Distributed resource aggregation method and system for power distribution network

Abstract

This invention relates to the field of distribution network technology, and more particularly to a method and system for distributed resource aggregation in distribution networks. The method includes acquiring real-time operational data of the distribution network and constructing a digital twin model of the aggregation scheduling scheme; converting the scheduling scheme into driving stimuli for virtual execution and generating multi-scenario simulations to obtain multi-dimensional pre-simulation results; based on the pre-simulation results, using a risk-benefit trade-off optimization framework to couple and iteratively optimize resource response timing, output allocation ratio, and reserve capacity until safety constraints are met and the aggregation benefit is optimal; issuing control commands and transmitting back actual operational data to dynamically calibrate the digital twin model. This method improves the security and economy of distributed resource aggregation in distribution networks.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, and in particular to a distributed resource aggregation method and system for power distribution networks. Background Technology

[0002] In existing distributed resource aggregation and scheduling technologies for distribution networks, the conventional approach typically involves mathematically modeling the distribution network based on physical mechanism models or simplified equivalent models. This treats fluctuations in renewable energy output and load demand deviations as random variables, handling uncertainties through scenario generation or robust optimization methods. Scheduling schemes often employ a combination of offline optimization and online adjustment, calculating the output plan, response sequence, and reserve capacity for resource aggregation on a daily or hourly basis, and periodically updating model parameters based on real-time measurement data. These methods rely on static analysis at a single time point, making it difficult to effectively capture the multi-timescale coupled physical characteristics of the distribution network's dynamic evolution.

[0003] The aforementioned conventional approach has two significant drawbacks. First, because the operating status of the distribution network changes rapidly with the increasing access of distributed resources, traditional models have long update cycles and cannot accurately reflect the transient characteristics of the current topology, equipment parameters, and power flow distribution in real time. Second, when there are drastic fluctuations in renewable energy output or load demand, aggregation scheduling schemes based on outdated models deviate significantly from actual operating conditions, leading to frequent safety issues such as voltage exceedances, line overloads, or malfunctions of protection devices. Third, conventional methods typically treat voltage exceedances, power flow margins, and protection malfunctions as post-event verification indicators, only checking them through static security analysis after optimization. This lacks proactive prediction and coupled trade-offs regarding multi-dimensional risks during the scheduling process. Consequently, aggregation schemes often suffer from inconsistencies, making it difficult to achieve optimal resource utilization efficiency and operational economy while ensuring system safety. Summary of the Invention

[0004] This invention provides a distributed resource aggregation method and system for power distribution networks, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a distributed resource aggregation method for power distribution networks, comprising:

[0006] Obtain real-time operational data of the distribution network physical system and distributed resource aggregation scheduling schemes to be executed, and construct a digital twin model of the distribution network based on the real-time operational data;

[0007] The aggregated scheduling scheme is converted into the driving incentive of the digital twin model through control sequence parsing and then virtually executed. By propagating the uncertainty of new energy output fluctuations and load demand deviations, a set of multiple scenarios is generated. In the digital twin model, multi-dimensional pre-simulation results including voltage over-limit probability distribution, power flow margin changes and protection maloperation risk are obtained through parallel simulation.

[0008] Based on the risk factors identified by the multi-dimensional simulation results, a risk-benefit trade-off optimization framework is constructed to couple the solution of resource response timing, output allocation ratio and reserve capacity configuration in the aggregated scheduling scheme, resulting in an adjusted aggregated scheduling scheme. The adjusted aggregated scheduling scheme is then converted into driving incentives for repeated virtual execution and parallel simulation to progressively update the multi-dimensional simulation results. The scheme is continuously optimized through gradient convergence criteria until the safety constraints are met and the aggregated benefit is optimal, and the optimized aggregated scheduling scheme is output.

[0009] Based on the optimized aggregation scheduling scheme, control commands are issued to the distribution network physical system through the communication interface. The actual operation data after execution is transmitted back to the digital twin model through the state synchronization mechanism. The mapping parameters of the digital twin model are dynamically calibrated through the parameter identification algorithm.

[0010] The aggregated scheduling scheme is converted into driving stimuli of the digital twin model through control sequence parsing for virtual execution. A multi-scenario set is generated by propagating the uncertainties of new energy output fluctuations and load demand deviations, including:

[0011] The resource scheduling instructions in the aggregation scheduling scheme are deconstructed in time sequence, and the power setting trajectory and response timing constraints of each distributed resource are extracted to construct a multi-dimensional control sequence matrix.

[0012] An incentive-response mapping function is established based on the multidimensional control sequence matrix, and the control quantities in the multidimensional control sequence matrix are converted into incentive signals for the corresponding resource nodes in the digital twin model. The incentive signals drive the digital twin model to perform state evolution to achieve virtual execution.

[0013] During the virtual execution process, historical samples of new energy output fluctuations and statistical distributions of load demand deviations are collected to construct a joint uncertainty probability space;

[0014] In the joint uncertainty probability space, a set of perturbation vectors covering extreme operating conditions is generated by Latin hypercube sampling. Each perturbation vector in the set of perturbation vectors is superimposed on the excitation signal to form a differentiated excitation scenario, driving the digital twin model to evolve in parallel to generate a set of multiple scenarios.

[0015] In the joint uncertainty probability space, a set of perturbation vectors covering extreme operating conditions is generated through Latin hypercube sampling. The perturbation vectors in this set are then superimposed onto the excitation signal to form differentiated excitation scenarios, including:

[0016] In the joint uncertainty probability space, an orthogonal mesh is constructed with the dimensions of new energy output fluctuation and load demand deviation. Based on the boundary cells of the orthogonal mesh, the boundary subspace corresponding to extreme operating conditions is located. A spatial sampling scheme is generated by configuring sampling density weights in the boundary subspace and the central region respectively through a hierarchical spatial filling strategy.

[0017] According to the spatial sampling scheme, a set of sampling points with spatial uniform distribution characteristics is generated in the joint uncertainty probability space. The coordinate values ​​of each sampling point in the set of sampling points in the dimension of new energy output fluctuation are mapped into the new energy output disturbance amplitude, and the coordinate values ​​of each sampling point in the dimension of load demand deviation are mapped into the load demand deviation amplitude, thereby constructing a set of disturbance vectors.

[0018] For each disturbance vector in the disturbance vector set, the disturbance amplitude of the new energy output is added to the power incentive benchmark value of the corresponding new energy node in the incentive signal, and the disturbance amplitude of the load demand deviation is added to the demand incentive benchmark value of the corresponding load node in the incentive signal, thereby generating a differentiated incentive scenario set that includes deterministic scheduling instructions and uncertain environmental disturbances.

[0019] Based on the risk factors identified by the multi-dimensional pre-simulation results, a risk-benefit trade-off optimization framework is constructed to couple the resource response timing, output allocation ratio, and reserve capacity configuration in the aggregated scheduling scheme, resulting in the adjusted aggregated scheduling scheme, including:

[0020] The distribution characteristics of power imbalance, voltage deviation and frequency deviation under each scenario are deconstructed from the multi-dimensional pre-simulation results. Based on the distribution characteristics, the risk severity index of each scenario is calculated. The operating conditions corresponding to the scenario whose risk severity index exceeds the safe operation threshold are abstracted as risk factors. The risk factors include a triplet description of triggering conditions, affected resource identifiers and deviation magnitude.

[0021] The probability of risk occurrence is quantified based on the triggering conditions, the amount of economic loss is assessed based on the magnitude of the deviation, a risk cost item is formed by the composite operation of the probability of risk occurrence and the amount of economic loss, the expected benefit of the aggregation scheduling scheme is set as the benefit return item, and a risk-benefit trade-off optimization framework including the risk cost item and the benefit return item is established.

[0022] In the risk-reward trade-off optimization framework, the relationship between the impact resource identifier and the response time-series variable, output allocation ratio variable, and reserve capacity configuration variable of the corresponding resource in the aggregated scheduling scheme is established. Based on the relationship, a complementary constraint relationship is established between the advance adjustment amount of the response time-series variable and the increase amount of the reserve capacity configuration variable. Through multi-variable collaborative iterative solution, a variable adjustment scheme that reduces the risk cost item and maintains the benefit return item is obtained. Based on the variable adjustment scheme, the aggregated scheduling scheme is reconstructed to obtain the adjusted aggregated scheduling scheme.

[0023] The probability of risk occurrence is quantified based on the triggering conditions, the economic loss is assessed based on the magnitude of the deviation, a risk cost item is formed through a composite calculation of the probability of risk occurrence and the economic loss, and the expected benefits of the aggregated scheduling scheme are set as a return item, including:

[0024] Using the time dimension characteristics, load fluctuation characteristics, and meteorological condition characteristics in the triggering conditions as input variables, a multi-dimensional mapping relationship is established based on the input variables. Through the multi-dimensional mapping relationship, the frequency of historical risk events corresponding to different combinations of triggering conditions is converted into probability quantification values ​​to obtain the probability of risk occurrence.

[0025] Based on the power deviation magnitude, duration, and impact range in the deviation magnitude, and combined with the unit capacity cost and default penalty coefficient of the resources involved in the aggregation scheduling scheme, a loss transmission path is constructed to correlate and solve the direct and indirect economic losses caused by a single risk event, thereby obtaining the amount of economic loss.

[0026] The risk cost term is obtained by performing a tensor product operation between the probability of the risk occurrence and the amount of economic loss, and by introducing a time discount factor to attenuate the risk cost in the future period.

[0027] By tracing the expected output timing, market transaction electricity price, and ancillary service compensation price of each resource from the aggregated scheduling scheme, the expected total revenue of each resource within the scheduling cycle is calculated to obtain the revenue return item.

[0028] The adjusted aggregation scheduling scheme is converted into driving incentives for repeated virtual execution and parallel inference to progressively update the multi-dimensional pre-simulation results. It is then continuously optimized using a gradient convergence criterion until safety constraints are met and the aggregation benefit is optimal. The optimized aggregation scheduling scheme is output as follows:

[0029] The resource response timing, output allocation ratio and reserve capacity configuration in the adjusted aggregation scheduling scheme are compiled into a driving instruction sequence. The driving instruction sequence is used to stimulate virtual execution and parallel simulation to re-execute the pre-simulation process and generate progressively updated multi-dimensional pre-simulation results.

[0030] From the progressively updated multi-dimensional pre-simulation results, the safety constraint deviation and the aggregated benefit change are extracted. A constraint violation gradient vector is constructed based on the safety constraint deviation, and a benefit optimization gradient vector is constructed based on the aggregated benefit change. The constraint violation gradient vector and the benefit optimization gradient vector are weighted and fused to form a composite gradient field.

[0031] The gradient convergence index is calculated using the magnitude and direction change rate of the composite gradient field. When the gradient convergence index meets the convergence criterion, it is determined that the adjusted aggregation scheduling scheme has reached the Pareto optimal state of safety constraints and aggregation benefits. The current adjusted aggregation scheduling scheme is then output as the optimized aggregation scheduling scheme.

[0032] Based on the optimized aggregation scheduling scheme, control commands are issued to the distribution network physical system through a communication interface. The executed actual operating data is then transmitted back to the digital twin model via a state synchronization mechanism. Dynamic calibration of the mapping parameters of the digital twin model is performed using a parameter identification algorithm, including:

[0033] The resource response timing, output allocation ratio, and reserve capacity configuration in the optimized aggregation scheduling scheme are parsed into a control instruction set, which is then sent to the distribution network physical system through a communication interface. The control instruction set carries a timing synchronization tag.

[0034] The power distribution network physical system performs power regulation according to the control command set, collects actual operating data, and transmits the actual operating data back to the digital twin model through the state synchronization mechanism. In the digital twin model, the actual operating data is aligned with the virtual execution prediction data based on the time synchronization label, and the power tracking error, time response error and state matching error are calculated to form a multi-dimensional identification error vector.

[0035] The multi-dimensional identification error vector is mapped to the mapping parameter space by a parameter identification algorithm, and a gradient relationship between the multi-dimensional identification error vector and the deviation of the mapping parameters is established. Based on the gradient relationship, the mapping parameter correction increment is calculated and superimposed on the current mapping parameter to complete the dynamic calibration of the mapping parameters of the digital twin model.

[0036] A second aspect of this invention provides a distributed resource aggregation system for power distribution networks, comprising:

[0037] A model building unit is constructed to acquire real-time operational data of the distribution network physical system and distributed resource aggregation scheduling schemes to be executed, and a digital twin model of the distribution network is constructed based on the real-time operational data.

[0038] The parallel simulation unit is used to convert the aggregate scheduling scheme into the driving incentive of the digital twin model through control sequence parsing for virtual execution. It generates a set of multiple scenarios by propagating the uncertainty of new energy output fluctuations and load demand deviations. In the digital twin model, it performs parallel simulation to obtain multi-dimensional pre-simulation results including voltage over-limit probability distribution, power flow margin changes and protection maloperation risk.

[0039] An optimization convergence unit is used to construct a risk-benefit trade-off optimization framework based on the risk factors identified by the multi-dimensional pre-simulation results. This framework is then coupled with the solution of resource response timing, output allocation ratio, and reserve capacity configuration in the aggregated scheduling scheme to obtain an adjusted aggregated scheduling scheme. The adjusted aggregated scheduling scheme is then converted into a driving incentive for repeated virtual execution and parallel simulation to progressively update the multi-dimensional pre-simulation results. The scheme is continuously optimized through gradient convergence criteria until the safety constraints are met and the aggregated benefit is optimal. Finally, the optimized aggregated scheduling scheme is output.

[0040] The dynamic calibration unit is used to issue control commands to the distribution network physical system through a communication interface based on the optimized aggregation scheduling scheme, transmit the actual operation data after execution back to the digital twin model through a state synchronization mechanism, and dynamically calibrate the mapping parameters of the digital twin model through a parameter identification algorithm.

[0041] A third aspect of the present invention provides an electronic device, comprising:

[0042] processor;

[0043] Memory used to store processor-executable instructions;

[0044] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0046] This method significantly improves the security level of distributed resource aggregation in distribution networks. Multi-scenario parallel simulations based on digital twin models can systematically capture the uncertainties of renewable energy output and load demand. The simulation results accurately characterize the voltage over-limit probability distribution, power flow margin change trajectory, and protection maloperation risk, enabling dispatchers to identify potential risk factors in advance. By explicitly incorporating risk factors into the optimization framework, overvoltage phenomena caused by resource response timing conflicts are effectively suppressed. Dynamically adjusting reserve capacity configuration enhances the system's robustness in extreme operating conditions, reducing voltage over-limit risk to below the safe threshold and ensuring the reliability of distribution network operation.

[0047] The optimization capability of the aggregation scheduling scheme is fundamentally enhanced. A risk-benefit trade-off mechanism establishes a correlation between resource output allocation ratio and reserve capacity configuration, resolving the conflict between safety and economy in traditional methods through coupled solution. The iterative deduction process accelerates optimization convergence with the gradient convergence criterion, maximizing the comprehensive benefits of distributed resource aggregation while satisfying safety boundaries such as power flow constraints and protection coordination. Progressive updates of multi-dimensional pre-simulation results enable automatic correction of the optimization direction, avoiding getting trapped in local optima, and significantly improving aggregation benefits compared to static optimization schemes.

[0048] This method endows the distribution network with adaptive adjustment capabilities to uncertainties. Real-time operational data drives the dynamic evolution of the digital twin model, and the parameter identification algorithm continuously calibrates the mapping parameters, ensuring high-fidelity synchronization between the twin model and the physical system. A state synchronization mechanism after control commands are issued forms a closed-loop feedback loop; operational deviations are triggered by data feedback to correct model parameters, enabling the scheduling scheme to automatically adjust to changes in grid operating conditions. This proactive adaptation mechanism eliminates decision-making biases caused by model mismatch, extends the effective cycle of the optimized scheme, and reduces the frequency and cost of repeated scheduling. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the distributed resource aggregation method for power distribution networks according to an embodiment of the present invention.

[0050] Figure 2 This is a flowchart of the aggregation scheduling optimization iteration based on digital twins in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating the distributed resource aggregation method for power distribution networks according to an embodiment of the present invention.

[0054] Distributed resource aggregation methods for distribution networks include:

[0055] Obtain real-time operational data of the distribution network physical system and distributed resource aggregation scheduling schemes to be executed, and construct a digital twin model of the distribution network based on the real-time operational data;

[0056] The aggregated scheduling scheme is converted into the driving incentive of the digital twin model through control sequence parsing and then virtually executed. By propagating the uncertainty of new energy output fluctuations and load demand deviations, a set of multiple scenarios is generated. In the digital twin model, multi-dimensional pre-simulation results including voltage over-limit probability distribution, power flow margin changes and protection maloperation risk are obtained through parallel simulation.

[0057] Based on the risk factors identified by the multi-dimensional simulation results, a risk-benefit trade-off optimization framework is constructed to couple the solution of resource response timing, output allocation ratio and reserve capacity configuration in the aggregated scheduling scheme, resulting in an adjusted aggregated scheduling scheme. The adjusted aggregated scheduling scheme is then converted into driving incentives for repeated virtual execution and parallel simulation to progressively update the multi-dimensional simulation results. The scheme is continuously optimized through gradient convergence criteria until the safety constraints are met and the aggregated benefit is optimal, and the optimized aggregated scheduling scheme is output.

[0058] Based on the optimized aggregation scheduling scheme, control commands are issued to the distribution network physical system through the communication interface. The actual operation data after execution is transmitted back to the digital twin model through the state synchronization mechanism. The mapping parameters of the digital twin model are dynamically calibrated through the parameter identification algorithm.

[0059] In one optional implementation, the aggregated scheduling scheme is converted into driving stimuli for virtual execution by control sequence parsing into the digital twin model. The generation of a multi-scenario set by propagating the uncertainties of new energy output fluctuations and load demand deviations includes:

[0060] The resource scheduling instructions in the aggregation scheduling scheme are deconstructed in time sequence, and the power setting trajectory and response timing constraints of each distributed resource are extracted to construct a multi-dimensional control sequence matrix.

[0061] An incentive-response mapping function is established based on the multidimensional control sequence matrix, and the control quantities in the multidimensional control sequence matrix are converted into incentive signals for the corresponding resource nodes in the digital twin model. The incentive signals drive the digital twin model to perform state evolution to achieve virtual execution.

[0062] During the virtual execution process, historical samples of new energy output fluctuations and statistical distributions of load demand deviations are collected to construct a joint uncertainty probability space;

[0063] In the joint uncertainty probability space, a set of perturbation vectors covering extreme operating conditions is generated by Latin hypercube sampling. Each perturbation vector in the set of perturbation vectors is superimposed on the excitation signal to form a differentiated excitation scenario, driving the digital twin model to evolve in parallel to generate a set of multiple scenarios.

[0064] When performing time-series deconstruction on the aggregated scheduling scheme, all distributed resource scheduling instructions involved in the scheme need to be broken down according to time and resource dimensions. Specifically, for each type of distributed resource in the scheme—including distributed photovoltaics, energy storage devices, adjustable loads, and small wind turbines—the power setpoints for each time step within the scheduling cycle, as well as the corresponding response timing constraints, such as ramp rate limits, minimum continuous response duration, and switching interval requirements, are extracted. This information is then organized into a multi-dimensional control sequence matrix. The rows of the matrix correspond to resource node numbers, the columns correspond to the discretized scheduling time steps, and the matrix elements record the power setpoint trajectory of each resource at the corresponding time step and its constraint markers. The construction of this matrix provides a structured data foundation for the accurate mapping of subsequent excitation signals, while preserving the temporal coupling relationships between resources and avoiding the loss of coordination constraint information when processing individual resource instructions.

[0065] Based on a multidimensional control sequence matrix, an excitation-response mapping function is established to convert the control quantities in the matrix into excitation signals for the corresponding resource nodes in the digital twin model. The core of the mapping function lies in establishing the correspondence between physical control quantities and the internal state variables of the digital twin model: for energy storage nodes, the power setpoint needs to be converted into charging / discharging current commands and state-of-charge boundaries; for adjustable load nodes, the load reduction needs to be mapped into equivalent admittance changes; for distributed photovoltaic nodes, the active power output setpoint needs to be converted into an injected current vector in conjunction with the inverter control mode. The excitation signals are injected into the digital twin model in a time-driven manner, triggering state updates for each resource node sequentially according to the scheduling time steps. This drives the model to complete the entire state evolution from the initial operating state to the end of the scheduling scheme execution, thereby realizing the virtual execution of the aggregated scheduling scheme in the digital twin model. During virtual execution, the power flow equations and node voltage equations within the model are solved in real time with each injection of the excitation signal, ensuring that the electrical state at each time step is in a physically consistent steady or quasi-steady state.

[0066] During virtual execution, historical sample sequences of renewable energy output fluctuations are simultaneously collected from the historical operation database, along with load demand deviation data from electricity consumption statistics. For renewable energy output fluctuations, the mean, standard deviation, and tail distribution characteristics are obtained by fitting the probability density function of historical samples. Considering the intraday correlation of photovoltaic output and the spatial correlation of wind power output, a multivariate joint probability distribution is constructed. For load demand deviations, deviation statistical models are established according to different electricity consumption types (residential load, commercial load, and industrial load), and a correlation coefficient matrix between loads is introduced to characterize the correlation structure of load deviations at different nodes on the same feeder. The renewable energy output fluctuation distribution and the load demand deviation distribution are coupled through a Copula function to construct a joint uncertainty probability space, enabling the sampled disturbance samples to simultaneously reflect the marginal distribution characteristics of the two types of uncertainty sources and their interdependence.

[0067] In the joint uncertainty probability space, a Latin hypercube sampling method is used to generate a set of perturbation vectors. Latin hypercube sampling uniformly divides the probability interval of each uncertainty dimension into several equally probable sub-intervals, randomly selecting a sample point within each sub-interval. By randomly arranging and combining the sample points of each dimension, good uniform coverage of the samples is ensured throughout the multidimensional probability space. Compared with Monte Carlo random sampling, Latin hypercube sampling can more fully cover the extreme regions of the probability space with the same number of samples, showing a significant advantage in capturing extreme conditions with low probability and high impact. Let the set of perturbation vectors generated by sampling contain... Each perturbation vector... ( It includes the output deviation component of each new energy node and the demand deviation component of each load node, and its dimension is consistent with the total number of nodes participating in uncertainty modeling in the distribution network.

[0068] Each perturbation vector in the perturbation vector set is superimposed on the reference excitation signal to form a differentiated excitation scenario. For the th Each scenario has its excitation signal. From the reference excitation signal With the corresponding perturbation vector Superposition, that is The overlay operation is performed synchronously in both the time and node dimensions, ensuring that the incentives received by each resource node in each scenario at each time step include the corresponding uncertainties and disturbances. Differentiated incentive scenarios are simultaneously injected into parallel computing instances of the digital twin model. Each instance independently executes the state evolution process, completing the entire simulation from the initial state to the end of the scheduling cycle without interference. Parallel evolution fully utilizes the multi-threaded or distributed computing capabilities of the digital twin platform, compressing the simulation of a large number of scenarios that originally needed to be executed serially into a single parallel computing cycle, significantly reducing the time overhead of generating multi-scenario sets.

[0069] Each parallel evolution instance outputs the full-time node voltage sequence, branch power flow sequence, and resource output realization sequence for the corresponding scenario, which are then aggregated to form a multi-scenario set. This multi-scenario set covers a broad distribution of operating states, from average to extreme conditions, providing complete scenario samples to support subsequent parallel simulations in the digital twin model to obtain multi-dimensional pre-simulation results such as voltage over-limit probability distribution, power flow margin changes, and protection maloperation risk. During the generation of the scenario set, the physical feasibility of the excitation signals for each scenario must be verified. Unreasonable scenarios where power setpoints exceed the equipment's rated range or violate ramping constraints due to superimposed disturbances are eliminated. Furthermore, boundary values ​​that meet the constraints are truncated to ensure that each scenario in the multi-scenario set corresponds to a physically achievable operating state.

[0070] For scenarios with high dimensionality in the joint uncertainty probability space and complex correlation structures among the uncertainty variables, the selection of the Copula function should be based on the tail correlation test results of historical data. Priority should be given to Clayton Copula or Gumbel Copula, which can characterize upper or lower tail correlation features, to ensure that extreme unidirectional deviation scenarios (such as significant photovoltaic power underperformance coupled with a significant load increase) are fully reflected in the sampling results. (Sample size...) In determining the sampling size, a balance needs to be struck between sufficient scene coverage and the consumption of parallel computing resources. Typically, the convergence speed of key risk indicators (such as the probability of voltage exceeding the limit) as the sample size increases is evaluated through pre-experiments, and the minimum sample size corresponding to the estimation error of the indicator is selected as the final sampling size.

[0071] In one optional implementation, a set of perturbation vectors covering extreme operating conditions is generated in the joint uncertainty probability space through Latin hypercube sampling. The perturbation vectors in the set are then superimposed onto the excitation signal to form differentiated excitation scenarios, including:

[0072] In the joint uncertainty probability space, an orthogonal mesh is constructed with the dimensions of new energy output fluctuation and load demand deviation. Based on the boundary cells of the orthogonal mesh, the boundary subspace corresponding to extreme operating conditions is located. A spatial sampling scheme is generated by configuring sampling density weights in the boundary subspace and the central region respectively through a hierarchical spatial filling strategy.

[0073] According to the spatial sampling scheme, a set of sampling points with spatial uniform distribution characteristics is generated in the joint uncertainty probability space. The coordinate values ​​of each sampling point in the set of sampling points in the dimension of new energy output fluctuation are mapped into the new energy output disturbance amplitude, and the coordinate values ​​of each sampling point in the dimension of load demand deviation are mapped into the load demand deviation amplitude, thereby constructing a set of disturbance vectors.

[0074] For each disturbance vector in the disturbance vector set, the disturbance amplitude of the new energy output is added to the power incentive benchmark value of the corresponding new energy node in the incentive signal, and the disturbance amplitude of the load demand deviation is added to the demand incentive benchmark value of the corresponding load node in the incentive signal, thereby generating a differentiated incentive scenario set that includes deterministic scheduling instructions and uncertain environmental disturbances.

[0075] In the joint uncertainty probability space, fluctuations in renewable energy output and load demand deviations constitute two independent yet jointly influential uncertainty dimensions affecting the distribution network's operational status. For these two dimensions, the probability space is divided into several equally probable intervals, and a Cartesian product operation is performed on both dimensions to form an orthogonal mesh covering the entire probability space. The rows of the mesh correspond to the renewable energy output fluctuation dimension, and the columns correspond to the load demand deviation dimension; each mesh cell represents an equally probable joint subspace. By statistically analyzing the fluctuation range of renewable energy output and the distribution of load demand deviations in historical operational data, the probability distribution functions for both dimensions are determined and mapped to a standardized probability space coordinate system, ensuring that each mesh cell has an equal weight in a probabilistic sense.

[0076] After establishing the orthogonal mesh, cells located in the grid boundary region are identified. These boundary cells correspond to extreme operating conditions such as maximum or minimum renewable energy output and peak or valley load demand. The boundary subspace is located based on the extreme values ​​of the grid coordinates; that is, when the grid index of a certain dimension is located in the first or last few layers, the cell is assigned to the boundary subspace. The central region corresponds to the set of grid cells in both dimensions that are within the middle probability range, reflecting typical operating conditions of the distribution network in daily operation. Considering that extreme operating conditions pose a significantly higher threat to the safe operation of the distribution network than normal operating conditions, a higher sampling density weight is configured in the boundary subspace, while a relatively lower sampling density weight is configured in the central region. This results in a more densely covered set of disturbance vectors in the extreme operating condition region, ensuring the sufficiency of the risk assessment. Let the sampling density weight of the boundary subspace be... The sampling density weight of the central region is The total number of sampling points is The number of sampling points allocated in the boundary subspace is then... The number of sampling points allocated to the central area is the same as the number of sampling points allocated to the remaining areas. Together, these two parts constitute a complete spatial sampling scheme.

[0077] Based on the aforementioned spatial sampling scheme, hierarchical spatial filling sampling is performed in the joint uncertainty probability space. Within each selected grid cell, coordinate values ​​are uniformly and randomly sampled within the probability interval of that cell to ensure that the sampling points do not overlap and are evenly distributed within the cell. For grid cells in the boundary subspace, multiple sampling points are densely arranged within them; for grid cells in the central region, fewer sampling points are arranged within each cell. The coordinate values ​​of the new energy output fluctuation dimension of all sampling points are... The probability interval corresponding to uniform coverage, and the coordinate values ​​of the load demand deviation dimension. Similarly, it uniformly covers the corresponding probability interval, where This serves as the index for the sampling points. The entire set of sampling points satisfies the characteristic of spatial uniform distribution in both dimensions, avoiding the problems of sample clustering or blank areas that occur in purely random sampling, thereby ensuring the representativeness of the multi-scenario inference results for the entire uncertainty space.

[0078] After the set of sampling points is generated, the coordinate values ​​of each sampling point are converted into the perturbation amplitude in terms of physical dimensions. For the dimension of new energy output fluctuation, the probability coordinates of the sampling points are converted using the inverse cumulative distribution function. Mapped to the amplitude of power output disturbance of new energy sources ,Right now ,in The inverse cumulative distribution function for the fluctuation of renewable energy output is obtained by fitting historical output data, and is typically modeled using a beta distribution or a normal mixture distribution. For the load demand deviation dimension, the probability coordinates of the sampling points are expressed using the corresponding inverse cumulative distribution function. Mapped to load demand deviation magnitude ,Right now ,in This is the inverse cumulative distribution function of the load demand deviation, obtained by fitting historical load data. The corresponding data for each sampling point is... and Combined, the perturbation vector corresponding to the sampling point is formed. It includes the output deviation component and demand deviation component of each node, and the disturbance vectors of all sampling points together constitute the disturbance vector set.

[0079] In the process of superimposing the perturbation vector onto the excitation signal, it is necessary to accurately map each component of the perturbation vector to a specific node in the digital twin model. Excitation signal It includes power incentive baseline values ​​for all new energy nodes and demand incentive baseline values ​​for all load nodes. These baseline values ​​are derived from the deterministic scheduling instructions obtained after control sequence parsing of the aggregated scheduling scheme. For the [missing information]... perturbation vector Extracting those belonging to new energy nodes Output disturbance component Superimposed by addition The power excitation reference value of this node The node is obtained at the 1st position. Power excitation value in each scenario Similarly, extract the load nodes. Demand deviation component This is superimposed on the corresponding demand incentive baseline value. ,get Reassemble the updated incentive values ​​of all nodes to form the first... Differentiated excitation signals for each scenario The excitation signal contains both deterministic scheduling instructions from the aggregation scheduling scheme and uncertain environmental disturbances from the disturbance vector.

[0080] Repeat the above superposition operation for all perturbation vectors in the perturbation vector set, and finally generate a vector containing... This set of incentive scenarios provides differentiated incentive scenarios. Each scenario corresponds to a specific combination of renewable energy output and load demand, and the entire set covers the complete uncertainty space from extreme to typical operating conditions. The incentive scenarios in the boundary subspace can effectively trigger safety risks such as voltage overruns and power flow overloads, providing sufficient scenario support for risk factor identification in subsequent multi-dimensional simulation results. The incentive scenarios in the central region reflect the aggregated benefit characteristics under normal operating conditions, providing a reliable benchmark reference for risk-benefit trade-off optimization. Through this hierarchical weighted spatial filling sampling strategy, the generated set of incentive scenarios, under the constraint of limited computing resources, simultaneously considers the safety assessment needs of extreme operating conditions and the economic assessment needs of typical operating conditions, providing high-quality input for the parallel extrapolation of subsequent digital twin models.

[0081] In one optional implementation, based on the risk factors identified by the multi-dimensional pre-simulation results, a risk-benefit trade-off optimization framework is constructed to couple and solve the resource response timing, output allocation ratio, and reserve capacity configuration in the aggregated scheduling scheme, resulting in an adjusted aggregated scheduling scheme including:

[0082] The distribution characteristics of power imbalance, voltage deviation and frequency deviation under each scenario are deconstructed from the multi-dimensional pre-simulation results. Based on the distribution characteristics, the risk severity index of each scenario is calculated. The operating conditions corresponding to the scenario whose risk severity index exceeds the safe operation threshold are abstracted as risk factors. The risk factors include a triplet description of triggering conditions, affected resource identifiers and deviation magnitude.

[0083] The probability of risk occurrence is quantified based on the triggering conditions, the amount of economic loss is assessed based on the magnitude of the deviation, a risk cost item is formed by the composite operation of the probability of risk occurrence and the amount of economic loss, the expected benefit of the aggregation scheduling scheme is set as the benefit return item, and a risk-benefit trade-off optimization framework including the risk cost item and the benefit return item is established.

[0084] In the risk-reward trade-off optimization framework, the relationship between the impact resource identifier and the response time-series variable, output allocation ratio variable, and reserve capacity configuration variable of the corresponding resource in the aggregated scheduling scheme is established. Based on the relationship, a complementary constraint relationship is established between the advance adjustment amount of the response time-series variable and the increase amount of the reserve capacity configuration variable. Through multi-variable collaborative iterative solution, a variable adjustment scheme that reduces the risk cost item and maintains the benefit return item is obtained. Based on the variable adjustment scheme, the aggregated scheduling scheme is reconstructed to obtain the adjusted aggregated scheduling scheme.

[0085] Electrical quantity distribution data for each scenario were extracted from multi-dimensional pre-simulation results. For each scenario, the statistical distribution characteristics of power imbalance, voltage deviation, and frequency deviation were statistically analyzed. Power imbalance reflects the difference between distributed resource output and actual load demand; voltage deviation describes the degree of deviation of node voltage from its rated value; and frequency deviation characterizes the magnitude of deviation between the system frequency and the nominal frequency. For these three types of electrical quantities, their mean, variance, and extreme quantiles were calculated for each scenario to comprehensively characterize their distribution characteristics. Based on this, the risk severity index for each scenario was calculated by combining the distribution characteristics of the three indicators. Specifically, Excessive amplitude of power imbalance Excessive voltage offset and the excessive amplitude of frequency deviation Weighted composition, with weight coefficients as follows: , and The calculation relationship is The weighting coefficients are calibrated according to the actual safe operation procedures of the distribution network. Exceeding the safe operating threshold The scenarios are selected, and the corresponding operating conditions are abstracted into risk factors. Each risk factor is described using a triple, with three components: triggering condition, affected resource identifier, and deviation magnitude. The triggering condition describes the specific electrical boundary condition that causes the risk severity to exceed the limit; the affected resource identifier points to the distributed resource node number directly associated with the risk; and the deviation magnitude quantifies the magnitude by which the relevant electrical quantity deviates from the safe operating boundary under this risk scenario. This triple abstraction method makes the risk factors traceable and operable, providing a clear input basis for the subsequent construction of the optimization framework.

[0086] Based on the triggering conditions in the triplet, the probability of each risk factor occurring is quantified. The operational boundary corresponding to the triggering condition has a clear statistical frequency in multi-scenario simulations. The ratio of the number of scenarios that meet the triggering conditions to the total number of scenarios is taken as the probability of risk occurrence. Based on the magnitude of the deviation in the triplet, assess the corresponding economic loss. The economic losses encompass equipment damage caused by voltage exceeding limits, penalties for wind and solar power curtailment due to power imbalance, and compensation costs for load interruptions caused by frequency deviations. This relates to the probability of the risk occurring. With economic losses Perform compound calculations to form a risk cost item. The calculation relationship is The total risk cost is obtained by summing up the risk costs of all identified risk factors. The process involves summing all identified risk factors. The sum of the expected electricity revenue, ancillary service revenue, and demand response subsidies that the aggregated dispatch scheme will obtain under normal operating conditions is defined as the revenue return item. To maximize net return, a risk-return trade-off optimization framework is established, incorporating risk cost and return terms. The objective function is to maximize net return. At the same time, system safety operation constraints are imposed, including node voltage upper and lower limit constraints, line power flow capacity constraints, and system frequency deviation constraints.

[0087] In the risk-reward trade-off optimization framework, the impact resource identifiers in the triples are used to establish an explicit association between each risk factor and the corresponding distributed resource in the aggregation scheduling scheme. For each associated resource, three types of optimization variables are introduced: response time-series variables... This represents the amount of advance adjustment of resource response actions relative to the original planned time, and is a variable representing the output allocation ratio. This indicates the proportion of power allocation for this resource in the total output of aggregated scheduling; it is a reserve capacity configuration variable. This represents the amount of reserve capacity reserved to cope with uncertainty. These three types of variables form a structured set of decision variables by influencing resource identifiers and their one-to-one correspondence with specific resource nodes.

[0088] Response time series variables Configuration variables with spare capacity There is an inherent complementary constraint relationship between them: when the response lead time of a resource increases, that resource can intervene in the adjustment process earlier, and the reserve capacity allocated to it can be appropriately reduced accordingly; conversely, when the response lead time is constrained by physical or contractual limitations and cannot be further increased, it is necessary to increase the reserve capacity allocation to compensate for the risk exposure caused by the response delay. This complementary constraint relationship can be expressed as: for each associated resource, and The adjustments are in opposite directions, and the sum of their weighted adjustments does not exceed the upper limit of the operational flexibility that the resource can withstand. ,Right now ,in In order to adjust the increment in advance in response to timing, Increase in reserve capacity and These represent the corresponding flexibility consumption coefficients. The introduction of complementary constraints ensures a coordinated balance between timing adjustments and backup configurations during the optimization process, avoiding situations where unilateral over-adjustment leads to insufficient safety margins in other dimensions.

[0089] After establishing the aforementioned variable relationships and complementary constraints, a multivariate collaborative iterative solution method is employed to solve the risk-reward tradeoff optimization framework. During the iteration process, each iteration simultaneously updates the values ​​of all related resources. , and The system employs three types of variables, calculating the gradient direction of the objective function with respect to each variable, adjusting the risk cost term along the gradient descent direction, and simultaneously ensuring the return term through constrained projection. Not lower than the preset minimum profit protection level After each iteration, the updated set of variables is substituted into the risk-reward tradeoff optimization framework to reassess the total risk cost. and return on investment To determine whether the gradient convergence criterion is met: when the absolute value of the change in the objective function between two adjacent iterations is lower than the convergence threshold. When the iteration is considered to have converged, the output variable adjustment scheme after convergence includes the incremental adjustment of the response timing of each resource. Adjustment amount of output distribution ratio and increase in reserve capacity .

[0090] Based on the variable adjustment scheme, the scheduling instructions for each resource in the original aggregate scheduling scheme are reconstructed: the original response time is advanced. Each time step will adjust the original output allocation ratio according to... The original backup capacity configuration will be increased. The reconstructed instruction set constitutes the adjusted aggregation scheduling scheme, which ensures that the aggregation benefit is no less than... Under the premise of ensuring safety, the risk cost items corresponding to each risk factor are significantly reduced, achieving a coordinated balance between safety and economy. The adjusted aggregation scheduling scheme will be transformed into a new driving incentive, re-input into the digital twin model for virtual execution and multi-scenario parallel simulation, to verify the adjustment effect and trigger the next round of progressive optimization iteration until all safety constraints are met and the aggregation benefit reaches the optimal level.

[0091] In one optional implementation, the probability of risk occurrence is quantified based on the triggering conditions, the economic loss is assessed based on the magnitude of the deviation, a risk cost item is formed through a composite operation of the probability of risk occurrence and the economic loss, and the expected benefit of the aggregation scheduling scheme is set as a return item, including:

[0092] Using the time dimension characteristics, load fluctuation characteristics, and meteorological condition characteristics in the triggering conditions as input variables, a multi-dimensional mapping relationship is established based on the input variables. Through the multi-dimensional mapping relationship, the frequency of historical risk events corresponding to different combinations of triggering conditions is converted into probability quantification values ​​to obtain the probability of risk occurrence.

[0093] Based on the power deviation magnitude, duration, and impact range in the deviation magnitude, and combined with the unit capacity cost and default penalty coefficient of the resources involved in the aggregation scheduling scheme, a loss transmission path is constructed to correlate and solve the direct and indirect economic losses caused by a single risk event, thereby obtaining the amount of economic loss.

[0094] The risk cost term is obtained by performing a tensor product operation between the probability of the risk occurrence and the amount of economic loss, and by introducing a time discount factor to attenuate the risk cost in the future period.

[0095] By tracing the expected output timing, market transaction electricity price, and ancillary service compensation price of each resource from the aggregated scheduling scheme, the expected total revenue of each resource within the scheduling cycle is calculated to obtain the revenue return item.

[0096] In quantifying the probability of risk occurrence, the time dimension features, load fluctuation features, and meteorological condition features of the triggering conditions are used as input variables to construct a multi-dimensional mapping relationship. The time dimension features include the peak / valley period identifier, holiday markers, and season category of the scheduling period; the load fluctuation features include the load ramp-up rate of the current period, the root mean square error of load prediction, and the historical fluctuation frequency; the meteorological condition features include wind speed, light intensity, temperature, and their prediction confidence intervals. These feature vectors are input into the multi-dimensional mapping relationship, and through kernel density estimation or nonparametric regression methods, the frequency of historical events in the historical risk event database that are similar to the current triggering condition combination is mapped to a probability quantification value, thus obtaining the probability of risk occurrence. Specifically, for each type of risk factor... The algorithm retrieves the most recent historical samples in the feature space that are closest to the current input variable from the historical dataset, calculates the frequency percentage of risk events triggered in these samples, and uses Laplace smoothing to avoid zero probability issues. Finally, it outputs the probability estimate of the risk occurrence under the combination of triggering conditions.

[0097] In assessing economic losses, the loss propagation path is constructed based on the power deviation magnitude, duration, and impact range within the deviation magnitude, combined with the unit capacity cost of resources involved in the aggregated dispatch scheme and the default penalty coefficient. Direct economic losses include the cost of reserve resource mobilization, wind and solar curtailment losses, and power imbalance penalties caused by power deviation; indirect economic losses include accelerated depreciation costs of equipment due to voltage overruns or protection malfunctions, user power outage compensation costs, and the converted value of market reputation losses. Let the power deviation magnitude be... (Unit: MW), duration is (Unit: hours), the affected area is the number of affected nodes. The unit capacity cost is (Yuan / MW·h), the default penalty coefficient is (yuan / MW), then the direct economic loss pass The indirect economic losses were calculated. By influence range coefficient Correlation between direct losses and the proportion of direct losses An estimate is made, and the final economic loss is obtained by adding up the direct and indirect economic losses. .

[0098] In the process of forming the risk cost item, the probability of risk occurrence is... With economic losses Tensor product operations are performed, and a time discounting factor is introduced to decay the risk cost for future periods. Time discounting factor Defined as the time interval between future time periods and the current time period. (Unit: hours) Expressed using the coefficient of exponential decay as ,in The discount rate parameter reflects the degree of time preference for future risk. For the [number]th [unit] within the scheduling period... Time period ( , Risk factors (total number of time periods divided for the scheduling cycle) The corresponding risk cost item pass Calculation, where and The first The probability of risk occurrence and the amount of economic loss during a given period. For the first The time discount factor corresponding to each time period. The total risk cost is obtained by summing the risk cost items for all time periods and all risk factors. The meaning of tensor product operation is that, in the case of multiple risk factors existing at the same time, the probability vector and loss vector of each risk factor are multiplied element by element and then summed, rather than a simple scalar multiplication, so as to accurately reflect the comprehensive cost effect of the coupling and superposition of multiple risk factors.

[0099] In setting revenue return items, the expected output sequence, market transaction price, and ancillary service compensation price of each resource are traced from the aggregated dispatch scheme. The expected output sequence provides the planned output value of each resource for each time period of the dispatch cycle; the market transaction price includes the time-of-use price in the spot market. (yuan / MW·h) and medium- and long-term contract electricity prices (RMB / MW·h); Ancillary service compensation prices include the unit price for frequency regulation service compensation. (RMB / MW·h) and standby service compensation unit price (Yuan / MW·h). Regarding resources ( , (the total number of resources participating in the aggregation scheduling), which in the th... The electricity revenue during the period is the expected output power. (MW) and duration (h) and the corresponding market electricity price, the ancillary service revenue is the resource in the [h]th month. Frequency modulation capacity provided during the time period (MW) and reserve capacity The product of (MW) and the corresponding compensation unit price. The expected total revenue of each resource over the entire scheduling cycle. The revenue return item of the aggregation scheduling scheme is obtained by summing the electricity revenue and ancillary service revenue for all time periods, and then summing the expected total revenue of all resources. .

[0100] In practical applications, the calculation of the above-mentioned revenue return item needs to consider the uncertainty of market electricity prices. For fluctuations in spot market electricity prices, a multi-scenario parallel extrapolation method, the same as the uncertainty scenario set, is adopted. The expected revenue value is calculated separately for each scenario, and then a weighted average is calculated using scenario weights to obtain a robust revenue return item estimate considering electricity price uncertainty. Meanwhile, the ancillary service compensation price will dynamically adjust according to the service quality provided by aggregated resources (such as frequency regulation response rate and standby activation success rate). Therefore, when tracing ancillary service revenue, it is necessary to evaluate the service quality parameters of each resource based on the extrapolation results of the digital twin model, and substitute the evaluation results into the compensation price correction formula to ensure that the calculation of the revenue return item is consistent with the actual implementation. This is achieved by using the risk cost item... With return on investment By incorporating them into a risk-reward trade-off optimization framework, we can achieve the optimal solution for aggregation benefits while ensuring the safety of aggregation scheduling.

[0101] In one optional implementation, the adjusted aggregation scheduling scheme is converted into driving incentives to repeatedly execute virtual execution and parallel inference to progressively update the multi-dimensional pre-simulation results. It is continuously optimized using a gradient convergence criterion until safety constraints are met and the aggregation benefit is optimal. The optimized aggregation scheduling scheme is then output, including:

[0102] The resource response timing, output allocation ratio and reserve capacity configuration in the adjusted aggregation scheduling scheme are compiled into a driving instruction sequence. The driving instruction sequence is used to stimulate virtual execution and parallel simulation to re-execute the pre-simulation process and generate progressively updated multi-dimensional pre-simulation results.

[0103] From the progressively updated multi-dimensional pre-simulation results, the safety constraint deviation and the aggregated benefit change are extracted. A constraint violation gradient vector is constructed based on the safety constraint deviation, and a benefit optimization gradient vector is constructed based on the aggregated benefit change. The constraint violation gradient vector and the benefit optimization gradient vector are weighted and fused to form a composite gradient field.

[0104] The gradient convergence index is calculated using the magnitude and direction change rate of the composite gradient field. When the gradient convergence index meets the convergence criterion, it is determined that the adjusted aggregation scheduling scheme has reached the Pareto optimal state of safety constraints and aggregation benefits. The current adjusted aggregation scheduling scheme is then output as the optimized aggregation scheduling scheme.

[0105] like Figure 2 As shown, the method includes:

[0106] After completing the coupled solution of the risk-reward trade-off optimization framework and obtaining the adjusted aggregation scheduling scheme, the scheme needs to be re-injected into the digital twin model for iterative verification to ensure that the final output optimization scheme simultaneously satisfies both the safety constraints and the optimal aggregation benefit.

[0107] The adjusted aggregated scheduling scheme incorporates three types of decision variables: resource response timing, output allocation ratio, and reserve capacity configuration. These variables are expanded along a timeline and mapped node by node, compiling into a structured sequence of driving instructions. Each instruction in the sequence corresponds to a control action for a specific distributed resource within a scheduling period, including fields such as target output value, response start time, duration, and reserve capacity. This instruction sequence is encapsulated in a format compatible with the digital twin model's interface protocol and injected into the digital twin model as the stimulus input for a new round of virtual execution, triggering the model to re-execute the multi-scenario parallel simulation process. The parallel simulation process remains consistent with the initial pre-simulation process, independently running power flow calculations and protection action simulations for each scenario in the generated multi-scenario set. The output includes progressively updated multi-dimensional pre-simulation results, including voltage over-limit probability distribution, power flow margin changes, and protection maloperation risk. Progressive updates mean that the pre-simulation results of each iteration are calculated based on the previous round of optimization decisions, rather than resetting to the initial state, thus ensuring the continuity of the optimization path and the consistency of the convergence direction.

[0108] From the progressively updated multi-dimensional pre-simulation results, two types of information are extracted: safety constraint deviation and aggregate benefit change. Safety constraint deviation characterizes the degree of violation of the current adjustment scheme at each safety boundary, specifically including the magnitude of node voltage deviation from the allowable range, the margin of branch power flow exceeding the thermal stability limit, and the excess amount of protection maloperation risk exceeding the acceptable threshold. For each constraint dimension, the deviation change of the current iteration result relative to the previous iteration result is calculated. After normalizing the deviation changes of each dimension according to the corresponding safety weight coefficient, they are combined into a constraint violation gradient vector. The number of its components is equal to the total number of dimensions of the safety constraints. Each component of the constraint violation gradient vector reflects the rate of improvement in safety margin that can be obtained by continuing to adjust the decision variables along that constraint dimension; the larger the absolute value of the component, the higher the sensitivity of that dimension to adjustment.

[0109] The method for extracting the aggregated revenue change is as follows: calculate the sum of the expected revenues of each resource within the scheduling cycle under the current iteration scheme, and the difference between this sum and the sum of the revenues of the previous iteration scheme. Then, decompose this difference into the dimensions of each decision variable to obtain the revenue optimization gradient vector. Each component of the profit optimization gradient vector represents the incremental profit resulting from a unit adjustment in the direction of the corresponding decision variable. Positive values ​​indicate that adjusting in that direction is beneficial to improving profit, while negative values ​​have the opposite effect.

[0110] Will the constraint violate the gradient vector With the gradient vector of benefit optimization Weighted fusion is performed to form a composite gradient field. The fusion method employs an adaptive weighting mechanism: when the deviation from safety constraints is large, the fusion weights for constraints that violate gradients are adjusted. Automatic improvement prioritizes solutions that converge towards those satisfying safety constraints; once safety constraints are largely met, the fusion weights of the benefit optimization gradients are adjusted. The corresponding increase is made to further explore the potential for improving aggregation returns. Specifically, the composite gradient field is expressed as... ,in Both weights are non-negative real numbers, dynamically calculated based on the ratio of the current total deviation from the safety constraint to the preset safety tolerance band. The composite gradient field comprehensively reflects the trade-off between the degree of satisfaction of safety constraints and the direction of benefit optimization, providing a unified evaluation benchmark for subsequent convergence judgment.

[0111] Gradient convergence index The magnitude is obtained by jointly calculating the magnitude and the rate of change of direction of the composite gradient field. This reflects the remaining room for improvement in the current optimization; when the modulus approaches zero, it indicates that the solution is close to a local optimum. (Rate of change of direction) Defined as the angle between the directions of the composite gradient fields in two adjacent iterations, it is used to measure whether the optimization path tends to be stable, and is calculated as follows:

[0112] ;

[0113] superscript and Representing the current iteration and the previous iteration respectively, the gradient convergence index, which combines the magnitude and the rate of change of direction, is expressed as: ,in The weighting coefficient for the rate of change of direction is used to balance the relative importance of the two convergence signals: magnitude decay and directional stability.

[0114] When the gradient convergence index The convergence threshold is lower than the preset convergence threshold for several consecutive iterations. When the current adjusted aggregation scheduling scheme has reached Pareto optimality in terms of safety constraints and aggregation benefits, it is determined that the current scheme can no longer improve aggregation benefits without compromising the satisfaction of safety constraints, nor can it further improve safety margins without reducing aggregation benefits. Upon reaching this state, the current adjusted aggregation scheduling scheme is directly output as the optimized aggregation scheduling scheme, and the iteration loop terminates. If the gradient convergence index does not meet the convergence criterion, the gradient step size of the decision variables is updated according to the direction of the composite gradient field, generating a new round of adjustment schemes. The driving instruction sequence is recompiled, triggering the next round of virtual execution and parallel deduction. This cycle continues until the convergence condition is met. The entire iterative optimization process is executed entirely in the virtual environment of the digital twin model, without causing any actual disturbance to the distribution network physical system, ensuring the safety and repeatability of the optimization process.

[0115] In one optional implementation, based on the optimized aggregation scheduling scheme, control commands are issued to the distribution network physical system through a communication interface. The executed actual operating data is then transmitted back to the digital twin model via a state synchronization mechanism. Dynamic calibration of the mapping parameters of the digital twin model using a parameter identification algorithm includes:

[0116] The resource response timing, output allocation ratio, and reserve capacity configuration in the optimized aggregation scheduling scheme are parsed into a control instruction set, which is then sent to the distribution network physical system through a communication interface. The control instruction set carries a timing synchronization tag.

[0117] The power distribution network physical system performs power regulation according to the control command set, collects actual operating data, and transmits the actual operating data back to the digital twin model through the state synchronization mechanism. In the digital twin model, the actual operating data is aligned with the virtual execution prediction data based on the time synchronization label, and the power tracking error, time response error and state matching error are calculated to form a multi-dimensional identification error vector.

[0118] The multi-dimensional identification error vector is mapped to the mapping parameter space by a parameter identification algorithm, and a gradient relationship between the multi-dimensional identification error vector and the deviation of the mapping parameters is established. Based on the gradient relationship, the mapping parameter correction increment is calculated and superimposed on the current mapping parameter to complete the dynamic calibration of the mapping parameters of the digital twin model.

[0119] After completing virtual simulation and gradient convergence verification, the optimized aggregation scheduling scheme needs to parse the three core control parameters—resource response timing, power allocation ratio, and reserve capacity configuration—into a set of directly quantifiable control commands. During parsing, a timing synchronization tag is attached to each control command. This tag records the planned execution time, scheduling cycle number, and virtual simulation timestamp from the digital twin model, thus establishing a precise correspondence between physical execution and virtual prediction in subsequent data feedback. The control command set is sent to the distribution network physical system in message form via a communication interface. The communication interface supports standardized industrial communication protocols to ensure the integrity and real-time performance of the commands during transmission. For response timing commands, the message includes a start-up time deviation tolerance window; for power allocation commands, the message includes the power setpoints of each resource node and their allowable fluctuation ranges; for reserve capacity commands, the message includes reserve activation trigger conditions and upper and lower capacity limits.

[0120] After receiving the control command set, the distribution network physical system executes power regulation actions sequentially according to the timing specified in the commands. During execution, sensors deployed at each resource node and key measurement point collect actual operating data at a fixed sampling frequency, including the measured power injection of each node, bus voltage amplitude and phase angle, branch power flow, and the actual response time of each resource. After local preprocessing (removing outliers and completing short-term missing measurements), this actual operating data is packaged and transmitted back to the digital twin model through a state synchronization mechanism. The state synchronization mechanism adopts a combination of event triggering and periodic reporting: when a significant change in the state of a node is detected, reporting is triggered immediately, and the state of the entire network is synchronized in batches at fixed periods to ensure that the digital twin model can obtain the latest operating status of the physical system in a timely manner.

[0121] Before entering the identification process, the actual operational data transmitted back to the digital twin model undergoes time alignment based on timing synchronization tags. This alignment process matches each sampling point in the actual operational data with the virtual execution prediction data of the digital twin model at the corresponding time. If a slight deviation exists between the timestamp of the transmitted data and the virtual prediction timestamp due to communication delays, a linear interpolation method is used to resample the prediction sequence, ensuring strict alignment of the two sets of data on the time axis. After time alignment, errors are calculated in three dimensions: power tracking, timing response, and state matching. The power tracking error is defined as the difference between the measured power injection at each resource node and the virtual extrapolated predicted power value; the timing response error is defined as the difference between the actual response start time of each resource and the planned start time in the control command; and the state matching error is defined as the deviation vector between the measured and predicted values ​​of the voltage amplitude, phase angle, and branch power flow of the entire network nodes. These three types of errors are concatenated along the node and time dimensions to form a multi-dimensional identification error vector. Its dimension is equal to the product of the number of measurement points involved in alignment and the number of the three error types.

[0122] Parameter identification algorithms identify error vectors from multiple dimensions. The input is [value], and the goal is to find [solution] that makes [the solution] true. The mapping parameter correction amount that approaches the zero vector. Let the mapping parameter vector of the digital twin model be... This includes line impedance parameters, transformer turns ratio parameters, distributed resource equivalent model parameters, and load characteristic parameters. In the current mapping parameters... Nearby, by applying small parameter perturbations to the digital twin model and observing the corresponding changes in the predicted output, the identification error vector is numerically calculated. For the mapping parameter vector Jacobian matrix ,in The Liede The row element represents the first row. The error component affects the first... Partial derivatives of each mapping parameter. Based on the Jacobian matrix. Establish a linearized gradient relationship between the error vector and the parameter deviation: ,in This is the mapping parameter deviation vector.

[0123] Mapping parameter correction increment This is obtained by solving the following weighted least squares problem: ,in To identify the error weight matrix, larger weights are assigned to measurement points with higher measurement accuracy, and smaller weights are assigned to measurement points with lower accuracy or greater uncertainty. is the regularization coefficient, used to suppress excessive amplification of parameter corrections when the Jacobian matrix is ​​ill-conditioned; The identity matrix. Regularization coefficients. The condition number of the Jacobian matrix is ​​adaptively adjusted: when the condition number is large, it is increased. To enhance numerical stability, the condition number is reduced when it is small. To improve recognition accuracy.

[0124] The mapping parameter correction increment is calculated. Then, it is superimposed onto the current mapping parameter vector to complete one dynamic calibration iteration: ,in This is the calibrated mapping parameter vector. To prevent model parameter oscillations caused by excessively large single calibration steps, the parameters are adjusted before stacking. Upper and lower bound constraints are applied to each component to limit the magnitude of a single correction to each parameter within its physically reasonable range. The calibrated mapping parameters... Immediately update each sub-model of the digital twin model to make the simulation behavior of the digital twin model more closely resemble the real response characteristics of the distribution network physical system.

[0125] After dynamic calibration, the digital twin model re-executes the virtual simulation with the updated mapping parameters to verify the calibration effect. If the calibrated identification error vector... If the norm of the parameter identification is lower than the preset convergence threshold, the dynamic calibration is considered successful, and the updated mapping parameters are persistently stored for use in virtual simulations of subsequent scheduling cycles. If the identification error still exceeds the threshold, the next round of parameter identification iteration is triggered until convergence or the maximum number of iterations is reached. Through the above closed-loop calibration mechanism, the digital twin model can continuously track parameter drift caused by equipment aging, topology changes, or changes in operating conditions in the distribution network physical system, ensuring the long-term validity of the virtual simulation results, thereby providing reliable digital twin support for the continuous optimization of distributed resource aggregation scheduling.

[0126] A second aspect of this invention provides a distributed resource aggregation system for power distribution networks, comprising:

[0127] A model building unit is constructed to acquire real-time operational data of the distribution network physical system and distributed resource aggregation scheduling schemes to be executed, and a digital twin model of the distribution network is constructed based on the real-time operational data.

[0128] The parallel simulation unit is used to convert the aggregate scheduling scheme into the driving incentive of the digital twin model through control sequence parsing for virtual execution. It generates a set of multiple scenarios by propagating the uncertainty of new energy output fluctuations and load demand deviations. In the digital twin model, it performs parallel simulation to obtain multi-dimensional pre-simulation results including voltage over-limit probability distribution, power flow margin changes and protection maloperation risk.

[0129] An optimization convergence unit is used to construct a risk-benefit trade-off optimization framework based on the risk factors identified by the multi-dimensional pre-simulation results. This framework is then coupled with the solution of resource response timing, output allocation ratio, and reserve capacity configuration in the aggregated scheduling scheme to obtain an adjusted aggregated scheduling scheme. The adjusted aggregated scheduling scheme is then converted into a driving incentive for repeated virtual execution and parallel simulation to progressively update the multi-dimensional pre-simulation results. The scheme is continuously optimized through gradient convergence criteria until the safety constraints are met and the aggregated benefit is optimal. Finally, the optimized aggregated scheduling scheme is output.

[0130] The dynamic calibration unit is used to issue control commands to the distribution network physical system through a communication interface based on the optimized aggregation scheduling scheme, transmit the actual operation data after execution back to the digital twin model through a state synchronization mechanism, and dynamically calibrate the mapping parameters of the digital twin model through a parameter identification algorithm.

[0131] A third aspect of the present invention provides an electronic device, comprising:

[0132] processor;

[0133] Memory used to store processor-executable instructions;

[0134] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0135] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0136] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for distribution grid oriented distributed resource aggregation, characterized in that, include: Obtain real-time operational data of the distribution network physical system and distributed resource aggregation scheduling schemes to be executed, and construct a digital twin model of the distribution network based on the real-time operational data; The aggregated scheduling scheme is converted into the driving incentive of the digital twin model through control sequence parsing and then virtually executed. By propagating the uncertainty of new energy output fluctuations and load demand deviations, a set of multiple scenarios is generated. In the digital twin model, multi-dimensional pre-simulation results including voltage over-limit probability distribution, power flow margin changes and protection maloperation risk are obtained through parallel simulation. Based on the risk factors identified by the multi-dimensional pre-simulation results, a risk-benefit trade-off optimization framework is constructed to couple the solution of resource response timing, output allocation ratio and reserve capacity configuration in the aggregated scheduling scheme, resulting in an adjusted aggregated scheduling scheme. The adjusted aggregated scheduling scheme is then converted into driving incentives for repeated virtual execution and parallel simulation to progressively update the multi-dimensional pre-simulation results. The scheme is continuously optimized through gradient convergence criteria until the safety constraints are met and the aggregated benefit is optimal, and the optimized aggregated scheduling scheme is output. Based on the optimized aggregation scheduling scheme, control commands are issued to the distribution network physical system through the communication interface. The actual operation data after execution is transmitted back to the digital twin model through the state synchronization mechanism. The mapping parameters of the digital twin model are dynamically calibrated through the parameter identification algorithm.

2. The method of claim 1, wherein, The aggregated scheduling scheme is converted into driving stimuli of the digital twin model through control sequence parsing for virtual execution. A multi-scenario set is generated by propagating the uncertainties of new energy output fluctuations and load demand deviations, including: The resource scheduling instructions in the aggregation scheduling scheme are deconstructed in time sequence, and the power setting trajectory and response timing constraints of each distributed resource are extracted to construct a multi-dimensional control sequence matrix. An incentive-response mapping function is established based on the multidimensional control sequence matrix, and the control quantities in the multidimensional control sequence matrix are converted into incentive signals for the corresponding resource nodes in the digital twin model. The incentive signals drive the digital twin model to perform state evolution to achieve virtual execution. During the virtual execution process, historical samples of new energy output fluctuations and statistical distributions of load demand deviations are collected to construct a joint uncertainty probability space; In the joint uncertainty probability space, a set of perturbation vectors covering extreme operating conditions is generated by Latin hypercube sampling. Each perturbation vector in the set of perturbation vectors is superimposed on the excitation signal to form a differentiated excitation scenario, driving the digital twin model to evolve in parallel to generate a set of multiple scenarios.

3. The method of claim 2, wherein, In the joint uncertainty probability space, a set of perturbation vectors covering extreme operating conditions is generated through Latin hypercube sampling. The perturbation vectors in this set are then superimposed onto the excitation signal to form differentiated excitation scenarios, including: In the joint uncertainty probability space, an orthogonal mesh is constructed with the dimensions of new energy output fluctuation and load demand deviation. Based on the boundary cells of the orthogonal mesh, the boundary subspace corresponding to extreme operating conditions is located. A spatial sampling scheme is generated by configuring sampling density weights in the boundary subspace and the central region respectively through a hierarchical spatial filling strategy. According to the spatial sampling scheme, a set of sampling points with spatial uniform distribution characteristics is generated in the joint uncertainty probability space. The coordinate values ​​of each sampling point in the set of sampling points in the dimension of new energy output fluctuation are mapped into the new energy output disturbance amplitude, and the coordinate values ​​of each sampling point in the dimension of load demand deviation are mapped into the load demand deviation amplitude, thereby constructing a set of disturbance vectors. For each disturbance vector in the disturbance vector set, the disturbance amplitude of the new energy output is added to the power incentive benchmark value of the corresponding new energy node in the incentive signal, and the disturbance amplitude of the load demand deviation is added to the demand incentive benchmark value of the corresponding load node in the incentive signal, thereby generating a differentiated incentive scenario set that includes deterministic scheduling instructions and uncertain environmental disturbances.

4. The method of claim 1, wherein, Based on the risk factors identified by the multi-dimensional pre-simulation results, a risk-benefit trade-off optimization framework is constructed to couple the resource response timing, output allocation ratio, and reserve capacity configuration in the aggregated scheduling scheme, resulting in the adjusted aggregated scheduling scheme, including: The distribution characteristics of power imbalance, voltage deviation and frequency deviation under each scenario are deconstructed from the multi-dimensional pre-simulation results. Based on the distribution characteristics, the risk severity index of each scenario is calculated. The operating conditions corresponding to the scenario whose risk severity index exceeds the safe operation threshold are abstracted as risk factors. The risk factors include a triplet description of triggering conditions, affected resource identifiers and deviation magnitude. The probability of risk occurrence is quantified based on the triggering conditions, the amount of economic loss is assessed based on the magnitude of the deviation, a risk cost item is formed by the composite operation of the probability of risk occurrence and the amount of economic loss, the expected benefit of the aggregation scheduling scheme is set as the benefit return item, and a risk-benefit trade-off optimization framework including the risk cost item and the benefit return item is established. In the risk-reward trade-off optimization framework, the relationship between the impact resource identifier and the response time-series variable, output allocation ratio variable, and reserve capacity configuration variable of the corresponding resource in the aggregated scheduling scheme is established. Based on the relationship, a complementary constraint relationship is established between the advance adjustment amount of the response time-series variable and the increase amount of the reserve capacity configuration variable. Through multi-variable collaborative iterative solution, a variable adjustment scheme that reduces the risk cost item and maintains the benefit return item is obtained. Based on the variable adjustment scheme, the aggregated scheduling scheme is reconstructed to obtain the adjusted aggregated scheduling scheme.

5. The method of claim 4, wherein, The probability of risk occurrence is quantified based on the triggering conditions, the economic loss is assessed based on the magnitude of the deviation, a risk cost item is formed through a composite calculation of the probability of risk occurrence and the economic loss, and the expected benefits of the aggregated scheduling scheme are set as a return item, including: Using the time dimension characteristics, load fluctuation characteristics, and meteorological condition characteristics in the triggering conditions as input variables, a multi-dimensional mapping relationship is established based on the input variables. Through the multi-dimensional mapping relationship, the frequency of historical risk events corresponding to different combinations of triggering conditions is converted into probability quantification values ​​to obtain the probability of risk occurrence. Based on the power deviation magnitude, duration, and impact range in the deviation magnitude, and combined with the unit capacity cost and default penalty coefficient of the resources involved in the aggregation scheduling scheme, a loss transmission path is constructed to correlate and solve the direct and indirect economic losses caused by a single risk event, thereby obtaining the amount of economic loss. The risk cost term is obtained by performing a tensor product operation between the probability of the risk occurrence and the amount of economic loss, and by introducing a time discount factor to attenuate the risk cost in the future period. By tracing the expected output timing, market transaction electricity price, and ancillary service compensation price of each resource from the aggregated scheduling scheme, the expected total revenue of each resource within the scheduling cycle is calculated to obtain the revenue return item.

6. The method of claim 1, wherein, The adjusted aggregation scheduling scheme is converted into driving incentives for repeated virtual execution and parallel inference to progressively update the multi-dimensional pre-simulation results. It is then continuously optimized using a gradient convergence criterion until safety constraints are met and the aggregation benefit is optimal. The optimized aggregation scheduling scheme is output as follows: The resource response timing, output allocation ratio and reserve capacity configuration in the adjusted aggregation scheduling scheme are compiled into a driving instruction sequence. The driving instruction sequence is used to stimulate virtual execution and parallel simulation to re-execute the pre-simulation process and generate progressively updated multi-dimensional pre-simulation results. From the progressively updated multi-dimensional pre-simulation results, the safety constraint deviation and the aggregated benefit change are extracted. A constraint violation gradient vector is constructed based on the safety constraint deviation, and a benefit optimization gradient vector is constructed based on the aggregated benefit change. The constraint violation gradient vector and the benefit optimization gradient vector are weighted and fused to form a composite gradient field. The gradient convergence index is calculated using the magnitude and direction change rate of the composite gradient field. When the gradient convergence index meets the convergence criterion, it is determined that the adjusted aggregation scheduling scheme has reached the Pareto optimal state of safety constraints and aggregation benefits. The current adjusted aggregation scheduling scheme is then output as the optimized aggregation scheduling scheme.

7. The method of claim 1, wherein, Based on the optimized aggregation scheduling scheme, control commands are issued to the distribution network physical system through a communication interface. The executed actual operating data is then transmitted back to the digital twin model via a state synchronization mechanism. Dynamic calibration of the mapping parameters of the digital twin model is performed using a parameter identification algorithm, including: The resource response timing, output allocation ratio, and reserve capacity configuration in the optimized aggregation scheduling scheme are parsed into a control instruction set, which is then sent to the distribution network physical system through a communication interface. The control instruction set carries a timing synchronization tag. The power distribution network physical system performs power regulation according to the control command set, collects actual operating data, and transmits the actual operating data back to the digital twin model through the state synchronization mechanism. In the digital twin model, the actual operating data is aligned with the virtual execution prediction data based on the time synchronization label, and the power tracking error, time response error and state matching error are calculated to form a multi-dimensional identification error vector. The multi-dimensional identification error vector is mapped to the mapping parameter space by a parameter identification algorithm, and a gradient relationship between the multi-dimensional identification error vector and the deviation of the mapping parameters is established. Based on the gradient relationship, the mapping parameter correction increment is calculated and superimposed on the current mapping parameter to complete the dynamic calibration of the mapping parameters of the digital twin model.

8. A distribution grid oriented distributed resource aggregation system for implementing the method of any of claims 1-7, characterized by, include: A model building unit is constructed to acquire real-time operational data of the distribution network physical system and distributed resource aggregation scheduling schemes to be executed, and a digital twin model of the distribution network is constructed based on the real-time operational data. The parallel simulation unit is used to convert the aggregate scheduling scheme into the driving incentive of the digital twin model through control sequence parsing for virtual execution. It generates a set of multiple scenarios by propagating the uncertainty of new energy output fluctuations and load demand deviations. In the digital twin model, it performs parallel simulation to obtain multi-dimensional pre-simulation results including voltage over-limit probability distribution, power flow margin changes and protection maloperation risk. An optimization convergence unit is used to construct a risk-benefit trade-off optimization framework based on the risk factors identified by the multi-dimensional pre-simulation results. This framework is then coupled with the solution of resource response timing, output allocation ratio, and reserve capacity configuration in the aggregated scheduling scheme to obtain an adjusted aggregated scheduling scheme. The adjusted aggregated scheduling scheme is then converted into a driving incentive for repeated virtual execution and parallel simulation to progressively update the multi-dimensional pre-simulation results. The scheme is continuously optimized through gradient convergence criteria until the safety constraints are met and the aggregated benefit is optimal. Finally, the optimized aggregated scheduling scheme is output. The dynamic calibration unit is used to issue control commands to the distribution network physical system through the communication interface based on the optimized aggregation scheduling scheme, transmit the actual operation data after execution back to the digital twin model through the state synchronization mechanism, and dynamically calibrate the mapping parameters of the digital twin model through the parameter identification algorithm.

9. An electronic device, comprising: include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images