Ramp demand node mapping method based on multi-source contribution degree allocation

Abstract

The application discloses a method for mapping climbing demand nodes based on multi-source contribution degree distribution, and relates to the technical field of power system operation and control. The method comprises the following steps: obtaining initial weights of source and load nodes and system-level climbing demand; constructing a multi-dimensional joint distribution model reflecting the correlation of fluctuations of heterogeneous resources by using a Vine Copula function, and calculating dynamic contribution degree coefficients; performing spatial mapping calculation according to the dynamic contribution degree coefficients to obtain an expected power change component; constructing a cooperative game model, quantifying marginal contributions to network congestion elimination by using Shapley values, and distributing capacity relaxation; and finally generating a mapping scheme meeting node physical constraints. The application is used to solve the problems of mapping precision distortion under a multi-source random fluctuation environment and the difficulty of existing schemes in considering network congestion constraints.

Description

Technical Field

[0001] This invention relates to the field of power system operation and control technology, and more specifically, to a method for mapping ramp demand nodes based on multi-source contribution allocation. Background Technology

[0002] The deepening construction of new power systems and the large-scale integration of high-proportion, volatile power sources such as wind and solar power have significantly increased the randomness and volatility of the system's net load. To address the risk of instantaneous power imbalance, power dispatching agencies urgently need to build a precise ramp-up demand mapping mechanism to scientifically decompose the system-level macro-ramp-up demand to each source-load node, so as to achieve refined allocation and safe placement of the network's flexible resources.

[0003] Currently, most mainstream ramp demand allocation technologies employ allocation methods based on static weights or linear proportional factors, which establish a mapping relationship between node capacity proportions and total system demand. Some improved schemes introduce conventional correlation analysis models to attempt to characterize the correlation between renewable energy output and system fluctuations, and use static sensitivity analysis to guide output adjustments at the node level.

[0004] However, existing technologies have significant limitations in handling the complex fluctuations of massive heterogeneous resources. On the one hand, traditional linear correlation models struggle to capture the nonlinear characteristics and high-order tail correlations of heterogeneous resources in multidimensional space, leading to significant distortions in mapping results under extreme weather or sudden operating conditions, and failing to accurately quantify the dynamic contribution of each source-load node to system demand. On the other hand, existing demand allocation schemes lack deep coupling with real-time physical constraints of the power grid, often ignoring the dynamic congestion risks of critical transmission sections. Furthermore, in the allocation logic of node-side flexibility relaxation, due to the lack of a scientific marginal benefit quantification mechanism, allocation schemes often fail to balance global security and physical feasibility, easily triggering voltage or frequency safety thresholds, and failing to construct a closed-loop mapping system from multi-source correlation perception to spatial congestion avoidance. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for mapping climbing demand nodes based on multi-source contribution allocation. This method utilizes the Vine Copula function to capture the higher-order fluctuation correlation of heterogeneous resources and coordinates with the Shapley value game model to quantify the marginal contribution of nodes to eliminate blockages. This addresses the accuracy distortion and the security bottleneck issues that traditional mapping methods face in multi-source random fluctuation environments, as well as the difficulty in simultaneously considering the physical constraints of the grid structure.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The method for mapping climbing demand nodes based on multi-source contribution allocation includes the following steps: obtaining the initial weights of each source-load node in the power network and determining the system-level climbing demand of the power network; constructing a multidimensional joint distribution model reflecting the correlation of fluctuations in various heterogeneous resources using the Vine Copula function, and calculating the dynamic contribution coefficient of each heterogeneous resource to the system-level climbing demand based on the multidimensional joint distribution model; performing spatial mapping calculation based on the initial weights and the dynamic contribution coefficients to obtain the expected power change components of each source-load node; constructing a cooperative game model, using Shapley values ​​to quantify the marginal contribution of each source-load node to eliminating network congestion, and allocating the climbing capacity relaxation amount of each source-load node accordingly; and generating a climbing demand mapping scheme that satisfies the physical constraints of the nodes based on the expected power change components and the climbing capacity relaxation amount.

[0007] In a preferred embodiment, obtaining the initial weights of each source-load node in the power network includes: obtaining the real-time topology feature matrix of the power network using a graph attention neural network; identifying key transmission sections in the topology feature matrix; and correcting the weights of downstream nodes of restricted branches based on branch load rates to obtain the initial weights.

[0008] In a preferred embodiment, the construction of a multidimensional joint distribution model reflecting the correlation of fluctuations in heterogeneous resources includes: extracting the prediction error sequence of each heterogeneous resource in historical operational data under a preset time period, and establishing the marginal distribution function of each sequence; determining the Vine structure that conforms to the current climate characteristics, and fitting the marginal distribution function by conditional probability iteration to obtain the Vine Copula joint distribution function that characterizes the higher-order tail correlation of heterogeneous resources.

[0009] In a preferred embodiment, the spatial mapping calculation includes: applying the dynamic contribution coefficient as a correction factor to the original mapping equation of each source load node; and aggregating various resource fluctuations on the same physical bus using vector summation to generate the expected net power fluctuation value of the node under the system ramp-up command.

[0010] In a preferred embodiment, quantifying the marginal contribution of each source load node to eliminating network congestion using the Shapley value includes: defining the nodes participating in the ramp response as a set of game participants, and establishing a game characteristic function with the overall network congestion payoff as the objective; calculating the congestion elimination increment caused by the target node joining different node combination subsets, and obtaining the Shapley value of the node by weighted averaging.

[0011] In a preferred embodiment, the allocation of ramp capacity slack for each source load node includes: spatially distributing the total ramp capacity slack of the system using the Shapley value of each source load node as a weighting factor; verifying whether the distributed node slack exceeds the adjustment limit of the corresponding node; if it does, then redistributing the remaining portion among the unsaturated nodes.

[0012] In a preferred embodiment, the process of generating the mapping scheme further includes: verifying whether the sum of the expected power change components and the allocated relaxation amounts of all source load nodes is consistent with the system-level ramping requirements; if not, updating the dynamic contribution coefficient through a negative feedback adjustment loop until the system balance constraint is met.

[0013] In a preferred embodiment, the method further includes: monitoring whether the grid state corresponding to the mapping scheme triggers a voltage or frequency safety threshold; if triggered, calculating the sensitivity of each source-load node to voltage exceedance using a node voltage sensitivity matrix, and adjusting the expected power change component accordingly.

[0014] In a preferred embodiment, the calculation of the dynamic contribution coefficient further includes: when the fluctuation characteristics of the system-level ramping demand meet the preset triggering conditions, switching to the enhanced distribution model; and increasing the weight ratio of resource nodes with volatility exceeding a preset threshold in the enhanced distribution model.

[0015] In a preferred embodiment, the method further includes: performing power dispatch according to the ramp demand mapping scheme; obtaining feedback results after the dispatch is performed; and using a reinforcement learning algorithm to update the model parameters of the VineCopula function online based on the feedback results.

[0016] The technical effects and advantages of the climbing demand node mapping method based on multi-source contribution allocation in this invention are as follows: This invention utilizes the Vine Copula function to construct a multidimensional joint distribution model, accurately capturing the high-order tail correlations among heterogeneous resource fluctuations in power networks. This effectively solves the accuracy distortion problem of traditional linear mapping methods when dealing with complex randomness, significantly improving the accuracy of dynamic contribution coefficient calculation. Furthermore, it collaboratively introduces a cooperative game model based on Shapley values, scientifically quantifying the marginal contribution of each source-load node to eliminating network congestion, ensuring the optimal allocation of ramp capacity slack in geographic space. Finally, by integrating initial weights, dynamic contribution, and physical constraints, a spatial mapping scheme that balances the flexibility requirements of the entire network with the safety boundaries of the grid structure is generated. This not only enhances the physical executability of dispatch instructions but also significantly improves the congestion avoidance capability and operational stability of the new power system in a multi-source complex fluctuation environment. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the climbing demand node mapping method based on multi-source contribution allocation provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the evolution curve of the dynamic contribution coefficient provided in an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the convergence curve of online update of model parameters based on the PPO algorithm provided in an embodiment of the present invention. Detailed Implementation

[0020] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1, Figure 1 The present invention provides a method for mapping climbing demand nodes based on multi-source contribution allocation, comprising the following steps: S1, obtain the initial weights of each source-load node in the power network, and determine the system-level ramping requirements of the power network.

[0022] In this embodiment, the system first obtains the operating benchmark of each source and load node in the power network through the data acquisition module, and determines the overall flexibility requirements of the system based on the deviation between load forecast and new energy output forecast.

[0023] Specifically, obtaining the initial weights of each source-load node includes: capturing the real-time topology feature matrix of the power network using a graph attention neural network (GAT); identifying key transmission sections in the topology feature matrix; and correcting the weights of downstream nodes of restricted branches based on branch load rates to obtain the initial weights. In this embodiment, the IEEE 39-node system is used as the research object, and the bus admittance matrix of the power system and the real-time power flow distribution data of each node (such as node voltage amplitude, phase angle, active / reactive power output) are used as input features of the graph attention neural network. The network input layer weights the adjacency matrix through a multi-head attention mechanism, and the hidden layer is set to 3 layers with 64 neurons in each layer. The LeakyReLU activation function is used to enhance the ability to capture nonlinear topology features. The output layer of the graph attention neural network is connected to a fully connected layer, and the output vector is normalized by the Softmax function to obtain the initial weights of each source-load node. Furthermore, the training sample set of this model originates from the marginal contribution data of nodes under the historical optimal scheduling scenario of the power system. Through supervised learning, the model has the ability to quickly map the importance of nodes.

[0024] During the identification of critical transmission sections, the system monitors the operating status of each branch in real time, marking branches with load rates exceeding the 90% warning threshold as restricted branches, and identifying downstream affected nodes of the restricted branches based on the power flow direction. To effectively mitigate potential congestion risks, the system employs an exponential decay function as a weight correction operator to reweight the original weights of the downstream affected nodes, shifting the adjustment task to unrestricted areas. In this embodiment, the decay adjustment coefficient... The value is set to 1.2 (or selected from the empirical range of 1.0 to 1.5) to balance the sensitivity of task transfer with the stability of system operation.

[0025] The formula for correcting the initial weights is as follows: (1) in, For nodes Corrected initial weights For nodes The original weights, The preset attenuation adjustment coefficient, To correspond to the real-time load rate of the restricted branch, The load rate threshold is set to 0.9 in this embodiment.

[0026] Meanwhile, system-level ramp-up and ramp-down requirements and This represents the maximum net load forecast deviation as assessed. The system obtains the net load forecast deviation within a preset time period and uses it as the basis for quantifying the upward and downward adjustment of system ramp-up demand. The confidence level used to quantify forecast uncertainty is preferably 95%, to balance the economy of adjusting resources with the system's safety in responding to extreme fluctuations.

[0027] The formula for calculating the system-level ramp-up and ramp-down requirements is as follows: (2) in, To meet the system-level ramp-up requirements, For system-level ramping requirements, and The first Time period and the Load forecast values ​​for the time period and These represent the sum of the projected output of various new energy sources for the corresponding time period. To adjust the forecast error quantile at the 95% confidence level, The down-adjusted prediction error quantile at the 95% confidence level.

[0028] To verify the effectiveness of the weight adjustment strategy, Table 1 presents simulated weight transfer data for some key nodes in the IEEE 39-node system before and after section blocking.

[0029] Table 1

[0030] As shown in Table 1, when an overload trend is detected in the branches where Node 16 and Node 19 are located ( When the system uses an exponential decay function to significantly strip away its weights, it shifts the center of gravity of the climbing response to unrestricted areas such as Node 12.

[0031] This step, through dynamic topology sensing and net load deviation assessment, solves the problems that static weights cannot adapt to real-time power flow changes in the power grid and that there is insufficient basis for demand quantification, ensuring the consistency of weight allocation with the physical grid status and actual regulation needs.

[0032] S2. A multidimensional joint distribution model reflecting the correlation of fluctuations of various heterogeneous resources is constructed using the Vine Copula function, and the dynamic contribution coefficient of each heterogeneous resource to the system-level ramping demand is calculated based on the multidimensional joint distribution model.

[0033] In this embodiment, the system further performs collaborative modeling of the random fluctuation characteristics of wind power, photovoltaics, and load.

[0034] Specifically, a multidimensional joint distribution model is constructed, including: extracting prediction error sequences of various heterogeneous resources from historical operational data under a preset time period, and establishing marginal distribution functions for each sequence; determining a Vine structure that conforms to current climate characteristics, and fitting a Vine Copula joint distribution function that characterizes the high-order tail correlation of heterogeneous resources by performing conditional probability iteration on the marginal distribution function. In this embodiment, the system extracts power prediction error data of each node with a sampling interval of 15 minutes from the historical database over the past year to form a high-dimensional time series sample. To accurately characterize the non-normal, peak-heavy-tailed characteristics of wind and solar load fluctuations, the system uses the non-parametric kernel density estimation (KDE) method to establish marginal distribution functions for the fluctuations of various heterogeneous resources. In this embodiment, the kernel function of KDE is preferably a Gaussian kernel function, and its bandwidth is automatically selected based on the Silverman reference rule to achieve an optimal balance between the smoothness of the distribution and the capture of local details.

[0035] When determining a Vine structure that conforms to the current climate characteristics and fitting parameters, the system adjusts the topology in real time based on meteorological monitoring data: In extreme weather conditions such as typhoons and cold waves with obvious geographic center characteristics, the system selects a C-vine structure (Canonical Vine) with a central node as the core; under normal weather conditions, a D-vine structure (Drawable Vine) is used to describe the linear chain-like correlation between nodes evolving with geographic or electrical distance. When selecting a pair-copula family of Vine structures, for resource pairs with strong tail correlations such as wind-solar fluctuations, the pair-copula is preferably a Gumbel Copula (used to capture upper tail correlations) or a t-copula (used to capture symmetric thick-tail correlations).

[0036] Furthermore, regarding the handling of extreme fluctuations, the calculation of the dynamic contribution coefficient also includes: when the fluctuation characteristics of the system-level ramp-up demand meet a preset trigger condition, switching to an enhanced distribution model; and increasing the weight ratio of resource nodes whose volatility exceeds a preset threshold in the enhanced distribution model. Specifically, this embodiment uses the absolute value of the second derivative of the total system load fluctuation as the trigger value; when this value exceeds... When the system reaches its preset extreme value, it is determined that it has entered an extreme ramp-up condition. At this time, the system automatically switches to the enhanced distribution model, and introduces a weight adjustment operator to process resource nodes whose volatility exceeds a preset threshold (such as 20% of the rated capacity). Specifically, under the enhanced distribution model, the correlation parameters of the corresponding Pair-Copula in the Vine structure are adjusted. This increases the Kendall rank correlation coefficient between the specific resource node and other nodes to 1.5 times the original value, thereby significantly improving the model's early warning sensitivity to sudden, high-intensity coordinated fluctuation risks.

[0037] The probability density calculation formula for the Vine Copula joint distribution function is as follows: (3) in, Let n be the joint probability density function of the fluctuations of various heterogeneous resources. For the first The marginal probability density function of a resource For the corresponding Pair-Copula density function, For based on The conditional probability distribution function.

[0038] The dynamic contribution coefficient of each heterogeneous resource to the climbing demand The calculation formula is as follows: (4) in, Represents the dynamic contribution coefficient for the corresponding type. For the first Power fluctuation components of each resource node The joint distribution function of the Vine Copula is... The integral region is used to meet the system-level ramp-up requirements triggering conditions, where j is the resource index. .

[0039] To solve the aforementioned high-dimensional integral formula, this embodiment employs a Monte Carlo method based on Latin hypercube sampling (LHS): First, generate... A group of random sample points conforming to the Vine Copula joint distribution is statistically analyzed and calculated to identify those falling into the integral region that meets the climbing requirement triggering condition. The sample mean within the range is used to quickly approximate the solution for the contribution coefficient of each resource.

[0040] To visually demonstrate the differences in the response of the multidimensional joint distribution model under different climatic characteristics, Figure 2 The evolution curve of the dynamic contribution coefficient is shown, which is fitted based on historical operation data of a provincial power grid.

[0041] like Figure 2 As shown, under normal operating conditions (corresponding to the smooth fluctuation zone in the figure), the contribution coefficients of wind, solar, and load are... The system maintains a relatively stable mean range; however, when the system encounters extreme weather events such as typhoons (corresponding to the surge area at t=45 minutes in the figure) and triggers the enhanced distribution model, the conditional prediction error of wind turbine nodes is captured by the higher-order tail correlation of Vine Copula, and its dynamic contribution coefficient instantly spikes to over 0.65. This indicates that under extreme operating conditions, the "traction" effect of the sudden drop in wind turbine output on the overall system ramp-up demand is significantly amplified.

[0042] This step, by introducing the Vine Copula model from the financial field and combining it with enhanced distribution adjustment, accurately quantifies the "simultaneous increase and decrease" effect of heterogeneous resources under extreme operating conditions, thus solving the technical problem of inaccurate division of responsibility for ramp-up demand.

[0043] S3. Based on the initial weights and the dynamic contribution coefficients, spatial mapping calculations are performed to obtain the expected power change components of each source-load node.

[0044] In this embodiment, the system maps macroscopic system-level ramping requirements to specific physical bus levels. By establishing a correlation matrix between the power changes of each node and the overall network flexibility requirements, the system achieves precise decomposition of power in geographic space.

[0045] Specifically, spatial mapping calculation includes applying the dynamic contribution coefficient as a correction factor to the original mapping equations of each source-load node. During the calculation process, to ensure that the total power after mapping strictly matches the total system demand, this embodiment specifies the initial weights of each source-load node. This represents the relative weight of the node within its corresponding resource set. For example, for a load node set, it satisfies... Similarly, for a set of photovoltaic or wind power nodes, the sum of their internal weights is 1.

[0046] The formulas for calculating the expected power variation components of each source load node are as follows: Expected ramp-up component at load nodes: (5) in, For the first The expected change in ramp-up power at each load node For nodes Corrected initial weights The dynamic contribution coefficient for the load category's upward slope. This is for system-level ramp-up requirements.

[0047] Expected ramp-up component of photovoltaic nodes: (6) in, For the first The expected change in ramp-up power for each photovoltaic node. This is the normalized initial weight of the photovoltaic node among all photovoltaic nodes. The dynamic contribution coefficient for the uphill climb of photovoltaic products.

[0048] Expected uphill component of wind power node: (7) in, For the first The expected uphill power change of each wind power node This represents the normalized initial weight of the wind turbine node among all wind turbine nodes. This is the dynamic contribution coefficient for wind power's uphill climbing.

[0049] Furthermore, the system aggregates various resource fluctuations on the same physical bus using a vector summation method. For integrated multi-energy complementary nodes that include wind power, photovoltaics, and loads, since different types of resource fluctuations may exhibit reverse cancellation characteristics over time, the system uses an internal cancellation mechanism to calculate the expected net power fluctuation of the physical node under the system ramp-up command. .

[0050] The physical bus Net power The mapping formula is as follows: (8) in, For the first The expected ramp-up power variation component of photovoltaic sub-type resources belonging to each physical bus; For the first The expected uphill power variation component of the wind electron resources belonging to each physical bus.

[0051] In this embodiment, it is specified that This indicates that the node lacks flexibility in the current time period and needs external support for upward climbing; conversely, if This indicates that the resource fluctuations of that node are redundant, providing a certain adjustment margin for the system.

[0052] This step, through the coupling calculation of normalized weight allocation and multidimensional factors, achieves accurate mapping and power balance of demand in geospatial space, significantly reduces the uncertainty caused by heterogeneous resource fluctuations, and provides a data foundation for subsequent congestion management.

[0053] S4. Construct a cooperative game model, use Shapley values ​​to quantify the marginal contribution of each source load node to eliminating network congestion, and allocate ramp capacity relaxation amount to each source load node accordingly.

[0054] In this embodiment, to optimize the distribution of redundancy across the entire network and ensure that the mapping scheme balances security and fairness, a cooperative game theory mechanism is introduced. First, the system assesses the safety margin required to maintain stable system operation based on the total number of limit exceedances at critical sections across the entire network and their safety thresholds, defining this margin as the system's total ramp capacity relaxation. .

[0055] Specifically, the Shapley value is used to quantify the marginal contribution of each source-load node to eliminating network congestion. This includes: defining the nodes participating in ramp-up response as a set of game participants and establishing a game characteristic function with the overall network congestion payoff as the objective; calculating the congestion elimination increment caused by adding the target node to different node combination subsets, and obtaining the Shapley value of the node through weighted averaging. The system defines the 10 in-service generator units participating in peak shaving and frequency regulation within the IEEE 39-node system and the critical load centers with demand response capabilities as the set of game participants. In this context, it is clear that load centers participate in the game through their adjustable loads, and if they do not have demand response capabilities, they cannot make a marginal contribution to eliminating system congestion.

[0056] Game characteristic function Set as a subset of nodes The reduction or exemption of penalties for exceeding limits at key transmission sections is achieved through coordinated adjustment of ramp output. In this embodiment, the characteristic function... The calculation is based on each node With its rated maximum climbing rate The system responds and, in conjunction with the power flow sensitivity matrix (PTDF), calculates the contribution of each node's output change to the voltage drop at the congestion-prone section, thereby obtaining the expected benefit of congestion elimination. For example, if the power flow at a critical 500kV section in the system exceeds its thermal stability limit of 2000MW, the system will calculate the congestion cost in real time at a unit price of 5000 yuan per megawatt; the characteristic function quantifies the node combination through the above sensitivity calculation. The economic benefits generated after responding to the ramp-up command.

[0057] Based on this, the ramp capacity slack is allocated to each source load node, including: spatially distributing the total ramp capacity slack of the system using the Shapley value of each source load node as a weighting factor. The system utilizes the marginal contribution reflected by the Shapley value to determine the total ramp capacity slack of the entire network. The capacity is distributed to each physical node, allowing nodes that contribute significantly to system security to receive higher capacity reservations.

[0058] During the verification process, the system introduces physical constraints for each source load node, such as the maximum output limit of thermal power units. For 600MW, minimum output limit The capacity is 150MW, and the rated ramp rate is 12MW / min. If the slack after allocation causes a unit to need to cross its maximum regulation limit in the next time period, the saturation overflow of that node will be extracted and then weighted again according to the proportion of Shapley values ​​of the other unsaturated nodes (i.e., nodes with regulation margins) until the constraints of the entire network are met.

[0059] The characteristic function The calculation formula is as follows: (9) in, For nodes For restricted branches The power transmission distribution factor (power flow sensitivity coefficient) is used to characterize the degree of influence of nodal output changes on the power flow of the cross section. To block the unit price, and The maximum value shall not exceed the total over-limit cost of that section.

[0060] The Shapley value of each node is calculated using the following formula: (10) in, For nodes Shapley value, The total number of participants in the game. For nodes not included Any subset of nodes, For nodes Join a subset The resulting blockage eliminates the incremental benefit.

[0061] The relaxation amount allocated to each node The calculation formula is as follows: (11) in, For nodes In the region The ramp capacity relaxation amount obtained from internal allocation. For nodes Shapley value, The sum of the Shapley values ​​of all participating nodes in this region. This represents the total ramp capacity relaxation of the system.

[0062] To visually demonstrate the game-theoretic allocation effect, Table 2 lists the marginal contribution quantification and relaxation allocation results of three typical peak-shaving units in a simulation example.

[0063] Table 2

[0064] Based on the calculation data in Table 2, it can be seen that although the output cost of unit G1 is relatively high, the system prioritizes allocating the most relaxation margin to it because its electrical position has the greatest marginal contribution to eliminating the blockage of the critical section (the Shapley value accounts for 45%). At the same time, for the part of the physical constraint overflow triggered by unit G2, the system accurately completed the secondary transfer between unsaturated nodes such as G3.

[0065] This step resolves the mismatch between the distribution of flexible resources and the demand for grid congestion through Shapley value game theory, achieving Pareto optimal allocation of the entire grid's slack, and fully considering the physical regulation characteristics of the resources on the source and load sides.

[0066] S5. Based on the expected power change component and the ramp capacity relaxation, generate a ramp demand mapping scheme that satisfies the node physical constraints.

[0067] In this embodiment, the system ultimately generates physically executable scheduling instructions and enters the closed-loop optimization stage to address real-time fluctuations and security risks in power system operation.

[0068] Specifically, the process of generating the mapping scheme also includes: verifying whether the sum of the expected power change components of all source load nodes and the allocated relaxation amount is consistent with the system-level ramping requirements; if not, updating the dynamic contribution coefficient through a negative feedback adjustment loop until the system balance constraint is met.

[0069] The system performs closed-loop balancing in the central control unit via a proportional-integral (PI) controller. The PI controller distributes the deviation... As the input signal, its scaling factor The value is 0.2 0.5, integral coefficient The value is 0.01 0.05. The system uses the output scalar of the PI controller as a gain coefficient to apply to the dynamic contribution coefficient. The magnitude of the vector is determined until the system equilibrium constraints are satisfied.

[0070] Furthermore, the system monitors whether the grid state corresponding to the mapping scheme triggers voltage or frequency safety thresholds. For frequency safety risks, a droop control formula is used for power correction. (12) in, For the first Frequency-assisted adjustment components generated by each node For the first The frequency response coefficient (i.e., droop coefficient) of each node. and These are the real-time monitoring system frequency and the system reference frequency (50Hz in this embodiment).

[0071] If a voltage or frequency safety threshold is triggered, the sensitivity of each source-load node to voltage exceedance is calculated using the node voltage sensitivity matrix, and the expected power change component is adjusted accordingly. The node voltage sensitivity correction calculation formula is as follows: (13) in, and These represent the expected power change components at the nodes before and after the correction. busbar voltage to node Sensitivity coefficient of active power, busbar Real-time voltage per unit value, This corresponds to the voltage safety boundary value; specifically, when hour, Take 1.05, when hour, Take 0.95. This is the preset correction step size, and using this symbol effectively avoids confusion with the load rate symbol.

[0072] Finally, the system performs power dispatch according to the aforementioned ramp-up demand mapping scheme; obtains the feedback results after dispatching, and updates the model parameters of the Vine Copula function online using a reinforcement learning algorithm (PPO). The observation state space of the PPO algorithm... This includes: real-time cross-sectional load rate vector, meteorological sensitivity characteristics of each node, the current comprehensive frequency deviation of the system, and real-time predicted output of each heterogeneous resource.

[0073] To guide model evolution, the system employs the following reward function. : (14) in, To correspond to the real-time load rate of the restricted branch, This represents the standard deviation of the per-unit voltage values ​​of all monitored busbars across the entire network. In this embodiment, this term is introduced into the reward function and assigned a weight. The aim is to guide reinforcement learning agents to consider not only active power balance but also the uniformity of the overall network voltage distribution when generating ramp-up mapping schemes, thereby preventing large local voltage fluctuations. Weights The values ​​are 0.5, 0.3, and 0.2 respectively.

[0074] The system balance constraint verification calculation formula is as follows: (15) in, Allocate deviations to system-level ramping requirements. For the total system-level ramp-up requirements, For the first Expected net power fluctuation value for each physical bus For nodes In the region The ramp capacity relaxation amount obtained from internal allocation. To determine the convergence criterion threshold, a threshold of 0.1MW is used in this embodiment.

[0075] To verify the effectiveness of closed-loop optimization, Figure 3 The convergence curve of online parameter update for the model based on the PPO algorithm is shown.

[0076] like Figure 3As shown, the horizontal axis represents the number of reinforcement learning interaction rounds (Episodes), and the vertical axis represents the comprehensive reward function value (Reward). In the initial 100-round exploration phase, due to the mapping strategy not yet adapting to the dynamic topology, the cross-sectional load rate penalty was frequently triggered, and the reward value exhibited violent oscillations between -200 and -150. With the continuous online updates, the system's comprehensive reward function value rapidly climbed and stabilized at an optimization range of around +350 after approximately 400 iterations, proving that the surrogate model had successfully learned the optimal mapping parameter adjustment strategy under complex weather and network constraints.

[0077] This step ensures the rigor and robustness of the ramp-up requirement mapping scheme under complex constraints through parameterized closed-loop adjustment, explicit safety boundary correction, and a high-dimensional awareness self-evolution mechanism.

[0078] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0079] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0080] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0081] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0082] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0083] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for mapping climbing demand nodes based on multi-source contribution allocation, characterized in that, Includes the following steps: Obtain the initial weights of each source-load node in the power network and determine the system-level ramping requirements; A multidimensional joint distribution model reflecting the correlation of fluctuations among heterogeneous resources is constructed using the Vine Copula function, and the dynamic contribution coefficient of each heterogeneous resource to the system-level ramping demand is calculated based on the multidimensional joint distribution model. Based on the initial weights and dynamic contribution coefficients, spatial mapping calculations are performed to obtain the expected power change components of each source-load node; The Shapley value is used to quantify the marginal contribution of each source load node to eliminating network congestion, and the ramp capacity relaxation amount of each source load node is allocated according to the marginal contribution. Based on the expected power change components and the ramp capacity relaxation, a ramp demand mapping scheme that satisfies the node physical constraints is generated.

2. The method according to claim 1, characterized in that, The process of obtaining the initial weights of each source-load node in the power network includes: Graph attention neural networks are used to obtain the real-time topology feature matrix of power networks; Key transmission sections in the topology feature matrix are identified, and the weights of downstream nodes of restricted branches are corrected based on the branch load rate to obtain the initial weights.

3. The method according to claim 1, characterized in that, The construction of the multidimensional joint distribution model reflecting the correlation of fluctuations in various heterogeneous resources includes: Extract the prediction error sequence of each heterogeneous resource in the historical operation data under a preset time period, and establish the marginal distribution function of each sequence; By identifying a Vine structure that conforms to the current climate characteristics, and by performing conditional probability iteration on the marginal distribution function, a Vine Copula joint distribution function characterizing the higher-order tail correlation of heterogeneous resources is obtained.

4. The method according to claim 1, characterized in that, The spatial mapping calculation includes: The dynamic contribution coefficient is used as a correction factor and applied to the original mapping equation of each source load node. By aggregating various resource fluctuations on the same physical bus using vector summation, the expected net power fluctuation value of that node under the system ramp-up command is generated.

5. The method according to claim 1, characterized in that, The method of quantifying the marginal contribution of each source load node to eliminating network congestion using Shapley values ​​includes: The nodes participating in the ramp response are defined as the set of game participants, and a game characteristic function with the overall network blocking payoff as the objective is established. Calculate the blockage elimination increment caused by adding the target node to different node combination subsets, and obtain the Shapley value of the node by weighted averaging.

6. The method for mapping climbing demand nodes based on multi-source contribution allocation according to claim 1, characterized in that, The ramp capacity relaxation amount allocated to each source load node includes: The total ramp capacity relaxation of the system is spatially allocated using the Shapley value of each source load node as a weighting factor. Verify whether the slack amount of the node after distribution exceeds the adjustment limit of the corresponding node. If it does, redistribute the remaining part among the unsaturated nodes.

7. The method according to claim 1, characterized in that, The process of generating the mapping scheme also includes: Verify whether the sum of the expected power change components and the allocated relaxation amount of all source load nodes is consistent with the system-level ramping requirements. If there is a discrepancy, the dynamic contribution coefficient is updated through a negative feedback adjustment loop until the system balance constraint is met.

8. The method according to claim 1, characterized in that, Also includes: Monitor whether the power grid state corresponding to the mapping scheme triggers the voltage or frequency safety threshold; If triggered, the sensitivity of each source-load node to voltage exceedance is calculated using the node voltage sensitivity matrix, and the expected power change component is adjusted accordingly.

9. The method according to claim 1, characterized in that, The calculation of the dynamic contribution coefficient also includes: When the fluctuation characteristics of system-level ramping demand meet the preset triggering conditions, switch to the enhanced distributed model; In the augmented distribution model, increase the weight of resource nodes whose volatility exceeds a preset threshold.

10. The method according to any one of claims 1 to 9, characterized in that, Also includes: Power dispatch is performed according to the aforementioned ramp demand mapping scheme; The feedback results after execution scheduling are obtained, and the model parameters of the VineCopula function are updated online based on the feedback results using a reinforcement learning algorithm.

Images