes voltage amplitude states that are not between the upper and lower voltage safety limits or the upper and lower frequency safety limits. The action space consisting of the action variables, the reward function, the first subset, and the second subset are collectively defined as components of the Markov decision process.

8. A knowledge-data fusion-driven transient control device for power distribution networks, characterized in that, The device includes: The acquisition module is used to acquire status data of transient control in the distribution network; The processing module is used to perform forward propagation processing on the state data based on a preset intelligent agent to obtain the action variables of the distribution network. The intelligent agent is obtained by embedding the control obstacle function representing the safety boundary of the distribution network as physical prior knowledge into a reinforcement learning framework based on Markov decision process, and then dynamically adjusting the safety constraints using a state-wise Lagrange multiplier network. The Markov decision process is obtained by modeling the transient control problem of the distribution network. The adjustment module is used to send the control reference value of the primary frequency regulation in the action variables to the corresponding target inverter, and drive the target inverter to adjust based on the control reference value to complete the transient control of the distribution network.

9. A knowledge-data fusion-driven transient control device for power distribution networks, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the knowledge-data fusion-driven power distribution network transient control method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the knowledge-data fusion-driven power distribution network transient control method as described in any one of claims 1 to 7.

Images