A method and system for generating rolling clearing schemes for post-disaster power distribution network recovery

Abstract

This application relates to the field of power system emergency control technology, specifically to a method and system for generating rolling clearing schemes for post-disaster distribution network recovery. The method includes: acquiring real-time data of the distribution network; constructing a context feature vector for the current moment, inputting it into a pre-trained decision generation network for forward inference, and outputting a preliminary clearing decision within a rolling window; using a pre-trained physical feasibility discrimination network to perform security verification on the preliminary decision; if the confidence level is lower than a threshold, correcting it through linear projection to obtain the final clearing decision; issuing and executing the first step instruction in the final clearing decision, and then rolling the decision time to the next time step for cyclical execution. The decision generation network and the discrimination network are co-trained using a simulation dataset, which is generated by solving a single-leader-multiple-follower model reconstructed from a post-disaster recovery single-layer mixed integer linear programming model. This invention enables dynamic closed-loop adjustment and rapid generation of post-disaster recovery schemes.

Description

Technical Field

[0001] This application relates to the field of power system emergency control technology, specifically to a method and system for generating a rolling clearing scheme for post-disaster distribution network recovery. Background Technology

[0002] Extreme natural disasters such as typhoons and earthquakes often cause physical disconnections between the distribution network and the upstream main grid, forcing the system to operate in islanded mode. In such post-disaster situations, distributed energy resources and virtual power plants become key resources to support sequential load recovery after disasters. However, the privatization of power assets makes it difficult for distribution system operators to directly dispatch private resources, requiring market-based mechanisms to guide their participation.

[0003] Existing technologies employ a model-predictive open-loop decision-making model to address post-disaster distribution network recovery. In terms of the decision-making framework, existing methods primarily use a model-predictive "open-loop" decision-making model to formulate recovery plans. This involves generating a scheduling sequence covering the entire recovery cycle in the initial stages based on predictions of future states. Regarding market mechanism design, existing methods mainly use a master-slave game model based on price signals to guide distributed resource allocation, with distribution system operators as leaders and virtual power plants as followers. Incentive pricing and scheduling schemes are generated by solving a bilevel programming problem. In terms of solution algorithms, existing methods primarily model the market clearing problem as a mixed-integer nonlinear programming problem and use decomposition algorithms such as Column-and-Constraint Generation (CCG) for iterative solving.

[0004] However, existing open-loop decision-making models based on model prediction have clear shortcomings: open-loop decision-making models based on long-term predictions struggle to incorporate real-time feedback on topological status and repair progress during the recovery process; pre-generated fixed scheduling sequences are prone to deviation from the dynamically evolving actual network state, leading to physically infeasible decision-making schemes or low recovery efficiency; while a single price-guided mechanism can coordinate vertical hierarchical relationships, it fails to effectively model and resolve horizontal decision-making coupling and conflict of interest among multiple virtual power plant entities due to shared physical networks, easily inducing strategic behavior or causing unstable equilibrium solutions; even with linear approximations of the model, traditional branch-and-bound mathematical programming solutions still face combinatorial explosion risks when dealing with large-scale, multi-time-period mixed integer programming problems, and their solution time is highly uncertain, making it difficult to meet the stringent requirements of post-disaster recovery for minute-level or even second-level real-time decision-making. Summary of the Invention

[0005] To address the technical challenges of existing open-loop decision-making methods based on model prediction for post-disaster power distribution network restoration, which are ill-suited to the dynamic evolution of the disaster, neglect of multi-stakeholder game conflicts by single-price guidance mechanisms, and the inability of traditional mathematical programming solutions to meet real-time requirements, this application provides a rolling clearing scheme generation method and system for post-disaster power distribution network restoration. This method adapts to dynamic changes through a rolling closed-loop decision-making mechanism, coordinates the interests of multiple stakeholders through a guidance mechanism that integrates physical verification and network game theory, and leverages neural networks to improve the generation speed of clearing schemes. It enables dynamic tracking and closed-loop adjustment of restoration schemes to the post-disaster power grid state, effectively guiding multiple stakeholders to achieve collaborative restoration equilibrium and achieving rapid clearing, thereby improving the reliability and economy of overall power supply restoration.

[0006] Firstly, this application provides a method for generating a rolling clearing scheme for post-disaster distribution network restoration, comprising the following steps:

[0007] S1. Obtain real-time data of the distribution network, including the physical topology status of the distribution network, branch repair status, system-level load demand, load survival value factor, renewable energy output level within the virtual power plant, and private load demand;

[0008] S2. Construct the current decision-making time based on the real-time data of the distribution network at the current decision-making time. Context feature vector ,Will Input pre-trained decision generation network Perform forward inference and output the timing scrolling window. Preliminary clearing decision within Preliminary clearing decision A decision vector whose elements include a time-series scrolling window. The incentive electricity price at each time step, the load recovery amount at each node, and the switching operation instructions for each branch;

[0009] Time-series scrolling window Defined as:

[0010]

[0011] in For time step, This represents the total duration of the time-series scrolling window;

[0012] S3. Using a pre-trained physical feasibility discrimination network Preliminary clearing decision Perform security checks and output decision confidence levels;

[0013] If the decision confidence level is lower than the preset confidence level threshold, the preliminary clearing decision is corrected through linear projection to obtain the final clearing decision. ;

[0014] If the decision confidence level is not lower than the preset confidence level threshold, then the preliminary clearing decision will be made directly. As the final clearing decision ;

[0015] S4. Based on the final clearing decision Release timeline scrolling window The internal incentive price instructions and load restoration control commands are only executed. The first step of the operation will be executed, and the current decision time will be set after execution. Scroll to the next time step, then return to step S1 and repeat;

[0016] Among them, decision generation network Physical feasibility assessment network All are deep learning neural networks, using simulation datasets. The simulation dataset was obtained after co-training. The construction steps are as follows:

[0017] Simulate different post-disaster power grid scenarios and construct a set of context feature vectors. Each context feature vector It includes the physical topology status of the distribution network, branch repair status, system-level load demand, load survival value factor, renewable energy output level within the virtual power plant, and private load demand;

[0018] For each context feature vector Solve the single-level mixed-integer linear programming model for disaster recovery to obtain its corresponding optimal decision vector. The post-disaster recovery single-level mixed integer linear programming model is reconstructed from a pre-built single-leader-multiple-follower model. During the reconstruction process, the McCormick envelope method is used to handle bilinear terms.

[0019] Record all feature-decision pairs To form a simulation dataset .

[0020] It should be further explained that in step S1, the physical topology of the distribution network... This includes the real-time on / off status of each line in each distribution network and the location information of operable switches;

[0021] Branch road repair status Specifically, the repair status binary parameters for each branch include the nodes. With nodes Branch roads between The repair status binary parameter is represented as Its value of 1 indicates a branch At any moment The system has been repaired to an operational state; a value of 0 indicates a branch. Not repaired;

[0022] Survival value factor Specifically, each node is assigned a weight coefficient based on its load type and preset recovery priority. At any moment The weighting coefficient is ;

[0023] Renewable energy output level For renewable energy power generation equipment at any time The active power output value; renewable energy power generation equipment includes at least one of photovoltaic power generation equipment and wind power equipment.

[0024] It should be further explained that the context feature vector Represented as:

[0025]

[0026] in, Represents the physical topology state;

[0027] Indicates the repair status of the branch road;

[0028] This indicates system-level load demand;

[0029] Indicates the survival value factor;

[0030] Indicates the level of renewable energy output;

[0031] This indicates the private load demand within the virtual power plant.

[0032] It should be further explained that the decision-making time step The value range is 1-15 min;

[0033] Total duration of the time-tracking scrolling window The value range is 15-60 min;

[0034] .

[0035] It should be further noted that the confidence threshold in step S3 ranges from 0.5 to 0.9.

[0036] It should be further noted that in step S3, the expression for correcting the initial clearing decision through linear projection is as follows:

[0037]

[0038] in, express The constraint matrix, where each row corresponds to a physical constraint and each column corresponds to... One of the elements;

[0039] express The constraint vector is a column vector, where each element represents... The boundary values ​​of the parallel constraints, which are obtained from the original boundaries of the physical constraints through mathematical transformation, include:

[0040] For upper limit constraints, simply take the upper limit value;

[0041] For lower bound constraints, take the negative value of the lower bound.

[0042] For bilateral constraints, they are decomposed into upper limit constraints and lower limit constraints, and then their values ​​are determined according to the corresponding rules.

[0043] It should be further explained that the single-leader-multiple-follower model is a decision optimization model with a two-layer interactive architecture, including a leader decision model and multiple follower decision models that are independently connected to the leader decision model. A horizontal equilibrium mechanism is established between the various follower decision models. The leader decision model corresponds to the power distribution system operator, and each follower decision model corresponds to a virtual power plant.

[0044] Among them, the decision objective of the leader decision-making model is in the scrolling window. Internally, we will coordinate and optimize the progress of the entire network load recovery, the cost of dispatching our own resources, and the incentive costs of purchasing electricity from virtual power plants.

[0045] The decision-making objective of the follower decision-making model is to maximize individual profits. After receiving the incentive electricity price signal issued by the leader, it optimizes the output and load recovery of its internal distributed energy resources to determine the net power injected into the distribution network.

[0046] The horizontal equilibrium mechanism constructs a global situation function to address the decision-making coupling and interest game among multiple followers due to the shared physical power grid. This transforms the Nash equilibrium problem of decentralized decision-making among multiple followers into a single-objective optimization problem of this global situation function.

[0047] It should be further noted that the objective function and constraints of the leader decision-making model include:

[0048]

[0049]

[0050] in, Indicates time No. The incentive electricity price for each virtual power plant is a decision variable for leaders; An index for virtual power plants. For time indexing;

[0051] Indicates at time node The actual restored active power is a decision variable for leaders. For node indexing, For time indexing;

[0052] Indicates at time The first generation of power distribution system operators The contribution of resources to resource recovery is a decision variable for leaders. For indexing proprietary resources, For time indexing;

[0053] Indicates at time branch road The switch state binary decision variable takes a value of 1 to indicate closed and 0 to indicate open.

[0054] Indicates at time node The load survival value factor is a non-negative weighting coefficient preset according to load type and recovery priority;

[0055] Indicates at time node System-level load requirements;

[0056] This indicates that the power distribution system operator owns the first Functions for the operation, fuel, or maintenance costs of resource recovery.

[0057] Indicates at time No. The net active power injected into the distribution network by each virtual power plant;

[0058] , These represent the weighting coefficients of the distribution system operator's cost of adjusting its own resources and its expenditure on purchasing electricity from virtual power plants in the objective function, respectively.

[0059] This represents the set of all nodes in a power distribution network.

[0060] This represents the set of all virtual power plants;

[0061] This represents the collection of all proprietary restoration resources owned by the power distribution system operator.

[0062] Represents dynamic topology evolution constraints;

[0063] This indicates network trend constraints;

[0064] This indicates the constraints on the operation of the power distribution system operator's own resources;

[0065] This indicates a load sequential recovery constraint;

[0066] Indicates the repair status of the branch road Discrete confidence sets;

[0067] Indicates system-level load demand Continuous confidence intervals;

[0068] Indicates system-level load demand The baseline value for prediction at any given time;

[0069] , This indicates the minimum and maximum allowable values ​​for the incentive electricity price;

[0070] Indicates the first The profit function of a virtual power plant.

[0071] It should be further noted that the objective function and constraints of the follower decision-making model include:

[0072]

[0073]

[0074] in, Indicates the first The profit function of a virtual power plant;

[0075] Indicates at time No. The net active power injected into the distribution network by a virtual power plant is a decision variable for followers;

[0076] Indicates at time No. The first virtual power plant The output value of energy resources is a decision variable for followers. Index for resource types;

[0077] Indicates time No. Incentive electricity prices for virtual power plants;

[0078] Indicates the first The first virtual power plant Operating cost function for energy resources;

[0079] Indicates the first The collection of all energy resources within a virtual power plant;

[0080] This represents the boundary constraints for the aggregation flexibility of virtual power plants;

[0081] express Time of the first Renewable energy within a virtual power plant Real-time prediction of maximum active power output Continuous confidence intervals;

[0082] express Time of the first Private load demand within a virtual power plant Continuous confidence intervals.

[0083] It should be further noted that the global situation function is:

[0084]

[0085] in, Denotes the global state function, where It is all virtual power plants at all times The output vector is composed of the output forces. It is all virtual power plants at all times The incentive price vector is composed of incentive prices;

[0086] Indicates the first The profit function of a virtual power plant;

[0087] Indicates at time , used to characterize the The output change of the first virtual power plant affects the second... Sensitivity coefficient for the degree of voltage impact at the grid connection point of a virtual power plant;

[0088] Indicates at time No. The net active power injected into the distribution network by each virtual power plant;

[0089] Indicates at time No. The net active power injected into the distribution network by each virtual power plant;

[0090] This represents the set of all virtual power plants.

[0091] It should be further noted that the constraints that the decision variables in the leader decision-making model need to satisfy include: dynamic topological evolution constraints. Network trend constraints Constraints on the operation of power distribution system operators' own resources Load sequential recovery constraints ;

[0092] The constraints that decision variables in the follower decision-making model need to satisfy include the virtual power plant aggregation flexibility boundary constraint. .

[0093] It should be further explained that dynamic topology evolution constraints The expression is:

[0094]

[0095] in, Indicates at time branch road The switch decision state is a binary variable, where 1 represents closed and 0 represents open.

[0096] Indicates at time branch road The physical repair status is a binary parameter, where 1 represents that the branch has been repaired and is operable, and 0 represents that it has not been repaired.

[0097] Indicates in scrolling window The maximum cumulative number of switch operations allowed within the specified range;

[0098] This represents the total number of nodes in the distribution network;

[0099] Indicates at time Representation Nodes A binary indicator variable indicating whether it is an active power supply root node, where 1 represents yes and 0 represents no;

[0100] Indicates at time Branches used to verify topological connectivity Virtual branch traffic;

[0101] Indicates at time Branches used to verify topological connectivity Virtual branch traffic;

[0102] Indicates at time Representation Nodes A binary auxiliary variable indicating whether a recovery pathway exists, where 1 represents yes and 0 represents no;

[0103] This represents the set of all branches in a power distribution network;

[0104] This represents the set of switch branches in a distribution network that have automatic operation capabilities.

[0105] This represents the set of all nodes in a power distribution network.

[0106] It represents a sufficiently large positive real number.

[0107] It should be further explained that network trend constraints The expression is:

[0108]

[0109] in, Indicates at time Flowing through the branch road The active power;

[0110] Indicates at time Flowing through the branch road reactive power;

[0111] Indicates at time The first generation of power distribution system operators The contribution of resources to recovery;

[0112] Indicates at time The first generation of power distribution system operators Recovering unused resources

[0113] Indicates at time No. The active power injected into the distribution network by a virtual power plant;

[0114] Indicates at time No. The reactive power injected into the distribution network by a virtual power plant;

[0115] Indicates at time node The actual restored active power load;

[0116] Indicates at time node The actual restoration of reactive power load;

[0117] Indicates at time node The square of the voltage amplitude;

[0118] Indicates at time node The square of the voltage amplitude;

[0119] Indicates a branch The resistance;

[0120] Indicates a branch The reactance;

[0121] This represents an auxiliary relaxation variable used to handle voltage drop relationships in disconnected topologies;

[0122] Indicates the reference voltage of the power distribution network;

[0123] Represents a node The power factor angle;

[0124] Indicates a branch The maximum active power limit that is allowed to pass;

[0125] Indicates a branch The maximum permissible reactive power limit;

[0126] This represents the minimum permissible square of the node voltage magnitude;

[0127] This represents the maximum permissible square of the node voltage magnitude;

[0128] Represents a node The set of parent nodes;

[0129] Represents a node The set of child nodes;

[0130] Indicates access node The distribution system operator's own set of restoration resources;

[0131] Indicates access node A collection of virtual power plants.

[0132] It should be further noted that the power distribution system operator's own resources are subject to operational constraints. The expression is:

[0133]

[0134] in, This indicates that the power distribution system operator owns the first The minimum allowable active power for resource recovery;

[0135] This indicates that the power distribution system operator owns the first The maximum allowable active power of the resource to be restored;

[0136] This indicates that the power distribution system operator owns the first The minimum permissible reactive power for resource recovery;

[0137] This indicates that the power distribution system operator owns the first The maximum permissible reactive power of the class-recovery resource;

[0138] This indicates that the power distribution system operator owns the first The maximum change in active power per unit time for a type of recovery resource.

[0139] It should be further explained that the load sequential recovery constraint The expression is:

[0140]

[0141] in, Indicates at time node The actual load recovers active power;

[0142] Indicates at time Representation Nodes A binary auxiliary variable indicating whether a recovery pathway exists, where 1 represents yes and 0 represents no;

[0143] Indicates at time node The system-level raw load active power demand.

[0144] It is further necessary to explain the boundary constraints of virtual power plant aggregation flexibility. The expression is:

[0145]

[0146] in, Indicates at time No. The active power injected into the distribution network by a virtual power plant;

[0147] Indicates at time No. The reactive power injected into the distribution network by a virtual power plant;

[0148] Indicates at time No. The first virtual power plant The active power output of the generator;

[0149] Indicates at time No. The first virtual power plant The reactive power output of the generator;

[0150] Indicates at time No. The first virtual power plant The actual active power output of each renewable energy source;

[0151] Indicates at time No. The first virtual power plant The actual reactive power output of a renewable energy source;

[0152] Indicates at time No. The charging power of the energy storage system inside a virtual power plant;

[0153] Indicates at time No. Discharge power of the energy storage system inside a virtual power plant;

[0154] Indicates at time No. The reactive power provided by the internal energy storage system of a virtual power plant;

[0155] Indicates at time No. The reactive power provided by the static var compensator inside the virtual power plant;

[0156] Indicates at time No. The actual restored active power of private loads within a virtual power plant;

[0157] Indicates the first The power factor angle of private loads within a virtual power plant;

[0158] Indicates at time Characterizing the first A binary variable representing the charging status of the energy storage system inside a virtual power plant, where 1 represents charging and 0 represents discharging.

[0159] Indicates at time No. The state of charge of the energy storage system within a virtual power plant;

[0160] Indicates the first Charging efficiency of the internal energy storage system of a virtual power plant;

[0161] Indicates the first Discharge efficiency of the energy storage system inside a virtual power plant;

[0162] Indicates the first The minimum allowable amount of active power that a virtual power plant can inject into the grid;

[0163] Indicates the first The maximum allowable amount of active power that a virtual power plant can inject into the grid;

[0164] Indicates the first The minimum allowable amount of reactive power that a virtual power plant can inject into the grid;

[0165] Indicates the first The maximum permissible amount of reactive power that a virtual power plant can inject into the grid;

[0166] Indicates the first Minimum allowable state of charge of the energy storage system within a virtual power plant;

[0167] Indicates the first The maximum permissible state of charge of the energy storage system within a virtual power plant;

[0168] Indicates the first The maximum allowable charging power of the internal energy storage system of a virtual power plant;

[0169] Indicates the first The maximum allowable discharge power of the internal energy storage system of a virtual power plant;

[0170] Indicates the first The internal energy storage system of each virtual power plant provides a minimum allowable value for reactive power;

[0171] Indicates the first The internal energy storage system of each virtual power plant provides the maximum permissible value of reactive power;

[0172] Indicates the first The first virtual power plant The minimum permissible active power output of the generator;

[0173] Indicates the first The first virtual power plant The maximum permissible active power output of the generator;

[0174] Indicates the first The first virtual power plant The minimum permissible reactive power output of the generator;

[0175] Indicates the first The first virtual power plant The maximum permissible reactive power output of the generator;

[0176] Indicates the first The first virtual power plant The rate at which the active power output of the generator climbs downhill is limited;

[0177] Indicates the first The first virtual power plant The rate at which the active power output of the generator climbs an incline is limited.

[0178] Indicates at time No. The first virtual power plant Real-time forecast of maximum active power output of renewable energy sources;

[0179] Indicates the first The first virtual power plant The minimum permissible reactive power output of a renewable energy source;

[0180] Indicates the first The first virtual power plant The maximum permissible reactive power output of each renewable energy source;

[0181] Indicates the first A collection of generator sets within a virtual power plant;

[0182] Indicates the first A collection of renewable energy sources within a virtual power plant;

[0183] Indicates at time No. The active power demand of private loads within a virtual power plant.

[0184] It should be further noted that the system-level load requirements involved in the leader decision-making model Branch road repair status And the maximum output of renewable energy involved in the follower decision-making model. and internal load demand The range of its uncertainty values ​​is determined by an uncertainty quantification system with statistical coverage guarantees; this uncertainty quantification system includes:

[0185] Baseline Prediction Layer: Used to extract multi-dimensional time-series features of the post-disaster distribution network, and outputs the predicted baseline values ​​of key parameters at each time point within the scrolling window through a data-driven model. ;

[0186] Boundary Representation Layer: Based on real-time error feedback and dynamic calibration mechanisms, it generates distribution-independent confidence boundaries for the predicted baseline values, forming a boundary with a probability of covering the true values ​​of no less than [value missing]. Continuous confidence intervals With discrete confidence sets Continuous confidence interval Including the first continuous confidence interval Second continuous confidence interval Third continuous confidence interval ;

[0187] The relevant uncertainty parameters in the leader decision-making model and the follower decision-making model are constrained within their corresponding confidence intervals or sets, i.e.:

[0188]

[0189] Among them, discrete confidence set This indicates the repair status of the distribution network branch within the rolling time window. The set of possible values;

[0190] First continuous confidence interval This indicates the system-level load demand within the rolling time window. The range of forecast uncertainty;

[0191] Second continuous confidence interval This represents the real-time predicted maximum active power output of renewable energy sources (such as solar and wind power) within the virtual power plant during the rolling time window. The range of uncertainty;

[0192] Third continuous confidence interval This represents the private load demand within the virtual power plant during the rolling time window. The range of prediction uncertainty.

[0193] It should be further noted that the baseline prediction layer uses a long short-term memory network as the prediction model. Through historical multidimensional time series data Predicting the baseline values ​​of source-load power at various times within the scrolling window. and the probability that the branch is in a connected state. .

[0194] It should be further noted that the formula for the dynamic update and calibration of the confidence boundary of the boundary representation layer is as follows:

[0195]

[0196] in, This indicates the corresponding continuous variable at the decision time. The deviation score;

[0197] This indicates the corresponding continuous variable at the decision time. Actual observed values

[0198] This indicates the corresponding continuous variable at the decision time. The predicted baseline value;

[0199] Indicates the corresponding continuous variable at time t. The predicted baseline value;

[0200] express Calibration threshold for continuous variables at any given time;

[0201] This indicates the corresponding discrete variable at the decision time. The actual observation status;

[0202] This indicates the corresponding discrete variable at the decision time. Predicted state and actual observed state The same probability;

[0203] Indicates the corresponding discrete variable at time t. The predicted probability distribution, including 、 ;

[0204] Indicates the corresponding discrete variable in The score for deviation at a given time;

[0205] express Calibration threshold for discrete variables at time points;

[0206] This is an out-of-bounds indicator function; if the current deviation score... Exceeded the current threshold If the value is 1, the indicator variable takes the value 1; otherwise, it takes the value 0.

[0207] The preset adaptive adjustment step size;

[0208] The preset error tolerance level, .

[0209] It should be further noted that the steps to reconstruct the single-leader-multiple-follower model into a post-disaster recovery single-level mixed-integer linear programming model include:

[0210] For each follower decision model, write out its Lagrange function. And derive the KKT condition equations that satisfy the first-order optimality condition;

[0211] By introducing the KKT condition equations as additional constraints into the constraint set of the leader decision-making model, the two-level decision-making model is transformed into a single-level mathematical programming problem with equilibrium constraints.

[0212] By introducing auxiliary binary variables and using the Big M method, the complementary relaxation conditions in the KKT conditions are transformed into a series of linear inequality constraints.

[0213] Using the McCormick envelope method, the incentive electricity price variables appearing in the leader's objective function and potential function in the transformed model are analyzed. With response power variables The bilinear product terms are linearized to obtain a single-level mixed-integer linear programming model that can be processed by the standard solver.

[0214] It should be further explained that the Lagrange function The KKT condition equations are as follows:

[0215]

[0216] in, This represents the Lagrangian function corresponding to the follower decision model;

[0217] Represents the vector of lower-level decision variables;

[0218] Representing equality constraints Lagrange multiplier vectors;

[0219] Inequality constraints The Lagrange multiplier vector.

[0220] It should be further explained that the specific form of linearizing the complementary relaxation condition using the Big M method is as follows:

[0221] For the follower decision model, the first Inequality constraints and its corresponding non-negative Lagrange multipliers Introducing auxiliary binary variables And add the following system of linear inequalities:

[0222]

[0223] in, It is a sufficiently large positive real number;

[0224] when At that time, constraints Corresponding It can take any positive value;

[0225] when hour, ,correspond .

[0226] It should be further noted that the specific form of linearizing the bilinear term using the McCormick envelope method is as follows:

[0227] Define auxiliary variables Using auxiliary variables The product term in the original objective function of the alternative leader decision-maker model ;

[0228] Based on variables and feasible domain boundary and Construct linear inequality constraints:

[0229]

[0230] This linear inequality constraint forms a compact convex hull in multidimensional space, confining the nonlinear bilinear surface within a rectangular envelope region.

[0231] It should be further noted that simulation datasets are used. For decision generation networks Physical feasibility assessment network The steps for collaborative training include:

[0232] Do not initialize the decision generation network and physical feasibility discrimination network Weight parameters;

[0233] fixed The parameters, using the dataset Samples in And additional randomly generated samples of physically infeasible decisions, used for supervised learning training. To enable them to distinguish between decision-making In a given scenario Does the following satisfy all physical constraints?

[0234] fixed The parameters are set to minimize decision-making costs and drive the output to meet physical feasibility, using a dataset. train ;

[0235] Using the training done in the current round ,Evaluate For training set samples Output decision Feasibility confidence level; select sample features whose confidence level is within a preset critical interval as difficult sample features. Return the features of the hard samples to the simulation dataset. The construction process involves solving for the corresponding optimal decision to generate new data pairs. and add to the dataset ;

[0236] The network was retrained using the expanded dataset for iterative optimization until... Decision performance and The discrimination accuracy of all converged.

[0237] It should be further explained that the training of the physical feasibility discrimination network... The loss function used The binary cross-entropy loss is expressed as follows:

[0238]

[0239] in, For network parameters;

[0240] As the truth label, when all physical constraints in the single-level mixed-integer linear programming model for disaster recovery are satisfied. ,otherwise .

[0241] It should be further explained that training the decision generation network The composite loss function used The expression is:

[0242]

[0243] in, The total linearization cost of a single-level mixed-integer linear programming model for post-disaster recovery;

[0244] This indicates the preset penalty coefficient.

[0245] It should be further noted that the pre-defined feasibility confidence threshold interval for the active learning mechanism to screen difficult sample features is [0.4, 0.6], that is:

[0246] when At that time, the corresponding scene features Features identified as difficult samples .

[0247] Secondly, this application provides a rolling clearing scheme generation system for post-disaster distribution network restoration, used to implement the above-mentioned rolling clearing scheme generation method, including:

[0248] The data acquisition module is used to acquire real-time data of the distribution network, including the physical topology status of the distribution network, the branch repair status, the system-level load demand, the load survival value factor, the renewable energy output level inside the virtual power plant, and the private load demand.

[0249] The preliminary clearing decision generation module is used to construct the current decision time based on the real-time data of the distribution network at the current decision time. Context feature vector ,Will Input pre-trained decision generation network Perform forward inference and output the timing scrolling window. Preliminary clearing decision within ;

[0250] The final clearing decision generation module is used to generate a pre-trained physical feasibility discrimination network. Preliminary clearing decision A security check is performed, and the decision confidence score is output. If the decision confidence score is lower than a preset confidence threshold, the initial clearing decision is corrected through linear projection to obtain the final clearing decision. If the decision confidence level is not lower than the preset confidence level threshold, then the preliminary clearing decision will be made directly. As the final clearing decision ;

[0251] The decision execution module is used to implement the final clearing decision. Release timeline scrolling window The internal incentive price instructions and load restoration control commands are only executed. The operation of the first step size.

[0252] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described rolling clearing scheme generation method.

[0253] Fourthly, this application provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described rolling clearing scheme generation method.

[0254] As can be seen from the above technical solutions, this application has the following advantages:

[0255] 1. This application solves the problems of existing open-loop decision-making models being unable to make closed-loop adjustments, master-slave game models having limited guiding effects and ignoring multi-agent game theory, and slow solution of complex models by executing a complete process including acquiring real-time data, outputting preliminary clearing decisions based on decision generation networks, verifying and correcting them using physical feasibility discrimination networks, and rolling release. It realizes rolling closed-loop optimization, safe and reliable guidance and rapid generation of post-disaster distribution network restoration decisions, significantly improving the dynamic adaptability, coordination and execution efficiency of the restoration process.

[0256] 2. This application solves the problem of decision delay caused by the complexity of the model and the large amount of computation in traditional optimization methods by using a pre-trained decision generation network for forward inference to directly output the preliminary clearing decision. It transforms the complex online solution process into offline network training and online fast forward propagation, which greatly improves the generation speed of the clearing scheme and can meet the real-time response requirements in extreme scenarios.

[0257] 3. This application solves the problem that traditional open-loop decision-making is difficult to use real-time status feedback to correct subsequent plans by introducing a time-series rolling window and a closed-loop mechanism that executes only the first step and rolls. Through the "partial execution, global rolling" approach, the market clearing decision can be adjusted in a closed loop according to the dynamic evolution of the power grid topology and repair status, thereby enhancing the adaptability of the recovery plan to the uncertain post-disaster environment.

[0258] 4. This application introduces a pre-trained physical feasibility discrimination network to verify the security of the initial decision and initiates a correction mechanism based on linear projection when the confidence level is insufficient. This solves the security problem that complex physical constraints may be overlooked during the rapid decision-making process. While ensuring the speed of decision-making, it adds a reliable physical security closed-loop verification and correction guarantee, which significantly improves the physical feasibility and execution reliability of the clearing scheme.

[0259] 5. By using a collaborative training method based on a simulation dataset to obtain the decision generation network and the physical feasibility discrimination network, the problem of lacking high-quality, physically consistent training samples in data-driven methods is solved. Training labels are generated by solving a single-layer mixed integer linear programming model for post-disaster recovery, ensuring the optimality and physical feasibility of the training data itself, thus laying a solid foundation for the reliable performance of the neural network. Attached Figure Description

[0260] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0261] Figure 1 This is a flowchart of a method for generating a rolling clearing scheme for post-disaster power distribution network recovery in one embodiment of this application.

[0262] Figure 2 This is a schematic block diagram of a rolling clearing scheme generation system for post-disaster power distribution network recovery in one embodiment of this application.

[0263] Figure 3 This is a schematic diagram of the hardware structure of an electronic device in one embodiment of this application. Detailed Implementation

[0264] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0265] The following describes in detail the rolling clearing scheme generation method involved in this application. Specific details, such as particular system structures and technologies, are presented for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details.

[0266] In the rolling clearing scheme generation method involved in this application, the term "comprising" indicates the presence of the described feature, integral, step, operation, element, and / or component, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0267] The terms "one embodiment" or "some embodiments" used in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this application do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized.

[0268] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0269] The rolling clearing scheme generation method provided in this application embodiment is executed by a computer device. Correspondingly, the rolling clearing scheme generation system for post-disaster power distribution network recovery runs in the computer device.

[0270] Figure 1 This is a flowchart illustrating a method for generating a rolling clearing scheme for post-disaster distribution network recovery, according to an embodiment of this application. Figure 1 The implementing entity can be a rolling clearing scheme generation system. Depending on different requirements, the order of steps in this flowchart can be changed, and some steps can be omitted.

[0271] like Figure 1 As shown, the method for generating a rolling clearing scheme for post-disaster power distribution network restoration includes:

[0272] S1. Obtain real-time data of the distribution network, including the physical topology status of the distribution network, branch repair status, system-level load demand, load survival value factor, renewable energy output level within the virtual power plant, and private load demand;

[0273] By acquiring comprehensive data in real time, including physical topology status, branch repair status, load survival value factor, and renewable energy output level, an accurate and dynamic information input basis is provided for generating clearing decisions that are in line with the current actual situation of the power grid.

[0274] In some specific embodiments, the physical topology of the distribution network This includes the real-time on / off status of each line in each distribution network and the location information of operable switches;

[0275] Branch road repair status Specifically, the repair status binary parameters for each branch include the nodes. With nodes Branch roads between The repair status binary parameter is represented as Its value of 1 indicates a branch At any moment The system has been repaired to an operational state; a value of 0 indicates a branch. Not repaired;

[0276] Survival value factor Specifically, each node is assigned a weight coefficient based on its load type and preset recovery priority. At any moment The weighting coefficient is ;

[0277] Renewable energy output level For renewable energy power generation equipment at any time The active power output value; renewable energy power generation equipment includes at least one of photovoltaic power generation equipment and wind power equipment.

[0278] By clearly defining the binary parameters of physical topology state, branch repair state, load survival value factor, and renewable energy output level, the model can accurately characterize the real-time structure and resource status of the power grid after a disaster, laying a clear parameterized foundation for accurate decision-making.

[0279] S2. Construct the current decision-making time based on the real-time data of the distribution network at the current decision-making time. Context feature vector ,Will Input pre-trained decision generation network Perform forward inference and output the timing scrolling window. Preliminary clearing decision within Preliminary clearing decision A decision vector whose elements include a time-series scrolling window. The incentive electricity price at each time step, the load recovery amount at each node, and the switching operation instructions for each branch;

[0280] Time-series scrolling window Defined as:

[0281]

[0282] in For time step, This represents the total duration of the time-series scrolling window.

[0283] By using a pre-trained decision generation network to perform forward reasoning on the integrated context feature vector, a preliminary clearing decision containing incentive electricity price, load recovery amount and switching command can be directly and quickly output, realizing a millisecond-level mapping from complex scenarios to solution generation.

[0284] In some specific embodiments, context feature vectors Represented as:

[0285]

[0286] in, Represents the physical topology state;

[0287] Indicates the repair status of the branch road;

[0288] This indicates system-level load demand;

[0289] Indicates the survival value factor;

[0290] Indicates the level of renewable energy output;

[0291] This indicates the private load demand within the virtual power plant.

[0292] In some specific embodiments, the decision time step The value range is 1-15 min;

[0293] Total duration of the time-tracking scrolling window The value range is 15-60 min;

[0294] .

[0295] By limiting the reasonable range and proportion of decision-making time step and total duration of the scrolling window, the real-time requirements of decision-making and the need for forward-looking optimization are balanced, enabling the scrolling mechanism to respond quickly and perform collaborative optimization with a limited step size.

[0296] S3. Using a pre-trained physical feasibility discrimination network Preliminary clearing decision Perform security checks and output decision confidence levels;

[0297] If the decision confidence level is lower than the preset confidence level threshold, the preliminary clearing decision is corrected through linear projection to obtain the final clearing decision. ;

[0298] If the decision confidence level is not lower than the preset confidence level threshold, then the preliminary clearing decision will be made directly. As the final clearing decision .

[0299] By introducing a physical feasibility discrimination network to verify the security of the initial decision and triggering linear projection correction based on the confidence threshold, a reliable physical constraint compliance guarantee is embedded in the rapid clearing process, ensuring the feasibility and security of the final decision.

[0300] In some specific embodiments, the confidence threshold ranges from 0.5 to 0.9.

[0301] In some specific embodiments, the expression for modifying the initial clearing decision through linear projection is as follows:

[0302]

[0303] in, express The constraint matrix, where each row corresponds to a physical constraint and each column corresponds to... One of the elements; express The constraint vector is a column vector, where each element represents... The boundary values ​​of the parallel constraints are obtained by mathematical transformation from the original boundaries of the physical constraints. These include: for upper limit constraints, the upper limit value is taken directly; for lower limit constraints, the negative value of the lower limit is taken; for bilateral constraints, they are decomposed into upper limit constraints and lower limit constraints and then the values ​​are taken according to the corresponding rules.

[0304] By employing a linear projection model aimed at minimizing deviation for decision correction, the optimization intent of the initial decision can be preserved to the greatest extent possible while ensuring that all physical constraints derived from real-time features are met.

[0305] S4. Based on the final clearing decision Release timeline scrolling window The internal incentive price instructions and load restoration control commands are only executed. The first step of the operation will be executed, and the current decision time will be set after execution. Scroll to the next time step, then return to step S1 and repeat.

[0306] Among them, decision generation network Physical feasibility assessment network All are deep learning neural networks, using simulation datasets. The simulation dataset was obtained after co-training. The construction steps are as follows:

[0307] Simulate different post-disaster power grid scenarios and construct a set of context feature vectors. Each context feature vector It includes the physical topology status of the distribution network, branch repair status, system-level load demand, load survival value factor, renewable energy output level within the virtual power plant, and private load demand;

[0308] For each context feature vector Solve the single-level mixed-integer linear programming model for disaster recovery to obtain its corresponding optimal decision vector. The post-disaster recovery single-level mixed integer linear programming model is reconstructed from a pre-built single-leader-multiple-follower model. During the reconstruction process, the McCormick envelope method is used to handle bilinear terms.

[0309] Record all feature-decision pairs To form a simulation dataset .

[0310] By executing the first step instruction within the scrolling window and rolling the decision time to the next time step in a loop mechanism, the clearing decision can be dynamically adjusted using the latest physical state feedback, thus achieving continuous adaptation of the recovery scheme to the post-disaster power grid evolution.

[0311] In some specific embodiments, the single-leader-multiple-follower model is a decision optimization model with a two-layer interactive architecture, including a leader decision model and multiple follower decision models that are independently connected to the leader decision model. A horizontal equilibrium mechanism is established between the various follower decision models. The leader decision model corresponds to the power distribution system operator, and each follower decision model corresponds to a virtual power plant.

[0312] Among them, the decision objective of the leader decision-making model is in the scrolling window. Internally, we will coordinate and optimize the progress of the entire network load recovery, the cost of dispatching our own resources, and the incentive costs of purchasing electricity from virtual power plants.

[0313] The decision-making objective of the follower decision-making model is to maximize individual profits. After receiving the incentive electricity price signal issued by the leader, it optimizes the output and load recovery of its internal distributed energy resources to determine the net power injected into the distribution network.

[0314] The horizontal equilibrium mechanism constructs a global situation function to address the decision-making coupling and interest game among multiple followers due to the shared physical power grid. This transforms the Nash equilibrium problem of decentralized decision-making among multiple followers into a single-objective optimization problem of this global situation function.

[0315] By constructing a single-leader-multiple-follower model with a two-layer interactive architecture and embedded horizontal equilibrium mechanism, the vertical incentive relationship between operators and multiple virtual power plants and the horizontal game coupling between virtual power plants were simulated, thus restoring the market structure foundation for global optimization.

[0316] In some specific embodiments, the objective function and constraints of the leader decision-making model include:

[0317]

[0318]

[0319] in, Indicates time No. The incentive electricity price for each virtual power plant is a decision variable for leaders; An index for virtual power plants. For time indexing; Indicates at time node The actual restored active power is a decision variable for leaders. For node indexing, For time indexing; Indicates at time The first generation of power distribution system operators The contribution of resources to resource recovery is a decision variable for leaders. For indexing proprietary resources, For time indexing; Indicates at time branch road The switch state binary decision variable takes a value of 1 to indicate closed and 0 to indicate open. Indicates at time node The load survival value factor is a non-negative weighting coefficient preset according to load type and recovery priority; Indicates at time node System-level load requirements; This indicates that the power distribution system operator owns the first Functions for the operation, fuel, or maintenance costs of resource recovery. Indicates at time No. The net active power injected into the distribution network by each virtual power plant; , These represent the weighting coefficients of the distribution system operator's cost of adjusting its own resources and its expenditure on purchasing electricity from virtual power plants in the objective function, respectively. This represents the set of all nodes in a power distribution network. This represents the set of all virtual power plants; This represents the collection of all proprietary restoration resources owned by the power distribution system operator. Represents dynamic topology evolution constraints; This indicates network trend constraints; This indicates the constraints on the operation of the power distribution system operator's own resources; This indicates a load sequential recovery constraint; Indicates the repair status of the branch road Discrete confidence sets; Indicates system-level load demand Continuous confidence intervals; Indicates system-level load demand The baseline value for prediction at any given time; , This indicates the minimum and maximum allowable values ​​for the incentive electricity price; Indicates the first The profit function of a virtual power plant.

[0320] By establishing a leader decision-making model and its constraint system aimed at maximizing weighted recovery volume and balancing the cost of proprietary resources with electricity purchase expenditures, the overall optimization criteria and risk management boundaries for power distribution system operators in post-disaster recovery are clarified.

[0321] In some specific embodiments, the objective function and constraints of the follower decision model include:

[0322]

[0323]

[0324] in, Indicates the first The profit function of a virtual power plant; Indicates at time No. The net active power injected into the distribution network by a virtual power plant is a decision variable for followers; Indicates at time No. The first virtual power plant The output value of energy resources is a decision variable for followers. Index for resource types; Indicates time No. Incentive electricity prices for virtual power plants; Indicates the first The first virtual power plant Operating cost function for energy resources;

[0325] Indicates the first The collection of all energy resources within a virtual power plant; This represents the boundary constraints for the aggregation flexibility of virtual power plants; express Time of the first Renewable energy within a virtual power plant Real-time prediction of maximum active power output Continuous confidence intervals; express Time of the first Private load demand within a virtual power plant Continuous confidence intervals.

[0326] By establishing a follower decision-making model with the goal of maximizing individual profits, we accurately characterize the selfish optimization behavior of virtual power plants after receiving incentive electricity price signals, providing an accurate proxy model for analyzing market-guided resource responses.

[0327] In some specific embodiments, the global situation function is:

[0328]

[0329] in, Denotes the global state function, where It is all virtual power plants at all times The output vector is composed of the output forces. It is all virtual power plants at all times The incentive price vector is composed of incentive prices; Indicates the first The profit function of a virtual power plant; Indicates at time , used to characterize the The output change of the first virtual power plant affects the second... Sensitivity coefficient for the degree of voltage impact at the grid connection point of a virtual power plant; Indicates at time No. The net active power injected into the distribution network by each virtual power plant; Indicates at time No. The net active power injected into the distribution network by each virtual power plant; This represents the set of all virtual power plants.

[0330] By constructing a global potential function that integrates the profit functions of all followers and deducts the coupling effects, the problem of solving the decentralized Nash equilibrium is transformed into a single-objective optimization of this potential function, which theoretically guarantees the existence, uniqueness, and solvability of the equilibrium in a multi-agent game.

[0331] In some specific embodiments, the constraints that the decision variables in the leader decision-making model need to satisfy include: dynamic topological evolution constraints. Network trend constraints Constraints on the operation of power distribution system operators' own resources Load sequential recovery constraints ;

[0332] The constraints that decision variables in the follower decision-making model need to satisfy include the virtual power plant aggregation flexibility boundary constraint. .

[0333] By clarifying that the decision variables of the leader decision-making model need to satisfy the constraints of dynamic topology evolution, network flow, operation of self-owned resources, and sequential load recovery, and that the decision variables of the follower decision-making model need to satisfy the constraints of virtual power plant aggregation flexibility, a complete and rigorous physical and operational boundary is established for the entire decision optimization problem.

[0334] In some specific embodiments, dynamic topology evolution constraints The expression is:

[0335]

[0336] in, Indicates at time branch road The switch decision state is a binary variable, where 1 represents closed and 0 represents open. Indicates at time branch road The physical repair status is a binary parameter, where 1 represents that the branch has been repaired and is operable, and 0 represents that it has not been repaired. Indicates in scrolling window The maximum cumulative number of switch operations allowed within the specified range; This represents the total number of nodes in the distribution network; Indicates at time Representation Nodes A binary indicator variable indicating whether it is an active power supply root node, where 1 represents yes and 0 represents no; Indicates at time Branches used to verify topological connectivity Virtual branch traffic; Indicates at time Branches used to verify topological connectivity Virtual branch traffic; Indicates at time Representation Nodes A binary auxiliary variable indicating whether a recovery pathway exists, where 1 represents yes and 0 represents no; This represents the set of all branches in a power distribution network; This represents the set of switch branches in a distribution network that have automatic operation capabilities. This represents the set of all nodes in a power distribution network. It represents a sufficiently large positive real number.

[0337] By establishing a dynamic topology evolution model that includes constraints such as repair status correlation, switching operation count limit, and virtual traffic balancing, it is ensured that the generated switching operation sequence conforms to the actual physical connectivity and operational feasibility after the disaster.

[0338] In some specific embodiments, network flow constraints The expression is:

[0339]

[0340] in, Indicates at time Flowing through the branch road The active power; Indicates at time Flowing through the branch road reactive power; Indicates at time The first generation of power distribution system operators The contribution of resources to recovery; Indicates at time The first generation of power distribution system operators The unproductive output of class recovery resources; Indicates at time No. The active power injected into the distribution network by a virtual power plant; Indicates at time No. The reactive power injected into the distribution network by a virtual power plant; Indicates at time node The actual restored active power load; Indicates at time node The actual restoration of reactive power load; Indicates at time node The square of the voltage amplitude; Indicates at time node The square of the voltage amplitude; Indicates a branch The resistance; Indicates a branch The reactance; This represents an auxiliary relaxation variable used to handle voltage drop relationships in disconnected topologies; Indicates the reference voltage of the power distribution network; Represents a node The power factor angle; , Indicates a branch The maximum active power limit and the maximum reactive power limit that are allowed to pass; , Represents the minimum and maximum allowable squares of the node voltage magnitude; Represents a node The set of parent nodes; Represents a node The set of child nodes; Indicates access node The distribution system operator's own set of restoration resources; Indicates access node A collection of virtual power plants.

[0341] By establishing a linearized DistFlow power flow constraint set based on branch state variable correction, the power balance and voltage security laws of distribution networks under different topology connections are accurately described, which is the core of ensuring the electrical feasibility of clearing decisions.

[0342] In some specific embodiments, the operating constraints of the power distribution system operator's own resources The expression is:

[0343]

[0344] in, , This indicates the first generation of power distribution system operators. The minimum and maximum allowable active power of the resource to be restored; , This indicates the first generation of power distribution system operators. The minimum and maximum allowable reactive power of the restored resource; This indicates the first generation of power distribution system operators. The maximum change in active power per unit time for a type of recovery resource.

[0345] By establishing operational constraints, including upper and lower limits of output and ramp rate, for the power distribution system operator's own resources, the actual adjustment capacity and physical limitations of such controllable resources during the post-disaster recovery process were accurately modeled.

[0346] In some specific embodiments, load sequential recovery constraints The expression is:

[0347]

[0348] in, Indicates at time node The actual load recovers active power; Indicates at time Representation Nodes A binary auxiliary variable indicating whether a recovery pathway exists, where 1 represents yes and 0 represents no; Indicates at time node The system-level raw load active power demand.

[0349] By establishing a sequential recovery constraint that ensures the load recovery amount is not reduced and is limited by the recovery path and the original demand, the restored load is prevented from going out of power again during the rolling decision-making process, thus ensuring the fairness and continuity of the recovery process.

[0350] In some specific embodiments, virtual power plant aggregation flexibility boundary constraints The expression is:

[0351]

[0352] in, , Indicates at time No. The active and reactive power injected into the distribution network by a virtual power plant; , Indicates at time No. The first virtual power plant The active and reactive power outputs of the generators; , Indicates at time No. The first virtual power plant The actual active power output and actual reactive power output of each renewable energy source; , Indicates at time No. The charging and discharging power of the energy storage system inside a virtual power plant; , Indicates at time No. The reactive power provided by the internal energy storage system and static var compensator of the virtual power plant; Indicates at time No. The actual restored active power of private loads within a virtual power plant; Indicates the first The power factor angle of private loads within a virtual power plant; Indicates at time Characterizing the first A binary variable representing the charging status of the energy storage system inside a virtual power plant, where 1 represents charging and 0 represents discharging. Indicates at time No. The state of charge of the energy storage system within a virtual power plant; , Indicates the first The charging and discharging efficiency of the energy storage system inside a virtual power plant. , Indicates the first The minimum and maximum allowable values ​​of active power injected into the grid by a virtual power plant; , Indicates the first The minimum and maximum allowable values ​​for reactive power injected into the grid by a virtual power plant; , Indicates the first The minimum and maximum allowable state of charge (SOC) values ​​for the energy storage system within a virtual power plant; , Indicates the first The maximum allowable charging power and maximum allowable discharging power of the energy storage system within a virtual power plant; , Indicates the first The internal energy storage system of each virtual power plant provides the minimum and maximum allowable values ​​for reactive power; , Indicates the first The first virtual power plant The minimum and maximum permissible active power output of the generator; , Indicates the first The first virtual power plant The minimum and maximum permissible reactive power output of the generator; , Indicates the first The first virtual power plant The active power output downhill ramp rate limit and the active power output uphill ramp rate limit of the generator. Indicates at time No. The first virtual power plant Real-time forecast of maximum active power output of each renewable energy source; , Indicates the first The first virtual power plant The minimum and maximum permissible reactive power output of each renewable energy source; Indicates the first A collection of generator sets within a virtual power plant; Indicates the first A collection of renewable energy sources within a virtual power plant; Indicates at time No. The active power demand of private loads within a virtual power plant.

[0353] By modeling in detail the aggregated operational constraints of various resources (generators, renewable energy, energy storage, and loads) within the virtual power plant, the physical boundary of its overall ability to provide flexible regulation to the power grid is accurately characterized.

[0354] In some specific embodiments, the system-level load requirements involved in the leader decision-making model Branch road repair status And the maximum output of renewable energy involved in the follower decision-making model. and internal load demand The range of its uncertainty values ​​is determined by an uncertainty quantification system with statistical coverage guarantees; this uncertainty quantification system includes:

[0355] Benchmark Prediction Layer: Used to extract multi-dimensional time-series features of the post-disaster distribution network, and outputs the predicted benchmark values ​​of key parameters at each time point within the scrolling window through a data-driven model. ;

[0356] Boundary Representation Layer: Based on real-time error feedback and dynamic calibration mechanisms, it generates distribution-independent confidence boundaries for the predicted baseline values, forming a boundary with a probability of covering the true values ​​of no less than [value missing]. Continuous confidence intervals With discrete confidence sets Continuous confidence interval Including the first continuous confidence interval Second continuous confidence interval Third continuous confidence interval ;

[0357] The relevant uncertainty parameters in the leader decision-making model and the follower decision-making model are constrained within their corresponding confidence intervals or sets, i.e.:

[0358]

[0359] Among them, discrete confidence set This indicates the repair status of the distribution network branch within the rolling time window. The set of possible values; the first continuous confidence interval This indicates the system-level load demand within the rolling time window. The range of prediction uncertainty; the second continuous confidence interval This represents the real-time predicted maximum active power output of renewable energy sources (such as solar and wind power) within the virtual power plant during the rolling time window. Uncertainty range; third continuous confidence interval This represents the private load demand within the virtual power plant during the rolling time window. The range of prediction uncertainty.

[0360] By constructing an uncertainty quantification system with statistical coverage guarantees, distribution-independent confidence intervals and sets are generated for key uncertain parameters, transforming the stochastic optimization problem into a deterministic robust optimization framework and reducing the sensitivity of decision-making to prediction errors.

[0361] In some specific embodiments, the baseline prediction layer uses a long short-term memory network as the prediction model. Through historical multidimensional time series data Predicting the baseline values ​​of source-load power at various times within the scrolling window. and the probability that the branch is in a connected state. .

[0362] By employing long short-term memory networks as the benchmark prediction model, the multidimensional temporal evolution of source-load power and topological state after a disaster can be effectively captured, providing a reliable prediction benchmark center for uncertainty quantification.

[0363] In some specific embodiments, the formula for dynamic updating and calibration of the confidence boundary of the boundary representation layer is as follows:

[0364]

[0365] in, This indicates the corresponding continuous variable at the decision time. The deviation score; This indicates the corresponding continuous variable at the decision time. The actual observed value; This indicates the corresponding continuous variable at the decision time. The predicted baseline value; Indicates the corresponding continuous variable at time t. The predicted baseline value; express Calibration threshold for continuous variables at any given time; This indicates the corresponding discrete variable at the decision time. The actual observation status; This indicates the corresponding discrete variable at the decision time. Predicted state and actual observed state The same probability; Indicates the corresponding discrete variable at time t. The predicted probability distribution, including 、 ; Indicates the corresponding discrete variable in The score for deviation at a given time; express Calibration threshold for discrete variables at time points; This is an out-of-bounds indicator function; if the current deviation score... Exceeded the current threshold If the value is 1, the indicator variable takes the value 1; otherwise, it takes the value 0. The preset adaptive adjustment step size; The preset error tolerance level, .

[0366] By employing a dynamic calibration mechanism based on real-time bias scores and adaptive threshold updates, the generated confidence boundary can be automatically adjusted in width according to the latest observation error, thus achieving online tracking and coverage guarantee of time-varying uncertainties.

[0367] In some specific embodiments, the steps of reconstructing a single-leader-multiple-follower model into a post-disaster recovery single-level mixed-integer linear programming model include:

[0368] For each follower decision model, write out its Lagrange function. And derive the KKT condition equations that satisfy the first-order optimality condition;

[0369] By introducing the KKT condition equations as additional constraints into the constraint set of the leader decision-making model, the two-level decision-making model is transformed into a single-level mathematical programming problem with equilibrium constraints.

[0370] By introducing auxiliary binary variables and using the Big M method, the complementary relaxation conditions in the KKT conditions are transformed into a series of linear inequality constraints.

[0371] Using the McCormick envelope method, the incentive electricity price variables appearing in the leader's objective function and potential function in the transformed model are analyzed. With response power variables The bilinear product terms are linearized to obtain a single-level mixed-integer linear programming model that can be processed by the standard solver.

[0372] By introducing the KKT conditions of the lower-level follower problem into the upper level and using the Big M method and McCormick envelope for linearization, the complex two-level game model is equivalently reconstructed into a single-level mixed-integer linear programming model that can be directly processed by commercial solvers.

[0373] In some specific embodiments, the Lagrange function The KKT condition equations are as follows:

[0374]

[0375] in, This represents the Lagrangian function corresponding to the follower decision model; Represents the vector of lower-level decision variables; Representing equality constraints Lagrange multiplier vectors; Inequality constraints The Lagrange multiplier vector.

[0376] By writing out the Lagrangian function of the follower decision model and listing the complete KKT condition equations, a rigorous and complete set of mathematical rules is provided for transforming a two-level model into a single-level programming problem.

[0377] In some specific embodiments, the linearization of complementary relaxation conditions using the Big M method takes the following form:

[0378] For the follower decision model, the first Inequality constraints and its corresponding non-negative Lagrange multipliers Introducing auxiliary binary variables And add the following system of linear inequalities:

[0379]

[0380] in, For a sufficiently large positive real number; when At that time, constraints Corresponding It can take any positive value; when hour, ,correspond .

[0381] By introducing auxiliary binary variables for each complementary relaxation condition and transforming them into a system of linear inequalities using the Big M method, the nonlinear complementary constraints in the model were successfully eliminated, which is a key step in achieving model linearization.

[0382] In some specific embodiments, the specific form of linearizing the bilinear term using the McCormick envelope method is as follows:

[0383] Define auxiliary variables Using auxiliary variables The product term in the original objective function of the alternative leader decision-maker model ;

[0384] Based on variables and feasible domain boundary and Construct linear inequality constraints:

[0385]

[0386] This linear inequality constraint forms a compact convex hull in multidimensional space, confining the nonlinear bilinear surface within a rectangular envelope region.

[0387] By utilizing the McCormick envelope method to construct a compact linear convex hull approximation for the bilinear product terms, the nonlinear terms in the objective function are transformed into a linear form with controllable accuracy loss, significantly reducing the difficulty of solving the model.

[0388] In some specific embodiments, simulation datasets are used. For decision generation networks Physical feasibility assessment network The steps for collaborative training include:

[0389] Do not initialize the decision generation network and physical feasibility discrimination network Weight parameters;

[0390] fixed The parameters, using the dataset Samples in And additional randomly generated samples of physically infeasible decisions, used for supervised learning training. To enable them to distinguish between decision-making In a given scenario Does the following satisfy all physical constraints?

[0391] fixed The parameters are set to minimize decision-making costs and drive the output to meet physical feasibility, using a dataset. train ;

[0392] Using the training done in the current round ,Evaluate For training set samples Output decision Feasibility confidence level; select sample features whose confidence level is within a preset critical interval as difficult sample features. Return the features of the hard samples to the simulation dataset. The construction process involves solving for the corresponding optimal decision to generate new data pairs. and add to the dataset ;

[0393] The network was retrained using the expanded dataset for iterative optimization until... Decision performance and The discrimination accuracy of all converged.

[0394] By employing an iterative collaborative training process that includes a fixed discriminant network, an optimized generator network, screening of difficult samples, and retraining with an expanded dataset, the dual neural networks continuously improve the quality of generated decisions and the accuracy of feasibility judgment through interaction.

[0395] In some specific embodiments, a physical feasibility discrimination network is trained. The loss function used The binary cross-entropy loss is expressed as follows:

[0396]

[0397] in, For network parameters; As the truth label, when all physical constraints in the single-level mixed-integer linear programming model for disaster recovery are satisfied. ,otherwise .

[0398] By training a physical feasibility discrimination network using a binary cross-entropy loss function, it can accurately learn to distinguish whether a given decision satisfies all physical constraints in a specific scenario, thus enabling it to accurately learn complex discrimination boundaries.

[0399] In some specific embodiments, the decision generation network is trained. The composite loss function used The expression is:

[0400]

[0401] in, The total linearization cost of a single-level mixed-integer linear programming model for post-disaster recovery;

[0402] This indicates the preset penalty coefficient.

[0403] By training the decision generation network with a composite loss function that integrates linearized cost and feasibility penalty terms, the network is driven to learn the economic optimality of decisions and actively avoid infeasible regions identified by the discriminant network.

[0404] In some specific embodiments, the preset feasibility confidence threshold interval for the active learning mechanism to screen difficult sample features is [0.4, 0.6], that is:

[0405] when At that time, the corresponding scene features Features identified as difficult samples .

[0406] By setting a critical range for feasibility confidence, we can actively filter difficult sample features and trigger additional data generation, enabling the training data to focus on covering complex scenarios near the decision boundary and improving the model's generalization ability under critical conditions.

[0407] The following are embodiments of the rolling clearing scheme generation system for post-disaster distribution network recovery provided in this application. This rolling clearing scheme generation system for post-disaster distribution network recovery belongs to the same inventive concept as the rolling clearing scheme generation method in the above embodiments. For details not described in detail in the embodiments of the rolling clearing scheme generation system, please refer to the embodiments of the rolling clearing scheme generation method for post-disaster distribution network recovery described above.

[0408] like Figure 2 As shown, the rolling clearing scheme generation system for post-disaster power distribution network restoration includes:

[0409] The data acquisition module is used to acquire real-time data of the distribution network, including the physical topology status of the distribution network, the branch repair status, the system-level load demand, the load survival value factor, the renewable energy output level inside the virtual power plant, and the private load demand.

[0410] The preliminary clearing decision generation module is used to construct the current decision time based on the real-time data of the distribution network at the current decision time. Context feature vector ,Will Input pre-trained decision generation network Perform forward inference and output the timing scrolling window. Preliminary clearing decision within ;

[0411] The final clearing decision generation module is used to generate a pre-trained physical feasibility discrimination network. Preliminary clearing decision A security check is performed, and the decision confidence score is output. If the decision confidence score is lower than a preset confidence threshold, the initial clearing decision is corrected through linear projection to obtain the final clearing decision. If the decision confidence level is not lower than the preset confidence level threshold, then the preliminary clearing decision will be made directly. As the final clearing decision ;

[0412] The decision execution module is used to implement the final clearing decision. Release timeline scrolling window The internal incentive price instructions and load restoration control commands are only executed. The operation of the first step size.

[0413] The rolling clearing scheme generation system in this embodiment is used to implement a rolling clearing scheme generation method for post-disaster power distribution network recovery.

[0414] This application also provides an electronic device for implementing the various embodiments of this application. Figure 3To illustrate the hardware structure of an electronic device according to various embodiments of this application, as shown in the following diagram... Figure 3 As shown, the electronic device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor.

[0415] Those skilled in the art will understand that the electronic device structure involved in the embodiments of this application does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0416] In embodiments of this application, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0417] In this application embodiment, the processor can be implemented using at least one of an Application-Specific Integrated Circuit (ASIC), a Digital Signal Processor (DSP), a Digital Signal Processing Device (DSPD), a processor, a controller, a microcontroller, a microprocessor, or an electronic unit designed to perform the functions described herein. In some cases, such implementations can be implemented within a controller. For software implementations, implementations such as processes or functions can be implemented with separate software modules that allow the performance of at least one function or operation. The software code can be implemented by a software application (or program) written in any suitable programming language, and the software code can be stored in memory and executed by the controller.

[0418] In addition, the electronic device includes some functional modules not shown, which will not be described in detail here.

[0419] Those skilled in the art will understand that the various aspects of the electronic device provided in this application can be implemented as a system, method, or program product. Therefore, the various aspects of this application can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0420] This application also provides a storage medium storing a program product capable of generating a rolling clearing scheme for post-disaster distribution network recovery. In some possible implementations, various aspects of this application can also be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this application.

[0421] The storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0422] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating a rolling clearing scheme for post-disaster power distribution network restoration, characterized in that, include: S1. Obtain real-time data of the distribution network, including the physical topology status of the distribution network, branch repair status, system-level load demand, load survival value factor, renewable energy output level within the virtual power plant, and private load demand; S2. Construct the current decision-making time based on the real-time data of the distribution network at the current decision-making time. Context feature vector ,Will Input pre-trained decision generation network Perform forward inference and output the timing scrolling window. Preliminary clearing decision within Preliminary clearing decision A decision vector whose elements include a time-series scrolling window. The incentive electricity price at each time step, the load recovery amount at each node, and the switching operation instructions for each branch; Time-series scrolling window Defined as: in For time step, This represents the total duration of the time-series scrolling window; S3. Using a pre-trained physical feasibility discrimination network Preliminary clearing decision Perform security checks and output decision confidence levels; If the decision confidence level is lower than the preset confidence level threshold, the preliminary clearing decision is corrected through linear projection to obtain the final clearing decision. ; If the decision confidence level is not lower than the preset confidence level threshold, then the preliminary clearing decision will be made directly. As the final clearing decision ; S4. Based on the final clearing decision Release timeline scrolling window The internal incentive price instructions and load restoration control commands are only executed. The first step of the operation will be executed, and the current decision time will be set after execution. Scroll to the next time step, then return to step S1 and repeat; Among them, decision generation network Physical feasibility assessment network All are deep learning neural networks, using simulation datasets. The simulation dataset was obtained after co-training. The construction steps are as follows: Simulate different post-disaster power grid scenarios and construct a set of context feature vectors. Each context feature vector It includes the physical topology status of the distribution network, branch repair status, system-level load demand, load survival value factor, renewable energy output level within the virtual power plant, and private load demand; For each context feature vector Solve the single-level mixed-integer linear programming model for disaster recovery to obtain its corresponding optimal decision vector. The post-disaster recovery single-level mixed integer linear programming model is reconstructed from a pre-built single-leader-multiple-follower model. During the reconstruction process, the McCormick envelope method is used to handle bilinear terms. Record all "feature-decision" pairs To form a simulation dataset .

2. The rolling clearing scheme generation method as described in claim 1, characterized in that, In step S3, the expression for correcting the initial clearing decision through linear projection is: in, express The constraint matrix, where each row corresponds to a physical constraint and each column corresponds to... One of the elements; express The constraint vector is a column vector, where each element represents... The boundary values ​​of the parallel constraints, which are obtained from the original boundaries of the physical constraints through mathematical transformation, include: For upper limit constraints, simply take the upper limit value; For lower bound constraints, take the negative value of the lower bound. For bilateral constraints, they are decomposed into upper limit constraints and lower limit constraints, and then their values ​​are determined according to the corresponding rules.

3. The rolling clearing scheme generation method as described in claim 1, characterized in that, The single-leader-multiple-follower model is a decision optimization model with a two-layer interactive architecture, including a leader decision model and multiple follower decision models that are independently connected to the leader decision model. A horizontal equilibrium mechanism is established between the various follower decision models. The leader decision model corresponds to the power distribution system operator, and each follower decision model corresponds to a virtual power plant. Among them, the decision objective of the leader decision-making model is in the scrolling window. Internally, we will coordinate and optimize the progress of the entire network load recovery, the cost of dispatching our own resources, and the incentive costs of purchasing electricity from virtual power plants. The decision-making objective of the follower decision-making model is to maximize individual profits. After receiving the incentive electricity price signal issued by the leader, it optimizes the output and load recovery of its internal distributed energy resources to determine the net power injected into the distribution network. The horizontal equilibrium mechanism constructs a global situation function to address the decision-making coupling and interest game among multiple followers due to the shared physical power grid. This transforms the Nash equilibrium problem of decentralized decision-making among multiple followers into a single-objective optimization problem of this global situation function.

4. The rolling clearing scheme generation method as described in claim 3, characterized in that, The objective function and constraints of the leader decision-making model include: in, Indicates time No. The incentive electricity price for each virtual power plant is a decision variable for leaders; An index for virtual power plants. For time indexing; Indicates at time node The actual restored active power is a decision variable for leaders. For node indexing, For time indexing; Indicates at time The first generation of power distribution system operators The contribution of resources to resource recovery is a decision variable for leaders. For indexing proprietary resources, For time indexing; Indicates at time branch road The switch state binary decision variable takes a value of 1 to indicate closed and 0 to indicate open. Indicates at time node The load survival value factor is a non-negative weighting coefficient preset according to load type and recovery priority; Indicates at time node System-level load requirements; This indicates that the power distribution system operator owns the first Functions for restoring the operating, fuel, or operation and maintenance costs of resources; Indicates at time No. The net active power injected into the distribution network by each virtual power plant; , These represent the weighting coefficients of the distribution system operator's cost of adjusting its own resources and its expenditure on purchasing electricity from virtual power plants in the objective function, respectively. This represents the set of all nodes in a power distribution network. This represents the set of all virtual power plants; This represents the collection of all proprietary restoration resources owned by the power distribution system operator. Represents dynamic topology evolution constraints; This indicates network trend constraints; This indicates the constraints on the operation of the power distribution system operator's own resources; This indicates a load sequential recovery constraint; Indicates the repair status of the branch road Discrete confidence sets; Indicates system-level load demand Continuous confidence intervals; Indicates system-level load demand The baseline value for prediction at any given time; , This indicates the minimum and maximum allowable values ​​for the incentive electricity price; Indicates the first The profit function of a virtual power plant.

5. The rolling clearing scheme generation method as described in claim 3, characterized in that, The objective function and constraints of the follower decision model include: in, Indicates the first The profit function of a virtual power plant; Indicates at time No. The net active power injected into the distribution network by a virtual power plant is a decision variable for followers; Indicates at time No. The first virtual power plant The output value of energy resources is a decision variable for followers. Index for resource types; Indicates time No. Incentive electricity prices for virtual power plants; Indicates the first The first virtual power plant Operating cost function for energy resources; Indicates the first The collection of all energy resources within a virtual power plant; This represents the boundary constraints for the aggregation flexibility of virtual power plants; express Time of the first Renewable energy within a virtual power plant Real-time prediction of maximum active power output Continuous confidence intervals; express Time of the first Private load demand within a virtual power plant Continuous confidence intervals.

6. The rolling clearing scheme generation method as described in claim 1, characterized in that, The global situation function is: in, Denotes the global state function, where It is all virtual power plants at all times The output vector is composed of the output forces. It is all virtual power plants at all times The incentive price vector is composed of incentive prices; Indicates the first The profit function of a virtual power plant; Indicates at time , used to characterize the The output change of the first virtual power plant affects the second... Sensitivity coefficient for the degree of voltage impact at the grid connection point of a virtual power plant; Indicates at time No. The net active power injected into the distribution network by each virtual power plant; Indicates at time No. The net active power injected into the distribution network by each virtual power plant; This represents the set of all virtual power plants.

7. The method for generating a rolling clearing scheme as described in any one of claims 5-6, characterized in that, System-level load requirements involved in the leader decision-making model Branch road repair status And the maximum output of renewable energy involved in the follower decision-making model. and internal load demand The range of its uncertainty values ​​is determined by an uncertainty quantification system with statistical coverage guarantees; this uncertainty quantification system includes: Benchmark Prediction Layer: Used to extract multi-dimensional time-series features of the post-disaster distribution network, and outputs the predicted benchmark values ​​of key parameters at each time point within the scrolling window through a data-driven model. ; Boundary Representation Layer: Based on real-time error feedback and dynamic calibration mechanisms, it generates distribution-independent confidence boundaries for the predicted baseline values, forming a boundary with a probability of covering the true values ​​of no less than [value missing]. Continuous confidence intervals With discrete confidence sets Continuous confidence interval Including the first continuous confidence interval Second continuous confidence interval Third continuous confidence interval ; The relevant uncertainty parameters in the leader decision-making model and the follower decision-making model are constrained within their corresponding confidence intervals or sets, i.e.: Among them, discrete confidence set This indicates the repair status of the distribution network branch within the rolling time window. The set of possible values; First continuous confidence interval This indicates the system-level load demand within the rolling time window. The range of forecast uncertainty; Second continuous confidence interval This represents the real-time predicted maximum active power output of renewable energy within the virtual power plant during the rolling time window. The range of uncertainty; Third continuous confidence interval This represents the private load demand within the virtual power plant during the rolling time window. The range of prediction uncertainty.

8. The rolling clearing scheme generation method as described in claim 1, characterized in that, The steps to refactor a single-leader-multiple-follower model into a single-level mixed-integer linear programming model for disaster recovery include: For each follower decision model, write out its Lagrange function. And derive the KKT condition equations that satisfy the first-order optimality condition; By introducing the KKT condition equations as additional constraints into the constraint set of the leader decision-making model, the two-level decision-making model is transformed into a single-level mathematical programming problem with equilibrium constraints. By introducing auxiliary binary variables and using the Big M method, the complementary relaxation conditions in the KKT conditions are transformed into a series of linear inequality constraints, resulting in the transformed model. Using the McCormick envelope method, the incentive electricity price variables appearing in the leader's objective function and potential function in the transformed model are analyzed. With response power variables The bilinear product terms are linearized to obtain a single-level mixed-integer linear programming model that can be processed by the standard solver.

9. The rolling clearing scheme generation method as described in claim 1, characterized in that, Using simulation datasets For decision generation networks Physical feasibility assessment network The steps for collaborative training include: Initialize the decision generation network and physical feasibility discrimination network Weight parameters; fixed The parameters, using the dataset Samples in And additional randomly generated samples of physically infeasible decisions, used for supervised learning training. To enable them to distinguish between decision-making In a given scenario Does the following satisfy all physical constraints? fixed The parameters are set to minimize decision-making costs and drive the output to meet physical feasibility, using a dataset. train ; Using the training done in the current round ,Evaluate For training set samples Output decision Feasibility confidence level; select sample features whose confidence level is within a preset critical interval as difficult sample features. Return the features of the hard samples to the simulation dataset. The construction process involves solving for the corresponding optimal decision to generate new data pairs. and add to the dataset ; The network was retrained using the expanded dataset for iterative optimization until... Decision performance and The discrimination accuracy of all converged.

10. A rolling clearing scheme generation system for post-disaster power distribution network restoration, characterized in that, A method for generating a rolling clearing scheme as described in any one of claims 1-9, comprising: The data acquisition module is used to acquire real-time data of the distribution network, including the physical topology status of the distribution network, the branch repair status, the system-level load demand, the load survival value factor, the renewable energy output level inside the virtual power plant, and the private load demand. The preliminary clearing decision generation module is used to construct the current decision time based on the real-time data of the distribution network at the current decision time. Context feature vector ,Will Input pre-trained decision generation network Perform forward inference and output the timing scrolling window. Preliminary clearing decision within ; The final clearing decision generation module is used to generate a pre-trained physical feasibility discrimination network. Preliminary clearing decision A security check is performed, and the decision confidence score is output. If the decision confidence score is lower than a preset confidence threshold, the initial clearing decision is corrected through linear projection to obtain the final clearing decision. If the decision confidence level is not lower than the preset confidence level threshold, then the preliminary clearing decision will be made directly. As the final clearing decision ; The decision execution module is used to implement the final clearing decision. Release timeline scrolling window The internal incentive price instructions and load restoration control commands are only executed. The operation of the first step size.

Images