System and method for intelligently arranging power-cut plan of power grid based on improved SAC

Abstract

The invention discloses an improved SAC-based intelligent arrangement system for a power-cut plan of a power grid. The system comprises an objective function construction module which establishes a multi-objective function of the power-cut plan of the power grid according to power demand data; the constraint condition construction module establishes a multi-dimensional constraint condition according to the safety operation data of the maintenance equipment; the Markov decision process modeling module constructs an intelligent learning framework by using a multi-objective function and a multi-dimensional constraint condition, wherein the intelligent learning framework comprises a state space, an action space and a reward function; the action cutting module is used for dynamically cutting the action space based on constraint conditions; the SAC network construction module learns an optimal strategy through intelligent agent and environment interaction to obtain a trained SAC network; and the SAC network application module deploys the strategy parameters in a power grid dispatching system to generate an intelligent arrangement scheme of the power-cut plan. According to the invention, the problems of low efficiency, difficult coordination, easy conflict and difficult global optimization caused by dependence on artificial experience in traditional power-cut plan arrangement are solved, and the intelligent level of arrangement and the decision-making efficiency are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power system operation and management technology, specifically to an intelligent scheduling system and method for power grid outage plans based on an improved SAC. Background Technology

[0002] Outage planning is a crucial aspect of power system operation and management, directly impacting grid safety, power supply reliability, and the orderly conduct of maintenance. With the continuous expansion of the power grid, the large-scale integration of renewable energy sources, and the deepening of power market reforms, outage planning faces increasingly complex constraints and challenges in balancing conflicting optimization objectives.

[0003] Existing methods for scheduling power outages primarily rely on the experience of dispatching personnel, manually coordinating factors such as equipment maintenance needs, grid safety constraints, and load power supply reliability. While this method depends to some extent on the professional knowledge and practical experience of operators, it suffers from low efficiency, difficulty in coordination, susceptibility to omissions and conflicts, and difficulty in achieving overall optimization. Especially during periods of intensive maintenance and fluctuating grid conditions, manual scheduling often struggles to accommodate the collaborative requirements of multiple departments, resulting in long scheduling cycles, frequent adjustments, and high execution risks.

[0004] In recent years, scholars both domestically and internationally have conducted extensive research on power outage planning optimization problems, introducing optimization methods such as mathematical programming and heuristic algorithms into power outage planning. These methods improve planning efficiency by establishing mathematical models or search strategies. However, when faced with large-scale, high-dimensional, nonlinear, and strongly coupled power outage planning problems, mathematical programming methods often suffer from a sharp decline in solution efficiency due to dimensionality; while traditional heuristic algorithms, although possessing strong global search capabilities, have slow convergence speeds, are prone to getting trapped in local optima, and struggle to effectively adapt to dynamically changing power grid operating environments and cope with uncertainties.

[0005] With the development of artificial intelligence technology, reinforcement learning, due to its ability to learn optimal decision-making strategies through trial and error in complex and dynamic environments, has been gradually applied to the field of power system optimization and scheduling. Reinforcement learning does not require pre-established precise mathematical models and can autonomously learn strategies through interaction with the environment, possessing strong adaptability and generalization capabilities. However, existing research mainly focuses on scenarios such as real-time scheduling, voltage control, or unit combination, and has not yet effectively addressed key challenges in outage planning, such as multi-objective, multi-constraint, and long-term scheduling decisions. Furthermore, how to construct a reasonable state space, action space, and reward function to accurately represent the complex relationship between grid operating status, maintenance task requirements, and safety constraints remains a core challenge restricting the application of deep reinforcement learning in this field. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent scheduling system and method for power grid outage plans based on improved SAC. This invention overcomes the shortcomings of existing technologies by introducing deep reinforcement learning algorithms into the field of outage plan optimization. By constructing an optimization model and reinforcement learning framework for outage plans, it achieves intelligent scheduling of outage plans, thereby improving the decision-making efficiency and reliability of power grid operation and management.

[0007] To achieve this objective, the present invention provides an intelligent scheduling system for power grid outage planning based on an improved SAC (System-Area Controlled Controller), comprising: The objective function construction module is used to build a multi-objective function for power grid outage plans based on electricity demand data. The constraint construction module is used to establish multi-dimensional constraints for power grid outage plans based on the safe operation data of maintenance equipment. The Markov decision process modeling module is used to construct an intelligent learning framework for power grid outage plans based on Markov decisions using the multi-objective function and multi-dimensional constraints. The intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function. The action trimming module, based on multi-dimensional constraints, eliminates unselectable plans set in the power grid action space when scheduling power outage plans on the decision day in the power grid state space, thus obtaining the trimmed power grid action space. The SAC network construction module is used to learn the optimal policy through the interaction between the agent and the environment based on the power grid state space, the pruned power grid action space and the power grid reward function, and obtain the trained SAC network. The SAC network application module is used to deploy the strategy network parameters of the trained SAC network into the power grid dispatch intelligent decision support system. In the power grid dispatch intelligent decision support system, a multi-dimensional state vector is constructed based on the periodically collected power grid operation data to obtain an intelligent scheduling scheme for power outage plans.

[0008] Preferably, in the objective function construction module, the multi-objective function for establishing a power grid outage plan based on electricity demand data is specifically used for: For monthly maintenance planning cycles, set penalty costs for failure to schedule maintenance. The specific formula is as follows: in, This refers to the number of equipment items scheduled for maintenance each month that are not included in the current period's maintenance plan. For the first The penalty coefficient for un-programmed equipment is set based on the equipment's importance, safety risk level, and maintenance time required. The specific formula for the total load loss during the monthly maintenance plan period is as follows: in, For the first The actual load loss caused by maintenance. The number of days in the monthly maintenance plan scheduling cycle; For minimizing renewable energy sources, the total amount of renewable energy sources that will be rendered useless during the monthly maintenance plan cycle is: in, For the first The abandoned electricity from Tianguang Photovoltaic Power Station; For the first The abandoned electricity of Tianfeng Power Plant; The linear weighted method is used to address the three optimization objectives: minimizing the penalty cost of unscheduled maintenance, the total load loss, and the total amount of renewable energy curtailment. The overall objective function is: ; in, , , They are respectively , , The target reward weighting coefficient.

[0009] Preferably, in the constraint construction module, the multi-dimensional constraint conditions for establishing the power grid outage plan based on the safe operation data of the maintenance equipment are specifically used for: Set maintenance time constraints for power outage plans: ; in, For the first The earliest executable time for the planned power outage; For the first The latest completion time for the planned power outage; For the first The actual start time of the planned power outage; For the first The estimated time required for maintenance of the planned power outage; Set maintenance continuity constraints for power outage plans: in, For the first The power outage plan was in the The execution status of the day, =1 indicates the first Heavenly The equipment corresponding to this power outage plan is under maintenance. =0 indicates the first Heavenly The equipment corresponding to this power outage plan is not under maintenance. If the first For power outage plans involving critical infrastructure, the timing must be determined by the applicant and cannot be adjusted. The planned power outage must meet the non-changeability constraint, and its start time for maintenance must be specified. It must be equal to the application time. .

[0010] When the The equipment maintenance corresponding to the planned power outage will affect the first When the equipment corresponding to the planned power outage is operating normally, maintenance of the affected equipment must be carried out simultaneously to reduce power outage losses, i.e., the first... The power outage plan and the first The planned power outage must meet the simultaneous outage constraint: in, The start and end times of the plan with the shorter power outage duration in a set of same-outage constraint plans; Let the first The power outage plan and the first If two power outage plans satisfy a mutual exclusion relationship, then the outage periods of the two plans cannot overlap, that is, the first... The power outage plan and the first The power outage plan must satisfy mutual exclusion constraints, and the specific formula is as follows: Setting up a power outage schedule must meet maintenance resource constraints, and the specific formula is as follows: in, This is the maximum number of devices that can be repaired simultaneously. This represents the total number of maintenance tasks planned for the current period. Set line transmission capacity constraints: in, For the first Skyline The actual transmitted active power; For the line The maximum allowed transmission power; Set power balance constraints: ; in, for Total system load during the time period for System load loss during the time period , , for Time period Taiwan thermal power unit, the first The first photovoltaic power station, the first The grid-connected power of each wind farm , , These represent the number of thermal power units, the number of photovoltaic power stations, and the number of wind farms, respectively.

[0011] The specific formula for setting upper and lower limit constraints is as follows: in, The first Minimum and maximum technical output of Taiwan's thermal power units; For the first Taiwan thermal power units Start / stop status during a time period =1 indicates the first The thermal power unit is in operation. =0 indicates the first The thermal power units in Taiwan are currently out of service.

[0012] Preferably, in the Markov decision process modeling module, an intelligent learning framework is constructed for the power grid outage plan based on Markov decisions. This intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function, specifically used for: The power grid outage plan is modeled as a Markov decision process, and the power grid state space, power grid action space, power grid state transition and power grid reward function are defined according to the characteristics of the outage plan problem. State Space Construction: A state vector integrating multiple dimensions was constructed. Specifically defined as , among which decision day Indicates the current optimization decision time and power outage plan attributes. This includes planned constraints and planned power outage duration, and the number of remaining days. This refers to the remaining programmable decision windows within the current cycle, and the equipment maintenance status. Reflects the real-time status of power outage equipment in the system and power flow information. As runtime parameters, the above-mentioned attribute parameters are integrated to construct... The state vector of a phased agent ; Constructing the action space: The action space is defined as the decision day. The set of executable power outage plans can be represented as , express The actions taken by the agent at each stage ,when hour, This means that no plans will be made for the time being, and the remaining plans will be considered on the next day; when At that time, it means that the plan will be implemented. Arranged to the number Maintenance will begin today; Constructing state transitions: when Actions taken by the phase agent At that time, power outage plan attributes and equipment maintenance status Remain unchanged, decision day Remaining days Calculate the updated power flow state information of the power grid. ,get Phase State ;when At that time, decision day Power outage plan attributes and remaining days Remain unchanged, according to the new arrangement plan Update equipment maintenance status Calculate the updated power flow state information of the power grid. ,get Phase State .

[0013] Constructing a reward function: A multi-dimensional reward function evaluation system is constructed to determine the penalty cost indicator for not scheduling maintenance, the total load loss indicator within the monthly maintenance plan cycle, and the total amount of renewable energy curtailment within the monthly maintenance plan cycle. in, The number of unplanned renewable energy abandonments on the decision date is [number to be filled in]. , The amount of abandoned electricity from the n-day photovoltaic power plant is determined on the decision-making date. The amount of abandoned electricity from wind farm n on the decision-making day. This is the final state, meaning that all power outage plans have been completed or the remaining power outage plans cannot be scheduled due to constraints. for In the stage orchestration state, R represents the reward for completing the orchestration of all plans in the final state. , , They are respectively , , The target reward weighting coefficient.

[0014] Preferably, in the SAC network construction module, the optimal policy is learned through the interaction between the agent and the environment to obtain the trained SAC network, specifically including: Initialize and set the network parameters and initialization status of the number of Critic networks and Actor networks, and set the total number of training epochs Q and the experience pool size. ; Determine the decision date If the application time is an unchangeable plan, then the unchangeable plan will be directly scheduled to the decision date if the constraints are met. Determine whether the decision date is the latest possible scheduling date for the unscheduled maintenance time constraint plan. If so, schedule the maintenance time constraint plan to the decision date, provided that the constraint conditions are met, and update the current status. Based on the current state, the action space is pruned, and normalized probabilities are sampled for action selection: selecting a non-zero action indicates that the plan will be scheduled for the decision date. For simultaneous stop plans... , , If you choose the plan Then, under the premise of satisfying the constraints, the plan is arranged simultaneously. and plans That is, the first The actual start time of the planned power outage is equal to the first The actual start time of the planned power outage. If you choose the plan Then update the first For the plan The latest time that can be scheduled; Based on the orchestration scheme corresponding to the selected action, the minimum load shedding and renewable energy curtailment under power flow constraints are obtained using the power system analysis toolkit, and the reward value is calculated. Update the next status and empirical samples The data is stored in the experience pool. A batch of experience samples is sampled from the experience pool to train the SAC network and update the network parameters. When the number of samples in the experience pool reaches the preset capacity limit, the data in the experience pool is deleted. When all planned arrangements are completed or no effective actions can be performed subsequently, the round termination condition is met, the power outage plan arrangement scheme is output, the current round ends, and the next round begins.

[0015] This invention also provides a method for intelligent scheduling of power grid outage plans based on improved SAC, comprising: Establish a multi-objective function for power grid outage planning based on electricity demand data; Establish multi-dimensional constraints for power grid outage plans based on the safe operation data of maintenance equipment; Using the aforementioned multi-objective function and multi-dimensional constraints, an intelligent learning framework for power grid outage planning is constructed based on Markov decision-making. The intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function. Based on multi-dimensional constraints, when scheduling power outage plans on the decision day in the power grid state space, the unselectable plans set in the power grid action space are eliminated to obtain the trimmed power grid action space. Based on the power grid state space, the pruned power grid action space, and the power grid reward function, the optimal policy is learned through the interaction between the agent and the environment, resulting in the trained SAC network. The policy network parameters of the trained SAC network are deployed in the intelligent decision support system for power grid dispatch. In the intelligent decision support system for power grid dispatch, a multi-dimensional state vector is constructed based on the periodically collected power grid operation data to obtain an intelligent scheduling scheme for power outage plans.

[0016] Preferably, the multi-objective function for establishing the power grid outage plan based on electricity demand data specifically includes: For monthly maintenance planning cycles, set penalty costs for failure to schedule maintenance. The specific formula is as follows: in, This refers to the number of equipment items scheduled for maintenance each month that are not included in the current period's maintenance plan. For the first The penalty coefficient for un-programmed equipment is set based on the equipment's importance, safety risk level, and maintenance time required. The specific formula for the total load loss during the monthly maintenance plan period is as follows: in, For the first The actual load loss caused by maintenance. The number of days in the monthly maintenance plan scheduling cycle; For minimizing renewable energy sources, the total amount of renewable energy sources that will be rendered useless during the monthly maintenance plan cycle is: in, For the first The abandoned electricity from Tianguang Photovoltaic Power Station; For the first The abandoned electricity of Tianfeng Power Plant; The linear weighted method is used to address the three optimization objectives: minimizing the penalty cost of unscheduled maintenance, the total load loss, and the total amount of renewable energy curtailment. The overall objective function is: ; in, , , They are respectively , , The target reward weighting coefficient.

[0017] Preferably, the multi-dimensional constraints for establishing the power grid outage plan based on the safe operation data of the maintenance equipment specifically include: Set maintenance time constraints for power outage plans: ; in, For the first The earliest executable time for the planned power outage; For the first The latest completion time for the planned power outage; For the first The actual start time of the planned power outage; For the first The estimated time required for maintenance of the planned power outage; Set maintenance continuity constraints for power outage plans: in, For the first The power outage plan was in the The execution status of the day, =1 indicates the first Heavenly The equipment corresponding to this power outage plan is under maintenance. =0 indicates the first Heavenly The equipment corresponding to this power outage plan is not under maintenance. If the first For power outage plans involving critical infrastructure, the timing must be determined by the applicant and cannot be adjusted. The planned power outage must meet the non-changeability constraint, and its start time for maintenance must be specified. It must be equal to the application time. .

[0018] When the The equipment maintenance corresponding to the planned power outage will affect the first When the equipment corresponding to the planned power outage is operating normally, maintenance of the affected equipment must be carried out simultaneously to reduce power outage losses, i.e., the first... The power outage plan and the first The planned power outage must meet the simultaneous outage constraint: in, The start and end times of the plan with the shorter power outage duration in a set of same-outage constraint plans; Let the first The power outage plan and the first If two power outage plans satisfy a mutual exclusion relationship, then the outage periods of the two plans cannot overlap, that is, the first... The power outage plan and the first The power outage plan must satisfy mutual exclusion constraints, and the specific formula is as follows: Setting up a power outage schedule must meet maintenance resource constraints, and the specific formula is as follows: in, This is the maximum number of devices that can be repaired simultaneously. This represents the total number of maintenance tasks planned for the current period. Set line transmission capacity constraints: in, For the first Skyline The actual transmitted active power; For the line The maximum allowed transmission power; Set power balance constraints: ; in, for Total system load during the time period for System load loss during the time period , , for Time period Taiwan thermal power unit, the first The first photovoltaic power station, the first The grid-connected power of each wind farm , , These represent the number of thermal power units, the number of photovoltaic power stations, and the number of wind farms, respectively.

[0019] The specific formula for setting upper and lower limit constraints is as follows: in, The first Minimum and maximum technical output of Taiwan's thermal power units; For the first Taiwan thermal power units Start / stop status during a time period =1 indicates the first The thermal power unit is in operation. =0 indicates the first The thermal power units in Taiwan are currently out of service.

[0020] Preferably, an intelligent learning framework for power grid outage planning is constructed based on Markov decision-making. This intelligent learning framework includes a power grid state space, a power grid action space, and a power grid reward function, specifically comprising: The power grid outage plan is modeled as a Markov decision process, and the power grid state space, power grid action space, power grid state transition and power grid reward function are defined according to the characteristics of the outage plan problem. State Space Construction: A state vector integrating multiple dimensions was constructed. Specifically defined as , among which decision day Indicates the current optimization decision time and power outage plan attributes. This includes planned constraints and planned power outage duration, and the number of remaining days. This refers to the remaining programmable decision windows within the current cycle, and the equipment maintenance status. Reflects the real-time status of power outage equipment in the system and power flow information. As runtime parameters, the above-mentioned attribute parameters are integrated to construct... The state vector of a phased agent ; Constructing the action space: The action space is defined as the decision day. The set of executable power outage plans can be represented as , express The actions taken by the agent at each stage ,when hour, This means that no plans will be made for the time being, and the remaining plans will be considered on the next day; when At that time, it means that the plan will be implemented. Arranged to the number Maintenance will begin today; Constructing state transitions: when Actions taken by the phase agent At that time, power outage plan attributes and equipment maintenance status Remain unchanged, decision day Remaining days Calculate the updated power flow state information of the power grid. ,get Phase State ;when At that time, decision day Power outage plan attributes and remaining days Remain unchanged, according to the new arrangement plan Update equipment maintenance status Calculate the updated power flow state information of the power grid. ,get Phase State .

[0021] Constructing a reward function: A multi-dimensional reward function evaluation system is constructed to determine the penalty cost indicator for not scheduling maintenance, the total load loss indicator within the monthly maintenance plan cycle, and the total amount of renewable energy curtailment within the monthly maintenance plan cycle. in, The number of unplanned renewable energy abandonments on the decision date is [number to be filled in]. , The amount of abandoned electricity from the n-day photovoltaic power plant is determined on the decision-making date. The amount of abandoned electricity from wind farm n on the decision-making day. This is the final state, meaning that all power outage plans have been completed or the remaining power outage plans cannot be scheduled due to constraints. for In the stage orchestration state, R represents the reward for completing the orchestration of all plans in the final state. , , They are respectively , , The target reward weighting coefficient.

[0022] Preferably, the optimal strategy is learned through interaction between the agent and the environment to obtain the trained SAC network, specifically including: The steps to obtain the trained SAC network include: Initialize and set the network parameters and initialization status of the number of Critic networks and Actor networks, and set the total number of training epochs Q and the experience pool size. ; Determine the decision date If the application time is an unchangeable plan, then the unchangeable plan will be directly scheduled to the decision date if the constraints are met. Determine whether the decision date is the latest possible scheduling date for the unscheduled maintenance time constraint plan. If so, schedule the maintenance time constraint plan to the decision date, provided that the constraint conditions are met, and update the current status. Based on the current state, the action space is pruned, and normalized probabilities are sampled for action selection: selecting a non-zero action indicates that the plan will be scheduled for the decision date. For simultaneous stop plans... , , If you choose the plan Then, under the premise of satisfying the constraints, the plan is arranged simultaneously. and plans That is, the first The actual start time of the planned power outage is equal to the first The actual start time of the planned power outage. If you choose the plan Then update the first For the plan The latest time that can be scheduled; Based on the orchestration scheme corresponding to the selected action, the minimum load shedding and renewable energy curtailment under power flow constraints are obtained using the power system analysis toolkit, and the reward value is calculated. Update the next status and empirical samples The data is stored in the experience pool. A batch of experience samples is sampled from the experience pool to train the SAC network and update the network parameters. When the number of samples in the experience pool reaches the preset capacity limit, the data in the experience pool is deleted. When all planned arrangements are completed or no effective actions can be performed subsequently, the round termination condition is met, the power outage plan arrangement scheme is output, the current round ends, and the next round begins.

[0023] The present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of a method for identifying key meteorological factors affecting icing of transmission lines as described above.

[0024] The beneficial effects of this invention are as follows: This invention enables intelligent and automated scheduling of power grid outage plans, significantly improving the efficiency and accuracy of plan formulation. By introducing the deep reinforcement learning SAC algorithm, it effectively overcomes the "curse of dimensionality" problem faced by traditional manual experience-based methods and mathematical programming methods. This invention can balance multiple objectives such as power grid safe operation, power supply reliability, and renewable energy consumption, dynamically adapting to complex conditions such as maintenance resources and power flow constraints, reducing load losses and renewable energy curtailment. This invention enhances the efficiency and convergence performance of strategy exploration through action space pruning and prior knowledge embedding, reducing the planning cycle and execution risk. This invention can adapt to changes in the power grid operating environment, support multi-source data fusion and online deployment, and improve the intelligence level and anti-interference capability of power grid dispatch management. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a flowchart illustrating the model of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0027] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Example 1 A smart scheduling system for power grid outage planning based on improved SAC, such as Figure 1 As shown, it includes: The objective function construction module is used to build a multi-objective function for power grid outage plans based on electricity demand data. The constraint construction module is used to establish multi-dimensional constraints for power grid outage plans based on the safe operation data of maintenance equipment. The Markov decision process modeling module is used to construct an intelligent learning framework for power grid outage plans based on Markov decisions using the multi-objective function and multi-dimensional constraints. The intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function. The action trimming module, based on multi-dimensional constraints, eliminates unselectable plans set in the power grid action space when scheduling power outage plans on the decision day in the power grid state space, thus obtaining the trimmed power grid action space. The SAC network construction module is used to learn the optimal policy through the interaction between the agent and the environment based on the power grid state space, the pruned power grid action space and the power grid reward function, and obtain the trained SAC network. The SAC network application module is used to deploy the strategy network parameters of the trained SAC network into the power grid dispatch intelligent decision support system. In the power grid dispatch intelligent decision support system, a multi-dimensional state vector is constructed based on the periodically collected power grid operation data to obtain an intelligent scheduling scheme for power outage plans.

[0028] For safe operation data, some preferred implementation schemes include maintenance time windows (i.e., the earliest executable time and the latest completion time of each plan), the number of days required to complete the maintenance, plan execution status, application time for non-changeable plans, co-shutdown constraints of different maintenance plans, mutual exclusion constraints of different maintenance plans, maintenance resource limits, line transmission capacity limits, active power output limits of each thermal power unit, power grid flow information (including node voltage, line power, etc.), load forecast data, and renewable energy output forecasts.

[0029] In some preferred implementation schemes, power grid operation data is periodically collected from the following multi-source platforms: The dispatch automation system provides real-time power grid operating status, such as power flow distribution and equipment status; The equipment outage management system provides relevant information about the outage plans to be scheduled, including plan attributes and constraints. The meteorological service platform provides weather forecast data to assess the impact of new energy output (such as wind speed and solar radiation) on the power grid; The new energy power prediction system provides output prediction data for new energy sources such as photovoltaics and wind power; After integrating the above system data, the core data includes the following: The power grid operating status is used to reflect the real-time dynamics of the power grid, such as node voltage, line power, and equipment commissioning status. The power outage plan to be scheduled includes information such as the plan's constraints, outage duration, and application time. Load forecasting data is used to predict total load demand for future periods and to assess power supply reliability. New energy output forecasting is used to predict the power generation of new energy sources such as photovoltaics and wind power, so as to promote consumption and reduce power curtailment; Other auxiliary data, such as decision time, remaining days, and equipment maintenance status, together constitute a multi-dimensional state vector.

[0030] In some preferred embodiments, it also includes an experience importance sampling module, which calculates the importance value and sampling probability of each experience sample based on the reward value of the experience sample when sampling experience samples from the experience pool during the SAC network training process, so as to strengthen the learning of key experience samples and optimize the training process of the SAC network.

[0031] In some preferred implementations, it also includes a verification module for verifying the intelligent scheduling scheme of the power outage plan to ensure that it meets the safety constraints of power grid operation; if the verification passes, a formal power outage plan scheduling scheme is generated and pushed to the dispatcher for secondary confirmation and corresponding manual adjustments; if the verification fails, a human-machine collaborative intervention mechanism is activated to indicate risk points and provide alternative schemes.

[0032] In an optional implementation, a modular architecture for the entire intelligent orchestration system is defined, including objective function construction, constraint construction, Markov decision process modeling, action pruning, SAC network construction, and application modules. This decomposes the complex power outage planning problem into manageable subsystems, enabling the system to automate multi-objective optimization and dynamic constraints, thereby reducing reliance on human experience. An integrated framework improves orchestration efficiency, ensuring grid operation safety, power supply reliability, and renewable energy integration. Simultaneously, the self-learning capability of the SAC algorithm adapts to changes in the grid environment, achieving seamless integration from data acquisition to decision output.

[0033] In some preferred embodiments, the objective function construction module, specifically the multi-objective function for establishing a power grid outage plan based on electricity demand data, is used for: For monthly maintenance planning cycles, set penalty costs for failure to schedule maintenance. The specific formula is as follows: in, This refers to the number of equipment items scheduled for maintenance each month that are not included in the current period's maintenance plan. For the first The penalty coefficient for un-programmed equipment is set based on the equipment's importance, safety risk level, and maintenance time required. The specific formula for the total load loss during the monthly maintenance plan period is as follows: in, For the first The actual load loss caused by maintenance. The number of days in the monthly maintenance plan scheduling cycle; For minimizing renewable energy sources, the total amount of renewable energy sources that will be rendered useless during the monthly maintenance plan cycle is: in, For the first The abandoned electricity from Tianguang Photovoltaic Power Station; For the first The abandoned electricity of Tianfeng Power Plant; The linear weighted method is used to address the three optimization objectives: minimizing the penalty cost of unscheduled maintenance, the total load loss, and the total amount of renewable energy curtailment. The overall objective function is: ; in, , , They are respectively , , The target reward weighting coefficient.

[0034] In the optional implementation, the objective function construction module is further refined, specifically defining the calculation formulas for the unscheduled penalty cost, total load loss, and total renewable energy curtailment. A linear weighted method is used to integrate multiple objectives, quantifying the abstract optimization objectives into calculable indicators. This provides a clear reward signal for the SAC algorithm, guiding the agent to learn to balance safety, supply, and consumption. This ensures that the power outage scheduling not only considers the urgency of equipment maintenance but also takes into account user electricity demand and environmental protection requirements, thereby avoiding suboptimal decisions caused by objective conflicts in traditional methods.

[0035] In some preferred embodiments, the constraint construction module, which establishes multi-dimensional constraints for the power grid outage plan based on the safe operation data of the maintenance equipment, is specifically used for: Set maintenance time constraints for power outage plans: ; in, For the first The earliest executable time for the planned power outage; For the first The latest completion time for the planned power outage; For the first The actual start time of the planned power outage; For the first The estimated time required for maintenance of the planned power outage; Set maintenance continuity constraints for power outage plans: in, For the first The power outage plan was in the The execution status of the day, =1 indicates the first Heavenly The equipment corresponding to this power outage plan is under maintenance. =0 indicates the first Heavenly The equipment corresponding to this power outage plan is not under maintenance. If the first For power outage plans involving critical infrastructure, the timing must be determined by the applicant and cannot be adjusted. The planned power outage must meet the non-changeability constraint, and its start time for maintenance must be specified. It must be equal to the application time. .

[0036] When the The equipment maintenance corresponding to the planned power outage will affect the first When the equipment corresponding to the planned power outage is operating normally, maintenance of the affected equipment must be carried out simultaneously to reduce power outage losses, i.e., the first... The power outage plan and the first The planned power outage must meet the simultaneous outage constraint: in, The start and end times of the plan with the shorter power outage duration in a set of same-outage constraint plans; Let the first The power outage plan and the first If two power outage plans satisfy a mutual exclusion relationship, then the outage periods of the two plans cannot overlap, that is, the first... The power outage plan and the first The power outage plan must satisfy mutual exclusion constraints, and the specific formula is as follows: Setting up a power outage schedule must meet maintenance resource constraints, and the specific formula is as follows: in, This is the maximum number of devices that can be repaired simultaneously. This represents the total number of maintenance tasks planned for the current period. Set line transmission capacity constraints: in, For the first Skyline The actual transmitted active power; For the line The maximum allowed transmission power; Set power balance constraints: ; in, for Total system load during the time period for System load loss during the time period , , for Time period Taiwan thermal power unit, the first The first photovoltaic power station, the first The grid-connected power of each wind farm , , These represent the number of thermal power units, the number of photovoltaic power stations, and the number of wind farms, respectively.

[0037] The specific formula for setting upper and lower limit constraints is as follows: in, The first Minimum and maximum technical output of Taiwan's thermal power units; For the first Taiwan thermal power units Start / stop status during a time period =1 indicates the first The thermal power unit is in operation. =0 indicates the first The thermal power units in Taiwan are currently out of service.

[0038] In the optional implementation scheme, a constraint construction module is specified in detail, including maintenance time, continuity, non-changeability, simultaneous shutdown, mutual exclusion, resource constraints, line capacity, power balance and unit output constraints. This comprehensively covers the physical and operational constraints in the actual operation of the power grid, ensuring that the generated plan is feasible and safe in addition to theoretical optimization. By modeling complex relationships in the real world (such as inter-equipment influence and resource bottlenecks), the scheme prevents power flow from exceeding limits or power outages caused by the scheduling, thereby reducing execution risks and improving the practicality of the scheme.

[0039] In some preferred embodiments, the Markov decision process modeling module constructs an intelligent learning framework for the power grid outage plan based on Markov decisions. This intelligent learning framework includes a power grid state space, a power grid action space, and a power grid reward function, specifically used for: The power grid outage plan is modeled as a Markov decision process, and the power grid state space, power grid action space, power grid state transition and power grid reward function are defined according to the characteristics of the outage plan problem. State Space Construction: A state vector integrating multiple dimensions was constructed. Specifically defined as , among which decision day Indicates the current optimization decision time and power outage plan attributes. This includes planned constraints and planned power outage duration, and the number of remaining days. This refers to the remaining programmable decision windows within the current cycle, and the equipment maintenance status. Reflects the real-time status of power outage equipment in the system and power flow information. As runtime parameters, the above-mentioned attribute parameters are integrated to construct... The state vector of a phased agent ; Constructing the action space: The action space is defined as the decision day. The set of executable power outage plans can be represented as , express The actions taken by the agent at each stage ,when hour, This means that no plans will be made for the time being, and the remaining plans will be considered on the next day; when At that time, it means that the plan will be implemented. Arranged to the number Maintenance will begin today; Constructing state transitions: when Actions taken by the phase agent At that time, power outage plan attributes and equipment maintenance status Remain unchanged, decision day Remaining days Calculate the updated power flow state information of the power grid. ,get Phase State ;when At that time, decision day Power outage plan attributes and remaining days Remain unchanged, according to the new arrangement plan Update equipment maintenance status Calculate the updated power flow state information of the power grid. ,get Phase State .

[0040] Constructing a reward function: A multi-dimensional reward function evaluation system is constructed to determine the penalty cost indicator for not scheduling maintenance, the total load loss indicator within the monthly maintenance plan cycle, and the total amount of renewable energy curtailment within the monthly maintenance plan cycle. in, The number of unplanned renewable energy abandonments on the decision date is [number to be filled in]. , The amount of abandoned electricity from the n-day photovoltaic power plant is determined on the decision-making date. The amount of abandoned electricity from wind farm n on the decision-making day. This is the final state, meaning that all power outage plans have been completed or the remaining power outage plans cannot be scheduled due to constraints. for In the stage orchestration state, R represents the reward for completing the orchestration of all plans in the final state. , , They are respectively , , The target reward weighting coefficient.

[0041] Based on the constraints of the power grid outage scheduling problem, the action space is dynamically pruned. That is, when scheduling outage plans on decision day n, unselectable plans in the action space are eliminated, including already scheduled plans, outage plans that do not meet maintenance time constraints, unchangeable constraints, simultaneous outage constraints, and mutual exclusion constraints. By setting the sampling probability of the corresponding actions of these plans to 0, it is ensured that these actions will not be selected, thereby improving the search efficiency of the algorithm.

[0042] In some preferred embodiments, the unselectable plans include those that have been scheduled, do not meet maintenance time constraints, cannot be changed, have simultaneous outage constraints, and have mutual exclusion constraints. The unselectable plans are eliminated by setting the sampling probability of the corresponding actions of these plans to 0.

[0043] For grid state transitions, in some preferred embodiments, an intelligent learning framework is constructed for grid outage plans based on Markov decisions. The intelligent learning framework includes grid state space, grid action space, grid reward function, and grid state transitions.

[0044] In an optional implementation, the power outage scheduling is modeled as a Markov decision process, with a clearly defined state space, action space, state transition, and reward function. This transforms the dynamic scheduling problem into a sequential decision problem that can be handled by reinforcement learning, enabling the SAC algorithm to learn long-term strategies through state awareness and action selection, capture the temporal changes in the power grid state, enhance the algorithm's adaptability to uncertainties, and thus improve the intelligence and robustness of the scheduling.

[0045] In some preferred embodiments, a SAC network guided by power outage plan constraints is proposed to achieve automatic scheduling of power outage plans. This is achieved by embedding prior knowledge into the SAC reinforcement learning framework to optimize policy exploration direction and improve decision-making efficiency. In the SAC network construction module, the optimal policy is learned through agent-environment interaction, resulting in a trained SAC network, specifically including: Initialize and set the network parameters and initialization status of the number of Critic networks and Actor networks, and set the total number of training epochs Q and the experience pool size. In an optional embodiment, the network parameters of four Critic networks and one Actor network are initialized. Determine the decision date If the application time is an unchangeable plan, then the unchangeable plan will be directly scheduled to the decision date if the constraints are met. Determine whether the decision date is the latest possible scheduling date for the unscheduled maintenance time constraint plan. If so, schedule the maintenance time constraint plan to the decision date, provided that the constraint conditions are met, and update the current status. Based on the current state, the action space is pruned, and normalized probabilities are sampled for action selection: selecting a non-zero action indicates that the plan will be scheduled for the decision date. For simultaneous stop plans... , , If you choose the plan Then, under the premise of satisfying the constraints, the plan is arranged simultaneously. and plans That is, the first The actual start time of the planned power outage is equal to the first The actual start time of the planned power outage. If you choose the plan Then update the first For the plan The latest time that can be scheduled; Based on the orchestration scheme corresponding to the selected action, the minimum load shedding and renewable energy curtailment under power flow constraints are obtained using the power system analysis toolkit, and the reward value is calculated. Update the next status and empirical samples The data is stored in the experience pool. A batch of experience samples is sampled from the experience pool to train the SAC network and update the network parameters. When the number of samples in the experience pool reaches the preset capacity limit, the data in the experience pool is deleted. When all planned arrangements are completed or no effective actions can be performed, the round termination condition is met, the power outage plan arrangement scheme is output, the current round q ends, and the next round begins.

[0046] In some preferred embodiments, the optimal strategy after training is deployed in the intelligent decision support system for power grid dispatching to automate the scheduling of power outage plans. Specifically, the online application process is as follows: First, the system periodically collects multi-source data, including grid operation status, outage plans to be scheduled, load forecast data, and renewable energy output forecast data, from the dispatch automation system, equipment outage management system, meteorological service platform, and renewable energy power forecast system. Based on preset data processing rules and coding mechanisms, it constructs a state vector that meets the model input requirements. Then, this state vector is input into a pre-trained SAC strategy network, which directly outputs the corresponding outage plan scheduling scheme. Simultaneously, the system verifies the output outage plan scheduling scheme to ensure it meets grid operation safety constraints. If the verification passes, a formal outage plan scheduling scheme is generated and pushed to the dispatcher for secondary confirmation and manual adjustments. If the verification fails, a human-machine collaborative intervention mechanism is activated, highlighting risk points and providing alternative solutions.

[0047] In an optional implementation, the specific steps of building the SAC network are described, including network initialization, constraint-guided action selection, experience pool management and round training. Prior knowledge (such as constraint pruning) is embedded to optimize the exploration process, accelerate convergence and avoid invalid actions. At the same time, interactive learning is used to achieve policy optimization, and an efficient and stable policy network is trained. It can learn autonomously in a simulated environment offline, reduce the computational overhead when applied online, and ensure that the orchestration scheme satisfies the constraints and is close to the global optimum.

[0048] In some preferred embodiments, such as Figure 2 As shown, taking the equipment under maintenance in the regional power grid as the research object, this paper constructs a multi-objective function and multi-dimensional constraints, and optimizes the scheduling of maintenance plans based on action pruning and scheduling rules. The specific steps include: S1. From the perspectives of ensuring the safe operation of the power grid, ensuring the electricity demand of users, and ensuring the effective consumption of new energy sources, establish the objective function for the power grid outage planning problem. S2. From the perspectives of the time requirements of the maintenance equipment itself, the influence relationship between maintenance equipment, and the maintenance resource requirements, establish the constraints for the power grid outage maintenance plan scheduling problem. S3. Based on the objective function and constraints of the power grid outage planning problem, the problem is modeled as a Markov decision process, and its state space, action space, state transition, reward function and other reinforcement learning elements are defined. A deep reinforcement learning model based on SAC is then constructed. Example 2 A smart scheduling method for power grid outage plans based on improved SAC, comprising: Establish a multi-objective function for power grid outage planning based on electricity demand data; Establish multi-dimensional constraints for power grid outage plans based on the safe operation data of maintenance equipment; Using the aforementioned multi-objective function and multi-dimensional constraints, an intelligent learning framework for power grid outage planning is constructed based on Markov decision-making. The intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function. Based on multi-dimensional constraints, when scheduling power outage plans on the decision day in the power grid state space, the unselectable plans set in the power grid action space are eliminated to obtain the trimmed power grid action space. Based on the power grid state space, the pruned power grid action space, and the power grid reward function, the optimal policy is learned through the interaction between the agent and the environment, resulting in the trained SAC network. The policy network parameters of the trained SAC network are deployed in the intelligent decision support system for power grid dispatch. In the intelligent decision support system for power grid dispatch, a multi-dimensional state vector is constructed based on the periodically collected power grid operation data to obtain an intelligent scheduling scheme for power outage plans.

[0049] In some preferred embodiments, the multi-objective function for establishing the power grid outage plan based on electricity demand data specifically includes: For monthly maintenance planning cycles, set penalty costs for failure to schedule maintenance. The specific formula is as follows: in, This refers to the number of equipment items scheduled for maintenance each month that are not included in the current period's maintenance plan. For the first The penalty coefficient for un-programmed equipment is set based on the equipment's importance, safety risk level, and maintenance time required. The specific formula for the total load loss during the monthly maintenance plan period is as follows: in, For the first The actual load loss caused by maintenance. The number of days in the monthly maintenance plan scheduling cycle; For minimizing renewable energy sources, the total amount of renewable energy sources that will be rendered useless during the monthly maintenance plan cycle is: in, For the first The abandoned electricity from Tianguang Photovoltaic Power Station; For the first The abandoned electricity of Tianfeng Power Plant; The linear weighted method is used to address the three optimization objectives: minimizing the penalty cost of unscheduled maintenance, the total load loss, and the total amount of renewable energy curtailment. The overall objective function is: ; in, , , They are respectively , , The target reward weighting coefficient.

[0050] In some preferred embodiments, the multi-dimensional constraints for establishing the power grid outage plan based on the safe operation data of the maintenance equipment specifically include: Set maintenance time constraints for power outage plans: ; in, For the first The earliest executable time for the planned power outage; For the first The latest completion time for the planned power outage; For the first The actual start time of the planned power outage; For the first The estimated time required for maintenance of the planned power outage; Set maintenance continuity constraints for power outage plans: in, For the first The power outage plan was in the The execution status of the day, =1 indicates the first Heavenly The equipment corresponding to this power outage plan is under maintenance. =0 indicates the first Heavenly The equipment corresponding to this power outage plan is not under maintenance. If the first For power outage plans involving critical infrastructure, the timing must be determined by the applicant and cannot be adjusted. The planned power outage must meet the non-changeability constraint, and its start time for maintenance must be specified. It must be equal to the application time. .

[0051] When the The equipment maintenance corresponding to the planned power outage will affect the first When the equipment corresponding to the planned power outage is operating normally, maintenance of the affected equipment must be carried out simultaneously to reduce power outage losses, i.e., the first... The power outage plan and the first The planned power outage must meet the simultaneous outage constraint: in, The start and end times of the plan with the shorter power outage duration in a set of same-outage constraint plans; Let the first The power outage plan and the first If two power outage plans satisfy a mutual exclusion relationship, then the outage periods of the two plans cannot overlap, that is, the first... The power outage plan and the first The power outage plan must satisfy mutual exclusion constraints, and the specific formula is as follows: Setting up a power outage schedule must meet maintenance resource constraints, and the specific formula is as follows: in, This is the maximum number of devices that can be repaired simultaneously. This represents the total number of maintenance tasks planned for the current period. Set line transmission capacity constraints: in, For the first Skyline The actual transmitted active power; For the line The maximum allowed transmission power; Set power balance constraints: ; Among them, for Total system load during the time period for System load loss during the time period , , for Time period Taiwan thermal power unit, the first The first photovoltaic power station, the first The grid-connected power of each wind farm , , These represent the number of thermal power units, the number of photovoltaic power stations, and the number of wind farms, respectively.

[0052] The specific formula for setting upper and lower limit constraints is as follows: in, The first Minimum and maximum technical output of Taiwan's thermal power units; For the first Taiwan thermal power units Start / stop status during a time period =1 indicates the first The thermal power unit is in operation. =0 indicates the first The thermal power units in Taiwan are currently out of service.

[0053] In some preferred implementations, an intelligent learning framework for grid outage planning is constructed based on Markov decision-making. This intelligent learning framework includes a grid state space, a grid action space, and a grid reward function, specifically comprising: The power grid outage plan is modeled as a Markov decision process, and the power grid state space, power grid action space, power grid state transition and power grid reward function are defined according to the characteristics of the outage plan problem. State Space Construction: A state vector integrating multiple dimensions was constructed. Specifically defined as , among which decision day Indicates the current optimization decision time and power outage plan attributes. This includes planned constraints and planned power outage duration, and the number of remaining days. This refers to the remaining programmable decision windows within the current cycle, and the equipment maintenance status. Reflects the real-time status of power outage equipment in the system and power flow information. As runtime parameters, the above-mentioned attribute parameters are integrated to construct... The state vector of a phased agent ; Constructing the action space: The action space is defined as the decision day. The set of executable power outage plans can be represented as , express The actions taken by the agent at each stage ,when hour, This means that no plans will be made for the time being, and the remaining plans will be considered on the next day; when At that time, it means that the plan will be implemented. Arranged to the number Maintenance will begin today; Constructing state transitions: when Actions taken by the phase agent At that time, power outage plan attributes and equipment maintenance status Remain unchanged, decision day Remaining days Calculate the updated power flow state information of the power grid. ,get Phase State ;when At that time, decision day Power outage plan attributes and remaining days Remain unchanged, according to the new arrangement plan Update equipment maintenance status Calculate the updated power flow state information of the power grid. ,get Phase State .

[0054] Constructing a reward function: A multi-dimensional reward function evaluation system is constructed to determine the penalty cost indicator for not scheduling maintenance, the total load loss indicator within the monthly maintenance plan cycle, and the total amount of renewable energy curtailment within the monthly maintenance plan cycle. in, The number of unplanned renewable energy abandonments on the decision date is [number to be filled in]. , The amount of abandoned electricity from the n-day photovoltaic power plant is determined on the decision-making date. The amount of abandoned electricity from wind farm n on the decision-making day. This is the final state, meaning that all power outage plans have been completed or the remaining power outage plans cannot be scheduled due to constraints. for In the stage orchestration state, R represents the reward for completing the orchestration of all plans in the final state. , , They are respectively , , The target reward weighting coefficient.

[0055] In some preferred implementations, the optimal policy is learned through interaction between the agent and the environment to obtain the trained SAC network, specifically including: Initialize and set the network parameters and initialization status of the number of Critic networks and Actor networks, and set the total number of training epochs Q and the experience pool size. ; Determine the decision date If the application time is an unchangeable plan, then the unchangeable plan will be directly scheduled to the decision date if the constraints are met. Determine whether the decision date is the latest possible scheduling date for the unscheduled maintenance time constraint plan. If so, schedule the maintenance time constraint plan to the decision date, provided that the constraint conditions are met, and update the current status. Based on the current state, the action space is pruned, and normalized probabilities are sampled for action selection: selecting a non-zero action indicates that the plan will be scheduled for the decision date. For simultaneous stop plans... , , If you choose the plan Then, under the premise of satisfying the constraints, the plan is arranged simultaneously. and plans That is, the first The actual start time of the planned power outage is equal to the first The actual start time of the planned power outage. If you choose the plan Then update the first For the plan The latest time that can be scheduled; Based on the orchestration scheme corresponding to the selected action, the minimum load shedding and renewable energy curtailment under power flow constraints are obtained using the power system analysis toolkit, and the reward value is calculated. Update the next status and empirical samples The data is stored in the experience pool. A batch of experience samples is sampled from the experience pool to train the SAC network and update the network parameters. When the number of samples in the experience pool reaches the preset capacity limit, the data in the experience pool is deleted. When all planned arrangements are completed or no effective actions can be performed subsequently, the round termination condition is met, the power outage plan arrangement scheme is output, the current round ends, and the next round begins.

[0056] Example 3 This embodiment provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in Embodiment 2.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the invention, but these changes, modifications or equivalent substitutions are all within the scope of protection of the pending claims of the invention.

[0061] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A smart scheduling system for power grid outage plans based on improved SAC, characterized in that, It includes: The objective function construction module is used to build a multi-objective function for power grid outage plans based on electricity demand data. The constraint construction module is used to establish multi-dimensional constraints for power grid outage plans based on the safe operation data of maintenance equipment. The Markov decision process modeling module is used to construct an intelligent learning framework for power grid outage plans based on Markov decisions using the multi-objective function and multi-dimensional constraints. The intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function. The action trimming module, based on multi-dimensional constraints, eliminates unselectable plans set in the power grid action space when scheduling power outage plans on the decision day in the power grid state space, thus obtaining the trimmed power grid action space. The SAC network construction module is used to learn the optimal policy through the interaction between the agent and the environment based on the power grid state space, the pruned power grid action space and the power grid reward function, and obtain the trained SAC network. The SAC network application module is used to deploy the strategy network parameters of the trained SAC network into the power grid dispatch intelligent decision support system. In the power grid dispatch intelligent decision support system, a multi-dimensional state vector is constructed based on the periodically collected power grid operation data to obtain an intelligent scheduling scheme for power outage plans.

2. The intelligent scheduling system for power grid outage planning based on improved SAC according to claim 1, characterized in that, In the objective function construction module, the multi-objective function for establishing a power grid outage plan based on electricity demand data is specifically used for: For monthly maintenance planning cycles, set penalty costs for failure to schedule maintenance. The specific formula is as follows: in, This refers to the number of equipment items scheduled for maintenance each month that are not included in the current period's maintenance plan. For the first The penalty coefficient for un-programmed equipment is set based on the equipment's importance, safety risk level, and maintenance time required. The specific formula for the total load loss during the monthly maintenance plan period is as follows: in, For the first The actual load loss caused by maintenance. The number of days in the monthly maintenance plan scheduling cycle; For minimizing renewable energy sources, the total amount of renewable energy sources that will be rendered obsolete within the monthly maintenance plan cycle is as follows: in, For the first The abandoned electricity from Tianguang Photovoltaic Power Station; For the first The abandoned electricity of Tianfeng Power Plant; The linear weighted method is used to address the three optimization objectives: minimizing the penalty cost of unscheduled maintenance, the total load loss, and the total amount of renewable energy curtailment. The overall objective function is: ； in, , , They are respectively , , The target reward weighting coefficient.

3. The intelligent scheduling system for power grid outage planning based on improved SAC according to claim 2, characterized in that, In the constraint construction module, the multi-dimensional constraint conditions for establishing the power grid outage plan based on the safe operation data of the maintenance equipment are specifically used for: Set maintenance time constraints for power outage plans: ； in, For the first The earliest executable time for the planned power outage; For the first The latest completion time for the planned power outage; For the first The actual start time of the planned power outage; For the first The estimated time required for maintenance of the planned power outage; Set maintenance continuity constraints for power outage plans: in, For the first The power outage plan was in the The execution status of the day, =1 indicates the first Heavenly The equipment corresponding to this power outage plan is under maintenance. =0 indicates the first Heavenly The equipment corresponding to this power outage plan is not under maintenance. If the first For power outage plans involving critical infrastructure, the timing must be determined by the applicant and cannot be adjusted. The planned power outage must meet the non-changeability constraint, and its start time for maintenance must be specified. It must be equal to the application time. ; When the The equipment maintenance corresponding to the planned power outage will affect the first When the equipment corresponding to the planned power outage is operating normally, maintenance of the affected equipment must be carried out simultaneously to reduce power outage losses, i.e., the first... The power outage plan and the first The planned power outage must meet the simultaneous outage constraint: in, The start and end times of the plan with the shorter power outage duration in a set of same-outage constraint plans; Let the first The power outage plan and the first If two power outage plans satisfy a mutual exclusion relationship, then the outage periods of the two plans cannot overlap, that is, the first... The power outage plan and the first The power outage plan must satisfy mutual exclusion constraints, and the specific formula is as follows: Setting up a power outage schedule must meet maintenance resource constraints, and the specific formula is as follows: in, This is the maximum number of devices that can be repaired simultaneously. This represents the total number of maintenance tasks planned for the current period. Set line transmission capacity constraints: in, For the first Skyline The actual transmitted active power; For the line The maximum allowed transmission power; Set power balance constraints: ； in, for Total system load during the time period for System load loss during the time period , , for Time period Taiwan thermal power unit, the first The first photovoltaic power station, the first The grid-connected power of each wind farm , , These represent the number of thermal power units, the number of photovoltaic power plants, and the number of wind farms, respectively. The specific formula for setting upper and lower limit constraints is as follows: in, The first Minimum and maximum technical output of Taiwan's thermal power units; For the first Taiwan thermal power units Start / stop status during a time period =1 indicates the first The thermal power unit is in operation. =0 indicates the first The thermal power units in Taiwan are currently out of service.

4. The intelligent scheduling system for power grid outage planning based on improved SAC according to claim 3, characterized in that, In the Markov decision process modeling module, an intelligent learning framework is constructed for the power grid outage plan based on Markov decisions. This intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function, specifically used for: The power grid outage plan is modeled as a Markov decision process, and the power grid state space, power grid action space, power grid state transition and power grid reward function are defined according to the characteristics of the outage plan problem. State Space Construction: A state vector integrating multiple dimensions was constructed. Specifically defined as , among which decision day Indicates the current optimization decision time and power outage plan attributes. This includes planned constraints and planned power outage duration, and the number of remaining days. This refers to the remaining programmable decision windows within the current cycle, and the equipment maintenance status. Reflects the real-time status of power outage equipment in the system and power flow information. As runtime parameters, the above-mentioned attribute parameters are integrated to construct... The state vector of a phased agent ; Constructing the action space: The action space is defined as the decision day. The set of executable power outage plans can be represented as , express The actions taken by the agent at each stage ,when hour, That is, no plans will be made for the time being, and the remaining plans will be considered on the next day; when At that time, it means that the plan will be implemented. Arranged to the number Maintenance will begin today; Constructing state transitions: when Actions taken by the phase agent At that time, power outage plan attributes and equipment maintenance status Remain unchanged, decision day Remaining days Calculate the updated power flow state information of the power grid. ,get Phase State ;when At that time, decision day Power outage plan attributes and remaining days Remain unchanged, according to the new arrangement plan Update equipment maintenance status Calculate the updated power flow state information of the power grid. ,get Phase State ; Constructing a reward function: A multi-dimensional reward function evaluation system is constructed to determine the penalty cost indicator for not scheduling maintenance, the total load loss indicator within the monthly maintenance plan cycle, and the total amount of renewable energy curtailment within the monthly maintenance plan cycle. in, The number of unplanned renewable energy abandonments on the decision date is [not specified]. , The amount of abandoned electricity from the n-day photovoltaic power plant is determined on the decision-making date. The amount of abandoned electricity from wind farm n on the decision-making day. This is the final state, meaning that all power outage plans have been completed or the remaining power outage plans cannot be scheduled due to constraints. for The stage orchestration state, where R is the reward for completing the orchestration of all plans in the final state. , , They are respectively , , The target reward weighting coefficient.

5. The intelligent scheduling system for power grid outage planning based on improved SAC according to claim 4, characterized in that, The SAC network construction module learns the optimal policy through the interaction between the agent and the environment, resulting in the trained SAC network, which specifically includes: Initialize and set the network parameters and initialization status of the number of Critic networks and Actor networks, and set the total number of training epochs Q and the experience pool size. ; Determine the decision date If the application time is an unchangeable plan, then the unchangeable plan will be directly scheduled to the decision date if the constraints are met. Determine whether the decision date is the latest possible scheduling date for the unscheduled maintenance time constraint plan. If so, schedule the maintenance time constraint plan to the decision date, provided that the constraint conditions are met, and update the current status. Based on the current state, the action space is pruned, and normalized probabilities are sampled for action selection: selecting a non-zero action indicates that the plan will be scheduled for the decision date. For simultaneous stop plans... , , If you choose the plan Then, under the premise of satisfying the constraints, the plan is arranged simultaneously. and plans That is, the first The actual start time of the planned power outage is equal to the first The actual start time of the planned power outage. If you choose the plan Then update the first For the plan The latest time that can be scheduled; Based on the orchestration scheme corresponding to the selected action, the minimum load shedding and renewable energy curtailment under power flow constraints are obtained using the power system analysis toolkit, and the reward value is calculated. Update the next status and empirical samples The data is stored in the experience pool. A batch of experience samples is sampled from the experience pool to train the SAC network and update the network parameters. When the number of samples in the experience pool reaches the preset capacity limit, the data in the experience pool is deleted. When all planned arrangements are completed or no effective actions can be performed subsequently, the round termination condition is met, the power outage plan arrangement scheme is output, the current round ends, and the next round begins.

6. A method for intelligent scheduling of power grid outage plans based on improved SAC, characterized in that, It includes: Establish a multi-objective function for power grid outage planning based on electricity demand data; Establish multi-dimensional constraints for power grid outage plans based on the safe operation data of maintenance equipment; Using the aforementioned multi-objective function and multi-dimensional constraints, an intelligent learning framework for power grid outage planning is constructed based on Markov decision-making. The intelligent learning framework includes the power grid state space, the power grid action space, and the power grid reward function. Based on multi-dimensional constraints, when scheduling power outage plans on the decision day in the power grid state space, the unselectable plans set in the power grid action space are eliminated to obtain the trimmed power grid action space. Based on the power grid state space, the pruned power grid action space, and the power grid reward function, the optimal policy is learned through the interaction between the agent and the environment, resulting in the trained SAC network. The policy network parameters of the trained SAC network are deployed in the intelligent decision support system for power grid dispatch. In the intelligent decision support system for power grid dispatch, a multi-dimensional state vector is constructed based on the periodically collected power grid operation data to obtain an intelligent scheduling scheme for power outage plans.

7. The intelligent scheduling method for power grid outage plans based on improved SAC according to claim 6, characterized in that, The multi-objective function for establishing a power grid outage plan based on electricity demand data specifically includes: For monthly maintenance planning cycles, set penalty costs for failure to schedule maintenance. The specific formula is as follows: in, This refers to the number of equipment items scheduled for maintenance each month that are not included in the current period's maintenance plan. For the first The penalty coefficient for un-programmed equipment is set based on the equipment's importance, safety risk level, and maintenance time required. The specific formula for the total load loss during the monthly maintenance plan period is as follows: in, For the first The actual load loss caused by maintenance. The number of days in the monthly maintenance plan scheduling cycle; For minimizing renewable energy sources, the total amount of renewable energy sources that will be rendered obsolete within the monthly maintenance plan cycle is as follows: in, For the first The abandoned electricity from Tianguang Photovoltaic Power Station; For the first The abandoned electricity of Tianfeng Power Plant; The linear weighted method is used to address the three optimization objectives: minimizing the penalty cost of unscheduled maintenance, the total load loss, and the total amount of renewable energy curtailment. The overall objective function is: ； in, , , They are respectively , , The target reward weighting coefficient.

8. The intelligent scheduling method for power grid outage plans based on improved SAC according to claim 7, characterized in that, The multi-dimensional constraints for establishing the power grid outage plan based on the safe operation data of the maintenance equipment specifically include: Set maintenance time constraints for power outage plans: ； in, For the first The earliest executable time for the planned power outage; For the first The latest completion time for the planned power outage; For the first The actual start time of the planned power outage; For the first The estimated time required for maintenance of the planned power outage; Set maintenance continuity constraints for power outage plans: in, For the first The power outage plan was in the The execution status of the day, =1 indicates the first Heavenly The equipment corresponding to this power outage plan is under maintenance. =0 indicates the first Heavenly The equipment corresponding to this power outage plan is not under maintenance. If the first For power outage plans involving critical infrastructure, the timing must be determined by the applicant and cannot be adjusted. The planned power outage must meet the non-changeability constraint, and its start time for maintenance must be specified. It must be equal to the application time. ; When the The equipment maintenance corresponding to the planned power outage will affect the first When the equipment corresponding to the planned power outage is operating normally, maintenance of the affected equipment must be carried out simultaneously to reduce power outage losses, i.e., the first... The power outage plan and the first The planned power outage must meet the simultaneous outage constraint: in, The start and end times of the plan with the shorter power outage duration in a set of same-outage constraint plans; Let the first The power outage plan and the first If two power outage plans satisfy a mutual exclusion relationship, then the outage periods of the two plans cannot overlap, that is, the first... The power outage plan and the first The power outage plan must satisfy mutual exclusion constraints, and the specific formula is as follows: Setting up a power outage schedule must meet maintenance resource constraints, and the specific formula is as follows: in, This is the maximum number of devices that can be repaired simultaneously. This represents the total number of maintenance tasks planned for the current period. Set line transmission capacity constraints: in, For the first Skyline The actual transmitted active power; For the line The maximum allowed transmission power; Set power balance constraints: ； in, for Total system load during the time period for System load loss during the time period , , for Time period Taiwan thermal power unit, the first The first photovoltaic power station, the first The grid-connected power of each wind farm , , These represent the number of thermal power units, the number of photovoltaic power plants, and the number of wind farms, respectively. The specific formula for setting upper and lower limit constraints is as follows: in, The first Minimum and maximum technical output of Taiwan's thermal power units; For the first Taiwan thermal power units Start / stop status during a time period =1 indicates the first The thermal power unit is in operation. =0 indicates the first The thermal power units in Taiwan are currently out of service.

9. A method for intelligent scheduling of power grid outage plans based on improved SAC according to claim 8, characterized in that, Based on Markov decision-making, an intelligent learning framework is constructed for power grid outage planning. This framework includes the power grid state space, the power grid action space, and the power grid reward function, specifically comprising: The power grid outage plan is modeled as a Markov decision process, and the power grid state space, power grid action space, power grid state transition and power grid reward function are defined according to the characteristics of the outage plan problem. State Space Construction: A state vector integrating multiple dimensions was constructed. Specifically defined as , among which decision day Indicates the current optimization decision time and power outage plan attributes. This includes planned constraints and planned power outage duration, and the number of remaining days. This refers to the remaining programmable decision windows within the current cycle, and the equipment maintenance status. Reflects the real-time status of power outage equipment in the system and power flow information. As runtime parameters, the above-mentioned attribute parameters are integrated to construct... The state vector of a phased agent ; Constructing the action space: The action space is defined as the decision day. The set of executable power outage plans can be represented as , express The actions taken by the agent at each stage ,when hour, That is, no plans will be made for the time being, and the remaining plans will be considered on the next day; when At that time, it means that the plan will be implemented. Arranged to the number Maintenance will begin today; Constructing state transitions: when Actions taken by the phase agent At that time, power outage plan attributes and equipment maintenance status Remain unchanged, decision day Remaining days Calculate the updated power flow state information of the power grid. ,get Phase State ;when At that time, decision day Power outage plan attributes and remaining days Remain unchanged, according to the new arrangement plan Update equipment maintenance status Calculate the updated power flow state information of the power grid. ,get Phase State ; Constructing a reward function: A multi-dimensional reward function evaluation system is constructed to determine the penalty cost indicator for not scheduling maintenance, the total load loss indicator within the monthly maintenance plan cycle, and the total amount of renewable energy curtailment within the monthly maintenance plan cycle. in, The number of unplanned renewable energy abandonments on the decision date is [not specified]. , The amount of abandoned electricity from the n-day photovoltaic power plant is determined on the decision-making date. The amount of abandoned electricity from wind farm n on the decision-making day. This is the final state, meaning that all power outage plans have been completed or the remaining power outage plans cannot be scheduled due to constraints. for The stage orchestration state, where R is the reward for completing the orchestration of all plans in the final state. , , They are respectively , , The target reward weighting coefficient.

10. The intelligent scheduling method for power grid outage plans based on improved SAC according to claim 9, characterized in that, The optimal policy is learned through interaction between the agent and the environment, resulting in the trained SAC network, which specifically includes: Initialize and set the network parameters and initialization status of the number of Critic networks and Actor networks, and set the total number of training epochs Q and the experience pool size. ; Determine the decision date If the application time is an unchangeable plan, then the unchangeable plan will be directly scheduled to the decision date if the constraints are met. Determine whether the decision date is the latest possible scheduling date for the unscheduled maintenance time constraint plan. If so, schedule the maintenance time constraint plan to the decision date, provided that the constraint conditions are met, and update the current status. Based on the current state, the action space is pruned, and normalized probabilities are sampled for action selection: selecting a non-zero action indicates that the plan will be scheduled for the decision date. For simultaneous stop plans... , , If you choose the plan Then, under the premise of satisfying the constraints, the plan is arranged simultaneously. and plans That is, the first The actual start time of the planned power outage is equal to the first The actual start time of the planned power outage. If you choose the plan Then update the first For the plan The latest time that can be scheduled; Based on the orchestration scheme corresponding to the selected action, the minimum load shedding and renewable energy curtailment under power flow constraints are obtained using the power system analysis toolkit, and the reward value is calculated. Update the next status and empirical samples The data is stored in the experience pool. A batch of experience samples is sampled from the experience pool to train the SAC network and update the network parameters. When the number of samples in the experience pool reaches the preset capacity limit, the data in the experience pool is deleted. When all planned arrangements are completed or no effective actions can be performed subsequently, the round termination condition is met, the power outage plan arrangement scheme is output, the current round ends, and the next round begins.

11. A computer storage medium, wherein the computer-readable storage medium stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for identifying key meteorological factors affecting icing of transmission lines as described in any one of claims 6-10.

Images