A wind power and photovoltaic power grid access dispatching method and system based on two-stage robust optimization

Abstract

The application provides a wind power photovoltaic grid access scheduling method and system based on two-stage robust optimization, relates to the technical field of grid scheduling, and the method comprises the following steps: initializing a grid scheduling state space based on a high-dimensional uncertainty disturbance set, and obtaining an initial scheduling potential energy distribution tensor; performing node interaction feature analysis on the initial scheduling potential energy distribution tensor, and mapping the data correlation degree between flexible adjustment resource nodes and source load fluctuation nodes into equivalent node coupling strength; and calculating the tightness distribution of scheduling constraints according to the equivalent node coupling strength, and obtaining a constraint force field topology graph. The application improves the execution degree and operation stability of the wind power photovoltaic grid access scheduling instruction.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching technology, and in particular to a method and system for dispatching wind and solar power grid access based on two-stage robust optimization. Background Technology

[0002] With the large-scale integration of new energy sources such as wind power and photovoltaics into the power system, the intermittency and volatility of their output pose certain challenges to the optimal scheduling of the power grid. Among the existing scheduling methods, two-stage robust optimization is often used to deal with the uncertainty of new energy sources. It usually generates correction scheduling instructions based on the real-time wind and solar power output deviation in the second stage.

[0003] However, in practical applications, traditional two-stage robust scheduling methods mostly use relatively fixed step sizes or linear rates for power adjustment in the correction commands generated in the second stage, rarely considering the jerk constraints on the power adjustment rates of flexible adjustment resources, such as energy storage devices and adjustable loads. The first derivative of the power adjustment rate with respect to time is defined as the power adjustment acceleration, and the second derivative is defined as the power adjustment jerk (i.e., the rate of change of the rate of change of the rate). The jerk constraint is essentially a key constraint to prevent sudden changes in the adjustment rate, as well as the inertial lag characteristics of the power flow response of the grid. For example, in the operation of a regional grid with wind and solar power integration, when the rapid changes in cloud cover in the afternoon cause a significant drop in photovoltaic output in a short period of time, the secondary correction command output with a conventional fixed step size may cause the energy storage system or adjustable load to lag or overshoot in actual output due to the rapid rate change during the tracking of the command. This can lead to node voltage fluctuations and line power oscillations, affecting the execution effect of the scheduling command and the safe and stable operation of the system. Summary of the Invention

[0004] This invention provides a two-stage robust optimization-based method and system for wind power and solar power grid connection scheduling, which improves the execution efficiency and operational stability of wind power and solar power grid connection scheduling instructions.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, a two-stage robust optimization-based wind and solar power grid access scheduling method is provided, the method comprising: Step 1: Initialize the power grid dispatch state space based on the set of high-dimensional uncertain disturbances to obtain the initial dispatch potential energy distribution tensor; Step 2: Analyze the interaction characteristics between nodes on the initial scheduling potential energy distribution tensor, and map the data correlation between the flexible adjustment resource nodes and the source load fluctuation nodes to the equivalent node coupling strength; calculate the tightness distribution of scheduling constraints based on the equivalent node coupling strength to obtain the constraint force field topology map; Step 3: Based on the constraint force field topology diagram, the power regulation rate of the flexible regulation resource node is optimized. During the optimization process, the rate trajectory is discretized into a time series using the upper and lower limits of the jerk of the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient, and the step size change rate and attenuation factor of the two-stage correction dispatch command are calculated to generate a robust dispatch command sequence. Step 4: Send the robust scheduling command sequence to the grid edge control node, collect the real-time wind and solar power output deviation, use the real-time wind and solar power output deviation to perform feedback correction and compensation on the robust scheduling command sequence, update the power imbalance gradient distribution in the constraint force field topology diagram, and obtain the final two-stage robust optimization scheduling strategy for wind and solar power grid connection.

[0006] Secondly, a wind power and solar power grid connection dispatch system based on two-stage robust optimization includes: The distributed construction module is used to initialize the power grid dispatch state space based on a high-dimensional set of uncertain disturbances, and obtain the initial dispatch potential energy distribution tensor. The constraint generation module is used to analyze the interaction characteristics between nodes of the initial scheduling potential energy distribution tensor, and to map the data correlation between flexible adjustment resource nodes and source load fluctuation nodes to equivalent node coupling strength; based on the equivalent node coupling strength, the tightness distribution of scheduling constraints is calculated to obtain the constraint force field topology map. The dispatching instruction module is used to optimize the power regulation rate of flexible regulation resource nodes based on the constraint force field topology diagram. During the optimization process, the rate trajectory is discretized into a time series by using the upper and lower limits of the jerk of the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient, and the step size change rate and attenuation factor of the two-stage correction dispatching instruction are calculated to generate a robust dispatching instruction sequence. The feedback update module is used to send the robust scheduling command sequence to the grid edge control node, collect the real-time wind and solar power output deviation, use the real-time wind and solar power output deviation to perform feedback correction and compensation on the robust scheduling command sequence, update the power imbalance gradient distribution in the constraint force field topology diagram, and obtain the final two-stage robust optimization scheduling strategy for wind and solar power grid connection.

[0007] The above-described solution of the present invention has at least the following beneficial effects: The initial construction of the grid dispatch state space is completed by using a set of high-dimensional uncertain disturbances to adapt to the uncertain operating conditions of the grid caused by random fluctuations in wind power and photovoltaic output. The coupling correlation between flexible regulation resource nodes and source-load fluctuation nodes is quantified by analyzing the interaction characteristics between grid nodes. The topology of the constraint force field is constructed according to the tightness of dispatch constraints to fully restore the internal constraint correlation logic and operating status of the grid. The upper and lower limits of jerk constraints are incorporated into the quadratic programming framework for optimizing the power regulation rate trajectory to avoid the impact of flexible regulation resource equipment and grid power oscillation caused by sudden rate changes. The power regulation rate curve is dynamically matched with the inertia coefficient of the grid power flow response, and the step size change rate and attenuation factor are calculated to make the rhythm of the correction command adapt to the lag characteristics of the grid power flow response and avoid node voltage fluctuations and line power oscillations caused by command overshoot. Attached Figure Description

[0008] Figure 1 This is a schematic flowchart of a wind power and solar power grid access scheduling method based on two-stage robust optimization, provided by an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of a wind power and photovoltaic grid access dispatch system based on two-stage robust optimization, provided by an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] like Figure 1 As shown, an embodiment of the present invention proposes a wind power and solar power grid access scheduling method based on two-stage robust optimization, the method comprising the following steps: Step 1: Initialize the power grid dispatch state space based on the set of high-dimensional uncertain disturbances to obtain the initial dispatch potential energy distribution tensor; Step 2: Analyze the interaction characteristics between nodes on the initial scheduling potential energy distribution tensor, and map the data correlation between the flexible adjustment resource nodes and the source load fluctuation nodes to the equivalent node coupling strength; calculate the tightness distribution of scheduling constraints based on the equivalent node coupling strength to obtain the constraint force field topology map; Step 3: Based on the constraint force field topology diagram, the power regulation rate of the flexible regulation resource node is optimized. During the optimization process, the rate trajectory is discretized into a time series using the upper and lower limits of the jerk of the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient, and the step size change rate and attenuation factor of the two-stage correction dispatch command are calculated to generate a robust dispatch command sequence. Step 4: Send the robust scheduling command sequence to the grid edge control node, collect the real-time wind and solar power output deviation, use the real-time wind and solar power output deviation to perform feedback correction and compensation on the robust scheduling command sequence, update the power imbalance gradient distribution in the constraint force field topology diagram, and obtain the final two-stage robust optimization scheduling strategy for wind and solar power grid connection.

[0012] In this embodiment of the invention, the initial construction of the grid dispatch state space is completed by a high-dimensional uncertainty disturbance set, which is adapted to the uncertain operating conditions of the grid caused by the random fluctuations in wind power and photovoltaic output. The coupling correlation between flexible regulation resource nodes and source-load fluctuation nodes is quantified by analyzing the interaction characteristics between grid nodes. The constraint force field topology is constructed according to the tightness of dispatch constraints, which fully restores the internal constraint correlation logic and operating status of the grid. The upper and lower limits of jerk constraints are incorporated into the quadratic programming framework for optimizing the power regulation rate trajectory, so as to avoid the impact of flexible regulation resource equipment and grid power oscillation caused by sudden rate changes. The power regulation rate curve is dynamically matched with the grid power flow response inertia coefficient, and the step size change rate and attenuation factor are calculated so that the rhythm of the correction command is adapted to the lag characteristics of the grid power flow response, thus avoiding node voltage fluctuations and line power oscillations caused by command overshoot.

[0013] In a preferred embodiment of the present invention, obtaining the wind and solar power output time-series prediction sequence, multi-source heterogeneous meteorological disturbance parameters, and network topology constraint data of the target power grid within a preset scheduling period, and constructing a high-dimensional uncertainty disturbance set, may include: In this embodiment of the invention, a regional power grid is taken as the object. This power grid includes 2 220kV substations, 5 110kV substations, and 12 35kV substations; 3 wind farms (total installed capacity 450MW, connected to nodes B1, B2, and B3 respectively), 4 photovoltaic power stations (total installed capacity 300MW, connected to nodes C1-C4 respectively); 2 energy storage power stations (total capacity 50MW / 100MWh, connected to nodes D1 and D2); and 3 adjustable industrial loads (total adjustable capacity 80MW, connected to nodes E1-E3). The preset scheduling cycle is 4 hours of intraday rolling scheduling (5-minute time resolution, 48 time periods) + 15 minutes of real-time scheduling (1-minute time resolution, 15 time periods). Based on the short-term fluctuation characteristics of wind and photovoltaic output and the timing requirements of the two-stage scheduling decision of the power grid, the preset scheduling cycle is set to 4 hours of intraday rolling scheduling (00:00-04:00, 04 ... (00:00-08:00, etc., rolling in segments) overlaid with 15-minute real-time scheduling, with time resolution differentiated according to scheduling level. The intraday rolling segment uses a 5-minute time interval, and the real-time scheduling segment uses a 1-minute time interval. This cycle range and resolution can fully cover the full-time characteristics of sudden fluctuations in wind and solar power output and dynamic response of power grid flow, matching the execution logic of two-stage robust scheduling pre-decision + real-time correction. When obtaining the wind and solar power output time series prediction sequence, the basic information such as the geographical location, installed capacity, and grid connection nodes of all grid-connected wind farms and photovoltaic power stations in the target grid is first sorted out. Relying on the grid-dedicated new energy power prediction system, the above 4-hour rolling data of each wind and solar farm is retrieved. The system generates power output prediction data within a dynamic cycle plus a 15-minute real-time cycle. Following predetermined time resolutions of 5 minutes and 1 minute, the hourly power output prediction values ​​for each station are compiled into a continuous time series. This also includes predictions for scenarios that can cause significant fluctuations in wind and solar power output, such as sudden changes in cloud cover, wind speed, and extreme temperatures. This comprehensively covers all characteristics of normal stable fluctuations and short-term sharp increases and decreases in wind and solar power output, forming a wind and solar power output time series prediction sequence that closely reflects the actual operating state of the target power grid. When acquiring multi-source heterogeneous meteorological disturbance parameters, the system uses the time range and resolution of the preset scheduling cycle as a benchmark, drawing data from provincial meteorological department numerical weather predictions and local meteorological monitoring terminals at wind and solar power stations. The system simultaneously collects meteorological information from various data sources, including regional meteorological micro-stations and meteorological satellite remote sensing data, covering the corresponding power grid coverage areas and various wind and solar power station locations. The collected data includes conventional indicators such as wind speed, wind direction, solar irradiance, ambient temperature, relative humidity, atmospheric pressure, cloud cover, and precipitation level. At the same time, it extracts disturbance characteristic data that can directly cause drastic fluctuations in wind and solar power output, such as severe convective weather, short-term strong winds, rapid cloud movement, and sudden changes in illumination. All meteorological data are time-series normalized at 1-minute / 5-minute intervals and spatially matched according to station coordinates and power grid node locations to form multi-dimensional, multi-type, multi-source heterogeneous meteorological disturbance parameters.When acquiring network topology constraint data, the steady-state operation of the power grid within a preset scheduling period is assumed. Complete grid structure information of the target power grid is retrieved from the power grid energy management system and topology configuration platform to clarify the physical connection relationships and node distribution of equipment such as buses, transmission lines, transformers, conventional thermal power units, energy storage systems, adjustable loads, and new energy grid connection points. Simultaneously, operational data such as the power transmission upper limit of each transmission line, the allowable operating range of node voltage, the upper and lower limits of equipment output, power flow stability safety boundaries, and grid connection interface constraints are collected within the preset period to completely lock down the network topology structure and rigid scheduling constraints of the target power grid within the scheduling period. After completing the comprehensive collection of the three types of data, based on the unified time axis of the preset scheduling cycle and the spatial coordinates of the power grid nodes, all data are aligned in both time and space. The predicted wind and solar power output, meteorological disturbance parameters, and network topology constraint data corresponding to the same power grid node / wind and solar power station at the same time are linked and integrated one by one. With random fluctuations in wind and solar power output, multi-dimensional meteorological disturbances, and rigid constraints of the power grid topology as the core dimensions, all aligned and integrated feature information is collected and fused to ultimately form a high-dimensional uncertainty disturbance set that can comprehensively characterize the superimposed effects of various uncertainties after wind and solar power are connected to the grid.

[0014] In a preferred embodiment of the present invention, step 1 above, which initializes the power grid dispatch state space based on a high-dimensional set of uncertain disturbances to obtain an initial dispatch potential energy distribution tensor, may include: In this embodiment of the invention, step 110 involves aligning the wind and solar power output time-series prediction sequence in the high-dimensional uncertainty disturbance set with the multi-source heterogeneous meteorological disturbance parameters using a spatiotemporal reference and normalizing the dimensions. This process extracts the source-load fluctuation characteristics and meteorological disturbance coupling vectors of each node within a preset scheduling period, constructing a standardized spatiotemporal disturbance feature sequence. Specifically, this includes performing spatiotemporal reference alignment, using a fixed time axis of the preset scheduling period, the unique numbers of all bus nodes in the power grid, and the fixed spatial coordinates of the wind and solar power station grid-connected nodes as a unified reference system. With multi-source heterogeneous meteorological disturbance parameters, each point was bound according to a completely consistent time segment and a perfectly matched spatial node. For the 48 5-minute time segments of intraday rolling scheduling and the 15 1-minute time segments of real-time scheduling, the timestamps of the wind and solar power output forecasts and meteorological parameters were checked one by one to eliminate the misalignment of the two types of data in the time dimension. At the same time, according to the grid node number and the grid connection location of wind and solar power plants, the meteorological parameters were accurately matched to the corresponding power generation nodes and load nodes to eliminate the point deviation in the spatial dimension and ensure that all data are accurately aligned within the same spatiotemporal framework. Then, dimensional normalization processing was performed to adjust the time... After empty alignment, all continuous physical parameters are normalized using extreme value normalization to eliminate dimensional differences and numerical magnitude gaps between different parameters. Normalization calculations are performed on all parameters, including wind and solar active power output, wind speed, solar irradiance, ambient temperature, electrical load power, and node voltage. After normalization, all parameters are mapped to the standard range of 0-1, completely eliminating the interference of dimensional differences on subsequent calculations. After normalization, source-load fluctuation characteristics are extracted time-series for each power source node, load node, and renewable energy grid connection point within the power grid. The fluctuation amplitude, slope, and magnitude of wind and solar power output are extracted from the source side. The duration and short-term output changes are varied; the power change rate, peak-valley deviation, and average load value of the load are extracted from the load side; at the same time, normalized meteorological parameters such as wind speed, solar irradiance, ambient temperature, and cloud cover are fused with the source-load fluctuation characteristics of the corresponding nodes point by point to form a meteorological disturbance coupling vector that can directly reflect the impact of meteorological disturbances on source-load output; the source-load fluctuation characteristics and meteorological disturbance coupling vectors of all nodes are arranged sequentially according to the grid node numbering order and the time section order of the preset scheduling cycle to construct a continuous, complete, and standardized spatiotemporal disturbance characteristic sequence without spatiotemporal deviation.

[0015] Step 111: Based on the standardized spatiotemporal disturbance feature sequence and combined with network topology constraint data, calculate the initial power transmission sensitivity matrix between each node, and map the initial power transmission sensitivity matrix to the grid steady-state operation boundary space to obtain the grid dispatch initial state space feature matrix; specifically, this includes: carrying out deep fusion matching of the standardized spatiotemporal disturbance feature sequence and network topology constraint data. The network topology constraint data fully includes the unique number, physical spatial location, and electrical connection attributes of all bus nodes in the target grid, the type, length, impedance parameters, and upper limits of active and reactive power transmission of transmission lines, the transformer ratio, rated capacity, and connection method, grid connection node information of conventional units, wind power, photovoltaic power, energy storage, and adjustable loads, the upper and lower limits of voltage amplitude operation of each bus node, the active and reactive power balance constraints of the entire network, and all rigid operation constraint information such as power flow constraints and node power constraints for grid steady-state operation; In the standardized spatiotemporal disturbance feature sequence, the source-load fluctuation characteristics and meteorological disturbance coupling vector corresponding to each power grid node are bound to the physical connection relationship, electrical parameters, and operational constraint boundaries of that node in the network topology constraint data on a node-by-node and time-by-time cross-section basis. This ensures that each set of disturbance features is completely matched with the topological attributes and operational boundaries of the corresponding node. Power transmission sensitivity is calculated node-by-node to construct an initial power transmission sensitivity matrix. Power transmission sensitivity is a core indicator that quantitatively describes the impact of power injection changes at a certain node on the electrical operating parameters of other nodes. It reflects the electrical coupling relationship and power transfer law between nodes. All power source nodes, load nodes, new energy grid connection points, and bus nodes in the target power grid are traversed. The sensitivity values ​​of all node combinations with direct or indirect electrical connections are calculated one by one. The sensitivity values ​​of node combinations without electrical connections are directly recorded as 0. For each group of nodes with coupling relationships... With nodes Extract the injection power change of node i at a single time segment within a preset scheduling period from the standardized spatiotemporal disturbance feature sequence, and simultaneously extract the node... Under the same time section, affected by the node The change in voltage amplitude caused by power variation, or the node With nodes The power flow changes of the interconnected lines are used to calculate the sensitivity value according to the formula, which is as follows: ,in For nodes in the power grid For nodes The power transmission sensitivity value; the larger the value, the stronger the node's sensitivity. Power fluctuations on nodes The more significant the impact of the operating parameters; For nodes The change in injected power at a single time segment within a preset scheduling period is directly extracted from the source load fluctuation characteristics of the standardized spatiotemporal disturbance characteristic sequence. For nodes At the corresponding time section, the node is affected. The voltage amplitude change or the power flow change of the two-node connected lines is calculated by combining the standardized spatiotemporal disturbance characteristic sequence with the grid topology connection relationship. According to the numbering order of all time sections and grid nodes in the preset scheduling period, the sensitivity values ​​of all node pairs are arranged in order. A two-dimensional sensitivity sub-matrix is ​​generated for each time section. Then, the sub-matrices of all time sections are stacked and integrated in chronological order to finally form an initial power transmission sensitivity matrix that can completely characterize the dynamic changes of the power transmission correlation characteristics between all network nodes within the preset scheduling period.

[0016] The initial power transmission sensitivity matrix is ​​mapped to the steady-state operation boundary space of the power grid to complete boundary verification and invalid feature removal. The steady-state operation boundary space of the power grid is the safe and legal operating range of the power grid, jointly defined by all rigid operating constraints in the network topology constraint data. All scheduling feature data must be within this space to have practical application value. The allowable range of node voltage amplitude, the upper limit of transmission line power transmission, the power balance equation of the whole network, and the power flow stability constraints in the network topology constraint data are used as rigid verification boundaries. The sensitivity values ​​and associated operating parameters of each time segment and each node pair in the initial power transmission sensitivity matrix are verified one by one. During the verification, the node voltage, line power flow, node power, and other parameters bound to the sensitivity values ​​are extracted first, and it is determined whether the parameters are within the steady-state operation boundary. If the parameter exceeds the constraint boundary, the sensitivity value and the corresponding associated feature are marked as invalid features and directly removed from the matrix; if the parameter is within the constraint boundary, the sensitivity value and the associated feature are marked as valid features. The initial power transmission sensitivity matrix is ​​fully preserved and mapped to the steady-state operation boundary space of the power grid through full and point-by-point boundary verification. Only valid feature data that meets the requirements for safe and steady-state operation of the power grid are retained. The valid feature data is reconstructed to generate the initial state space feature matrix of the power grid dispatch. All valid feature data retained after boundary verification are rearranged and integrated according to unified rules. The first dimension is sorted according to the unique number of all nodes in the power grid. The second dimension is sorted according to the time section of the preset dispatch period. The third dimension is sorted according to the electrical parameter types such as node voltage, line power flow, power sensitivity, and source load power. The integrated multi-dimensional valid feature data is constructed into a structured matrix. This matrix fully contains all core dispatch state information such as the steady-state operation parameters of each node in the power grid within the preset dispatch period, the power transmission correlation characteristics between nodes, and the source load disturbance coupling characteristics. It can reflect the steady-state operation state of the power grid under the uncertainty disturbance of wind and solar power, and finally form the initial state space feature matrix of the power grid dispatch.

[0017] Step 112: Based on the initial state space feature matrix of the power grid dispatch, calculate the power mismatch risk and voltage over-limit probability of each spatiotemporal node under the influence of uncertain disturbances, and perform multimodal weighted fusion of the power mismatch risk and voltage over-limit probability to form a primary potential energy distribution matrix characterizing the overall grid dispatch pressure situation. Specifically, this includes: performing data extraction and traversal range determination before calculation; using a preset dispatch cycle as the time benchmark, locking all calculation time segments, including 48 five-minute time segments for 4 hours of daily rolling dispatch and 15 one-minute time segments for 15 minutes of real-time dispatch; using the node dimension of the initial state space feature matrix of the power grid dispatch as the spatial benchmark, traversing all power source nodes, load nodes, new energy grid connection points, and bus nodes within the target power grid, covering all grid nodes participating in dispatch; and from the initial state space feature matrix of the power grid dispatch... In the feature matrix, corresponding operating parameters are extracted for each time segment and each node. Simultaneously, the rated power baseline and voltage operating limits for each node are retrieved from the network topology constraint data, and the wind, solar, and meteorological disturbance sample sets for each node are retrieved from the high-dimensional uncertainty disturbance set, providing complete basic data for risk indicator calculation. Power mismatch risk is calculated for each spatiotemporal node. The power mismatch risk is used to accurately quantify the degree of mismatch between the total power output and total load power of a single grid node at a given moment under the influence of wind and solar uncertainty disturbances. It is a core indicator reflecting the risk of power supply and demand imbalance at a node; a larger value indicates a more severe power supply and demand gap or redundancy at the node, and higher scheduling pressure. For each time t and each node k, three core parameters are extracted from the initial state space feature matrix of the power grid scheduling. The first parameter is the total power output of node k at time t. This value integrates the total active power output of all power sources, including wind power, photovoltaic power, conventional thermal power units, and energy storage discharge; the second term is the total load power of node k at time t. This value summarizes the total power consumption of all residential loads, industrial loads, and adjustable loads under this node; the third item is the rated power reference value of node k. This value, obtained from network topology constraint data, represents the rated capacity of the node's grid-connected interface and serves as a standardized benchmark for the degree of power mismatch. Based on these parameters, the power mismatch risk is calculated point-by-point according to the formula. ,in This represents the power mismatch risk level of node k at time t; It represents the absolute difference between the power output and load power of node k at time t, eliminating the positive and negative differences between supply and demand surplus and deficit, and only retaining the imbalance magnitude; according to the entire time section of the preset scheduling cycle and the numbering order of all nodes in the power grid, the calculation is completed one by one and the values ​​are recorded to form a power mismatch risk dataset covering the entire time and space range.

[0018] The voltage exceedance probability is calculated for each spatiotemporal node. This probability quantifies the likelihood that a node voltage will exceed its steady-state safe operating range under uncertain disturbances. It is a core indicator reflecting the node voltage safety risk; a higher value indicates a higher risk of voltage exceedance and poorer grid stability. For each time t and each node k, using meteorological and wind / solar power output disturbance scenarios from a high-dimensional uncertain disturbance set as statistical samples, the voltage operating value of node k under each disturbance scenario is simulated to determine whether it exceeds the upper and lower limits of voltage operation specified by the network topology constraint data. Two types of sample counts are recorded: one is the number of disturbance samples where the voltage of node k exceeds the allowable operating range at time t. One type is the total number of disturbance scenarios involving voltage exceeding limits; the other type is the total number of disturbance samples involving node k at time t. This refers to the total number of all wind and weather disturbance scenarios corresponding to this node in the high-dimensional uncertainty disturbance set. Based on the above statistical data, the voltage over-limit probability is calculated point by point according to the formula. ,in The value represents the voltage over-limit probability of node k at time t, ranging from 0 to 1. The closer the value is to 1, the higher the probability of voltage over-limit. The values ​​are calculated and recorded one by one in a spatiotemporal order that is completely consistent with the power mismatch risk level, forming a voltage over-limit probability dataset covering the entire spatiotemporal range.

[0019] Multimodal weighted fusion of two types of risk indicators is performed to calculate the primary potential energy value. The primary potential energy value is a comprehensive indicator that integrates power mismatch risk and voltage safety risk, used to intuitively characterize the overall dispatch pressure of a single node at a certain moment. Weighted fusion is needed to achieve quantitative coupling of the two types of risks. Weighting coefficients are set according to the priority of grid dispatch safety. Grid dispatch takes power supply and demand balance as its core objective and voltage stability as its key constraint; therefore, the weighting coefficients for power mismatch risk are... The weighting coefficient for the voltage over-limit probability is set to 0.6. The coefficients are set to 0.4, and the sum of the two types of coefficients is 1 to ensure the standardization and rationality of the fusion results. For each spatiotemporal node, the power mismatch risk and voltage over-limit probability calculated synchronously are substituted into the fusion formula to calculate the primary potential energy value of that node at that moment. ,in This represents the primary potential energy value of node k at time t. The larger the value, the greater the impact of uncertainty disturbances on the node at that time, and the higher the scheduling and control pressure. The primary potential energy values ​​of all nodes and all time sections are calculated according to a unified spatiotemporal order. The primary potential energy values ​​are then systematically integrated to construct a primary potential energy distribution matrix. The primary potential energy values ​​of all spatiotemporal nodes are arranged in a two-dimensional order according to the spatial order of the unique grid node numbers from smallest to largest and the temporal order of the time sections of the preset scheduling cycle from first to last. The row dimension corresponds to the numbers of all grid nodes, and the column dimension corresponds to all time sections of the preset scheduling cycle. Each element in the matrix corresponds to the primary potential energy value of a unique node and a unique time. This matrix can completely and intuitively represent the spatiotemporal distribution of scheduling pressure of all nodes in the entire network under the uncertainty disturbances of wind and solar power within the preset scheduling cycle, ultimately forming the primary potential energy distribution matrix.

[0020] Step 113 involves performing multidimensional tensor reconstruction and orthogonal basis projection transformation on the primary potential energy distribution matrix. Tensor product operations and feature dimensionality reduction are performed along the node topology dimension, temporal evolution dimension, and perturbation confidence dimension to obtain the initial scheduling potential energy distribution tensor. Specifically, this includes performing multidimensional tensor reconstruction on the primary potential energy distribution matrix. The primary potential energy distribution matrix output in step 112 is a two-dimensional matrix containing only node topology and temporal evolution. The row dimension corresponds to the unique number of all operating nodes in the power grid, and the column dimension corresponds to all time sections within the preset scheduling period. Each element in the matrix represents a pair of nodes. The primary potential energy values ​​at corresponding nodes and times are used to fully represent the confidence characteristics of wind and solar power uncertainties. To fully represent the confidence characteristics of these uncertainties, the two-dimensional matrix needs to be expanded into a three-dimensional structured tensor, adding a perturbation confidence dimension. This dimension is divided into high, medium, and low levels based on the confidence levels of wind and solar power output prediction and meteorological perturbations. High confidence level corresponds to scenarios with wind and solar power prediction errors less than 5% and stable meteorological perturbations; medium confidence level corresponds to scenarios with wind and solar power prediction errors of 5%-10% and minor fluctuations in meteorological perturbations; and low confidence level corresponds to scenarios with wind and solar power prediction errors greater than 10% and severe fluctuations in meteorological perturbations. During the reconstruction process, the two-dimensional primary potential energy distribution matrix is ​​expanded according to the high, medium, and low perturbation confidence levels. The primary potential energy values ​​at the same node and time under different perturbation confidence levels are integrated into individual elements of the three-dimensional tensor, ultimately forming a three-dimensional initial tensor with dimensions of node topology × temporal evolution × perturbation confidence, thus completing the structured representation of the multi-dimensional characteristics of the scheduling state.

[0021] An orthogonal basis projection transformation is performed on the reconstructed 3D tensor. An orthogonal basis decomposition algorithm is used to remove redundant features and refine core features. The specific implementation process is as follows: First, the initial 3D tensor is expanded into two-dimensional matrices along three dimensions: node topology, temporal evolution, and perturbation confidence. Expanding along the node topology dimension yields the node feature matrix, along the temporal evolution dimension yields the temporal feature matrix, and along the perturbation confidence dimension yields the perturbation confidence feature matrix. For each expanded two-dimensional matrix, singular value decomposition is performed sequentially, extracting the left singular vectors of the matrix as orthogonal basis vectors for the corresponding dimensions. The magnitude of the singular values ​​directly represents the corresponding... The importance of a feature is determined by the singular value; a larger singular value indicates a stronger ability of the feature to represent scheduling pressure and uncertainty. The principal component orthogonal basis vectors extracted from the three dimensions are combined to construct an orthogonal basis space covering all dimensions. The original three-dimensional initial tensor is then projected into this orthogonal basis space to complete the orthogonal transformation of the features. During the projection process, redundant feature components with singular values ​​less than a set threshold are directly removed, and only core disturbance features with large singular values ​​that can reflect the risk of power grid mismatch, voltage over-limit risk, node coupling relationship, and time-series fluctuation are retained. This process eliminates redundant information and interference noise in the tensor, resulting in a purified three-dimensional core feature tensor.

[0022] Tensor product operations are performed sequentially along the three dimensions of the core feature tensor to achieve deep coupling and fusion of multi-dimensional features. The specific implementation process is as follows: The tensor product operation is a third-order tensor outer product operation, used to fuse the core features of the three independent dimensions of node topology, temporal evolution, and disturbance confidence into a unified feature with coupling correlation. First, the core feature vectors of the node topology dimension, the temporal evolution dimension, and the disturbance confidence dimension are extracted from the core feature tensor after orthogonal basis projection transformation. Then, the third-order tensor product operation is performed with the node topology feature vector as the spatial reference, the temporal evolution feature vector as the temporal reference, and the disturbance confidence feature vector as the uncertainty reference. The feature vectors of the three dimensions are combined by the outer product of all dimensions. During the operation, each feature combination will generate a coupled feature value. This value contains three information: node spatial correlation, temporal dynamic change, and disturbance confidence level. It can completely describe the coupling law of power grid dispatch potential energy under different nodes, different times, and different disturbance intensities. By using tensor product operations, the isolation of features in a single dimension is completely broken, and the intrinsic relationship between node topology, temporal evolution, and perturbation confidence is strengthened. This allows the tensor to simultaneously represent spatial topological relationships, temporal dynamic characteristics, and uncertain perturbation patterns. Feature dimensionality reduction is then performed on the coupled feature tensor after tensor product operations. First, the variance contribution rate of all feature components in the coupled feature tensor is calculated. The variance contribution rate represents the degree of contribution of a single feature component to the overall scheduling state representation. All feature components are sorted from largest to smallest according to their contribution rate, and the contribution rates of the feature components are accumulated sequentially. When the accumulated contribution rate reaches 95%, the filtering stops, and minor interfering features with extremely low contribution rates are removed. After dimensionality reduction and filtering, the initial scheduling potential energy distribution tensor is finally obtained.

[0023] Through progressive end-to-end processing, the spatiotemporal calibration, feature extraction, risk quantification, and tensor quantification modeling of wind and solar uncertainty data are completed from the source, characterizing the distribution of grid dispatch pressure and uncertainty, and improving the two-stage robust dispatch and grid operation stability.

[0024] In a preferred embodiment of the present invention, step 2 above, which involves parsing the inter-node interaction characteristics of the initial scheduling potential energy distribution tensor and mapping the data correlation between the flexible adjustment resource nodes and the source load fluctuation nodes to the equivalent node coupling strength; calculating the tightness distribution of scheduling constraints based on the equivalent node coupling strength to obtain the constraint force field topology graph, may include: In this embodiment of the invention, step 220 involves extracting the power fluctuation feature sequence in the time dimension and the adjacency association feature vector in the topological dimension for each flexible adjustment resource node and each source-load fluctuation node based on the initial scheduling potential energy distribution tensor; concatenating the power fluctuation feature sequence and the adjacency association feature vector, and calculating the normalized dynamic time curvature distance of the feature sequence between the concatenated nodes to obtain the initial association difference between the nodes; specifically, this includes: traversing and filtering the node types of the initial scheduling potential energy distribution tensor, identifying all grid operation nodes one by one along the node topological dimension of the tensor, and dividing them into two types of target nodes: flexible adjustment resource nodes and source-load fluctuation nodes. Flexible adjustment resource nodes include energy storage system nodes, adjustable load nodes, electricity-to-gas device nodes, and gas turbine unit nodes; source-load fluctuation nodes include wind power grid-connected nodes, photovoltaic grid-connected nodes, fluctuating residential load nodes, and fluctuating industrial load nodes; pairing each flexible adjustment resource node with each source-load fluctuation node to form an independent node pair to be analyzed; after completing the node pair construction, targeting... For each node pair, the power fluctuation feature sequence in the time-series dimension is extracted from the flexible adjustment resource node and the source-load fluctuation node. Following the time-series evolution dimension of the initial scheduling potential energy distribution tensor, and using all time segments of the preset scheduling cycle as a benchmark (including 48 5-minute time segments for intraday rolling scheduling and 15 1-minute time segments for real-time scheduling), four core power fluctuation features are extracted from each time segment: the power fluctuation amplitude between the current and previous time segments, the power fluctuation slope within a unit time segment, the duration of the current power fluctuation state, and the short-term power change within three consecutive time segments. The four fluctuation features corresponding to each time segment are combined in a fixed order to form a single-time-segment feature vector. Then, following the order of the time segments in the preset scheduling cycle from first to last, all single-time-segment feature vectors are concatenated to form a time-series power fluctuation feature sequence with a length completely consistent with the time segment of the scheduling cycle. This yields the time-series power fluctuation feature sequences for the flexible adjustment resource node and the source-load fluctuation node, respectively.

[0025] For each node pair, a topological adjacency correlation feature vector is extracted. Following the node topological dimension of the initial scheduling potential energy distribution tensor, and based on the physical topology and electrical connection relationship of the power grid, two core topological correlation features between nodes are extracted: the electrical connection distance between the flexible regulation resource node and the source-load fluctuation node, the number of directly adjacent nodes of each node, the length of the indirect topological path between the two nodes, and the correlation level of the two nodes in the overall power grid topology. These four topological features are integrated into a vector form with a unified dimension in a fixed order, forming an adjacency correlation feature vector that only represents the topological correlation relationship. This ensures that the topological feature vector dimensions of different node pairs are consistent. After completing the extraction of both types of features, the features of the same node pair are sequentially concatenated. The time-series power fluctuation feature sequences of the flexible regulation resource node and the source-load fluctuation node are used as the preorder part, and their topological adjacency correlation feature vectors are used as the postorder part. The concatenation follows the rule of time-series features first, topological features last, forming a joint feature sequence specific to that node pair.

[0026] The original dynamic time curvature distance between two sets of joint feature sequences is calculated. This distance measures the similarity between two time-series feature sequences in terms of fluctuation trends and patterns of change. It can adapt to scenarios where the two sequences have different lengths and shifts in fluctuation phase. A smaller distance indicates a better fit between the power fluctuation time-series patterns and topological correlation characteristics of the two nodes, and a higher potential for interaction and correlation between the nodes. The two joint feature sequences to be calculated are determined, and the joint feature sequence corresponding to the flexible adjustment resource node is denoted as... The sequence length is ,Right now ,in Represents a sequence The Middle The feature elements at each position; the joint feature sequence corresponding to the source load fluctuation node is denoted as... The sequence length is ,Right now ,in Represents a sequence The Middle Feature elements at each position; calculate the local distance matrix, which is used to characterize the sequence. The Middle Feature elements and sequences The Middle The degree of instantaneous difference between each feature element is determined by calculating the pairwise local distances between all elements using Euclidean distance, forming... OK Local distance matrix of columns The first in the matrix Line 1 Column elements The calculation formula is ,in The dimension of a single feature element. For sequence No. The element of the first 3D eigenvalues For sequence No. The element of the first 1D eigenvalues; then, a cumulative distance matrix is ​​constructed based on the local distance matrix, and the cumulative distance matrix is... OK Column matrix Used to record the sequence from the starting point (1, 1) to the current position ( , The minimum cumulative bending distance is constructed in two steps: boundary initialization and dynamic programming recursion. The first step is to perform matrix boundary initialization, setting the element in the first row and first column of the cumulative distance matrix to the corresponding value of the local distance. The remaining elements in the first row are accumulated forward along the sequence direction, i.e. The remaining elements in the first column are accumulated forward along the sequence direction, i.e. The second step is to perform dynamic programming recursive calculations on the elements within the matrix. and any position The minimum cumulative distance is selected according to the dynamic programming recursive formula, which is as follows: ,in For position The minimum cumulative bending distance is determined; after constructing the cumulative distance matrix, a dynamic programming backtracking algorithm is used to search for the final bending path. The final bending path is the bending path that connects the start and end points of the sequence and has the minimum total cumulative distance. The backtracking process starts from the end point of the matrix. Begin by working backwards point by point to the starting point (1, 1), following the minimum cumulative distance rule at each step and comparing the current position. left side , lower side lower left side Three cumulative distance values ​​are used, and the position with the smallest value is selected as the previous path node. This backtracking step is repeated until the starting point (1,1) is reached. All the backtracked positions are connected in sequence to obtain the final curved path of the two sequences. Based on the final curved path, the original dynamic time curved distance is calculated, and the local distances corresponding to all positions passed on the final curved path are calculated. By performing point-by-point summation, the total accumulated value is the original dynamic time-bending distance between the two joint feature sequences. This value fully reflects the overall difference between the two sequences in terms of power fluctuation timing and topological correlation characteristics.

[0027] Based on the original dynamic time bending distance, normalization is performed to eliminate the calculation bias caused by differences in the joint feature sequence length among different nodes, resulting in the normalized dynamic time bending distance. The calculation formula is as follows: ,in For flexible adjustment of resource nodes With source load fluctuation node The normalized dynamic time curvature distance of the node pairs; For nodes With nodes The original dynamic time bending distance of the joint feature sequence; For nodes The total length of the joint feature sequence; For nodes The total length of the joint feature sequence; the calculated normalized dynamic time bending distance is defined as the initial correlation difference between the flexible adjustment resource node and the source load fluctuation node. The larger the value, the greater the difference between the temporal fluctuation and topological correlation between the nodes and the lower the degree of interaction correlation. The smaller the value, the smaller the difference and the higher the degree of interaction correlation.

[0028] Step 221: Based on the initial correlation difference, combined with the adjustable capacity and response dead zone threshold of each flexible regulation resource node, and the fluctuation amplitude level of each source-load fluctuation node, calculate the data correlation coefficient of each node under different scheduling periods; perform time-series weighted average of the data correlation coefficients for all periods to obtain the comprehensive data correlation between nodes; specifically, for each group of flexible regulation resource nodes and source-load fluctuation nodes, complete the extraction and calibration of three types of key operating parameters. The first type is the adjustable capacity of the flexible regulation resource node. This parameter is extracted from the power grid equipment operation ledger and scheduling control procedures. It refers to the maximum active power adjustment amplitude that a single flexible regulation resource node can achieve within the safe operating range, including the sum of upward and downward adjustment capacity. It represents the upper limit of the node's power response capability to external fluctuations. The value is fixed and only slightly varies with the equipment operating status. The second category is the response dead zone threshold of flexible adjustment resource nodes. This parameter is a fixed value determined by the equipment factory calibration and on-site commissioning. It refers to the power range within which the flexible adjustment resource node will not trigger any adjustment action when the power fluctuation amplitude is less than this threshold. This is used to avoid equipment losses caused by frequent actions due to small fluctuations. The node will only perform power adjustment when the fluctuation exceeds this threshold. The third category is the fluctuation amplitude level of source-load fluctuation nodes. Based on the actual power fluctuation amplitude of the source-load fluctuation node within the preset scheduling cycle, the fluctuation amplitude is divided into three levels: high, medium, and low, and the values ​​are quantized. The fluctuation amplitude is less than 10% of the node's rated capacity, which is the low level, and the quantization value is 1; the fluctuation amplitude is between 10% and 30% of the node's rated capacity, which is the medium level, and the quantization value is 2; the fluctuation amplitude is more than 30% of the node's rated capacity, which is the high level, and the quantization value is 3. After quantization, it can be directly used in numerical calculations.

[0029] After parameter extraction, based on the initial correlation difference and the three types of parameters mentioned above, the data correlation coefficient for each scheduling period is calculated. This coefficient characterizes the real-time matching correlation between flexible adjustment resource nodes and source load fluctuation nodes at a given moment. The calculation formula is as follows: ,in denoted as the data correlation coefficient between the k-th flexible adjustment resource node and the source load fluctuation node at time t. The larger the coefficient, the closer the real-time correlation between the nodes. The initial association dissimilarity of the k-th node pair at time t is calculated by step 220; Let be the adjustable capacity of the flexible adjustment resource nodes in the k-th node pair at time t; Let t be the response dead zone threshold of the flexible adjustment resource node in the k-th node pair at time t; Let be the quantized value of the fluctuation amplitude level of the k-th group of nodes to the source load fluctuation node at time t; The absolute difference between the adjustable capacity and the response dead zone threshold reflects the effective adjustment capability of the node. It is calculated by iterating through all time segments within the preset scheduling period to obtain the set of data correlation coefficients for the entire time period. A time-series weighted average method is then used to fuse the data correlation coefficients for the entire time period, given the total number of time segments in the preset scheduling period. The total number of time slots is 63, consisting of 48 five-minute time slots for intraday rolling scheduling and 15 one-minute time slots for real-time scheduling; time sequence weight. A linear increment rule is adopted, meaning that the closer the time is to the current scheduling time, the larger the weight value. The initial weight of a time period is 1, and the weight value increases by 1 for each subsequent time period, ensuring that the weight distribution matches the real-time scheduling requirements. The formula for calculating the time-series weighted average is as follows: ,in , which represents the comprehensive data correlation degree of the k-th node pair, and is a stable representation value of the correlation characteristics throughout the entire time period; This is the weighted sum of the correlation coefficients of the data across all time periods. The sum of time-series weights for all time periods is used to calculate the overall data correlation degree.

[0030] Step 222: The comprehensive data correlation degree is converted into a primary coupling strength value within a preset interval according to a preset nonlinear mapping function. The primary coupling strength value is then subjected to neighborhood iterative averaging and boundary truncation processing to finally obtain the equivalent node coupling strength, which characterizes the degree of interaction between flexible regulation resource nodes and source-load fluctuation nodes. Specifically, this includes: performing nonlinear mapping processing to convert the comprehensive data correlation degree to a standard numerical range of 0 to 1 to obtain the primary coupling strength value; using an S-shaped nonlinear mapping function, which can amplify the distinguishability of high correlation degree values ​​and compress the differences of low correlation degree values, making the numerical distribution of coupling strength more consistent with the actual dispatch logic of the power grid; the calculation formula is as follows: ,in Let be the primary coupling strength value of the k-th node pair; This is the mapping scaling factor, fixed at 5, used to adjust the steepness of the mapping curve, ensuring that small changes in the overall data correlation can be clearly reflected in the primary coupling strength value; The comprehensive data correlation degree of the kth node pair is calculated in step 221. After mapping, the primary coupling strength value is limited to the range of 0 to 1. The closer the value is to 1, the tighter the coupling relationship between the nodes, and the closer the value is to 0, the looser the coupling relationship.

[0031] The primary coupling strength value is subjected to neighborhood iterative averaging to eliminate drastic jumps in the primary coupling strength value caused by local data anomalies and topological abrupt changes, making the numerical distribution smoother and more consistent with the continuous characteristics of the power grid topology. The neighborhood range is determined by selecting adjacent node pairs in the power grid topology with an electrical distance of less than a set threshold and a direct or indirect topological association, centered on the k-th node pair, to form the neighborhood node pair set of that node pair. Then set the iterative convergence condition, that is, the difference between the primary coupling strength values ​​after two adjacent iterations is less than 1. When the value is considered to have converged, the iteration stops, and the formula for calculating the iterative average is: ,in For the k-th node pair The initial coupling strength value after the next iteration; This represents the initial coupling strength value of the k-th node pair after the m-th iteration. This is the sum of the initial coupling strength values ​​of all node pairs in the neighborhood node pair set after the m-th iteration; The value represents the number of node pairs in the neighborhood node pair set. During the iteration process, each calculation merges the values ​​of the target node pair and the neighborhood node pairs, gradually smoothing out local mutations until the values ​​converge. Boundary truncation is performed to constrain the range of the converged values ​​to ensure the standardization of the final result. The converged values ​​are compared with the standard interval of 0 to 1. If the value is less than 0, it is forcibly assigned the value of 0; if the value is greater than 1, it is forcibly assigned the value of 1; if the value is within the interval of 0 to 1, it remains unchanged. The final value after boundary truncation is the equivalent node coupling strength.

[0032] Step 223: Based on the equivalent node coupling strength, construct a coupling potential energy edge between each pair of coupled flexible regulation resource nodes and source load fluctuation nodes, and assign an initial density weight to each coupling potential energy edge, resulting in a set of coupling potential energy edges with initial density weights. Specifically, this includes: for each pair of coupled flexible regulation resource nodes and source load fluctuation nodes, constructing a coupling potential energy edge connecting the two nodes, with the two nodes as endpoints. The coupling potential energy edge is used to characterize the correlation between power transfer and regulation response between nodes; assigning an initial density weight to each coupling potential energy edge, the initial density weight being proportional to the equivalent node coupling strength and inversely proportional to the electrical distance between nodes, where the electrical distance... Per-unit values ​​are used, calculated based on the per-unitized line impedance, which is dimensionless. The calculation formula is as follows: ,in The initial tightness weights of the coupling potential energy edge; Let be the equivalent node coupling strength of the k-th node pair; To flexibly adjust the electrical distance between resource nodes and source load fluctuation nodes, all constructed coupled potential energy edges are integrated according to the node numbering order and topological connection relationship to form a set of coupled potential energy edges with initial density weights.

[0033] Step 224: Based on the coupled potential energy edge set, and combined with the upper and lower limits of each node's own power regulation and ramp rate limits as node constraint boundary values, traverse all branch power flow constraints and node power balance constraints in the power grid, and calculate the Lagrange multiplier estimate for each constraint in the current scheduling period; mark constraints with Lagrange multiplier estimates greater than a preset threshold as active constraints, and sort the active constraints in descending order according to the size of the Lagrange multiplier estimates to obtain a constraint tightness ranking list; specifically, this includes: extracting the self-operation constraints of each flexible regulation resource node and source-load fluctuation node, using the node power regulation upper limit, power regulation lower limit, and ramp rate limits as node constraint boundary values, then traversing all branch power flow constraints and node power balance constraints in the power grid, constructing constraint equations for each constraint, and calculating the Lagrange multiplier estimate for each constraint in the current scheduling period based on the coupled potential energy edge set and node constraint boundary values. The Lagrange multiplier estimate is used to characterize the degree of constraint tension; the larger the value, the more urgent the constraint. The specific calculation method is as follows: establish the Lagrange function. ,in The objective function of power grid dispatch (e.g., minimizing adjustment costs). ≤0 is the first Constraint functions, For the corresponding Lagrange multipliers, the estimated values ​​of the Lagrange multipliers are obtained by solving the KKT conditions under the current scheduling period. ,Right now This estimated value directly reflects the degree of constraint tension. A Lagrange multiplier judgment threshold is set. This threshold is not arbitrarily set, but determined based on engineering practice of new power systems with high penetration of wind and solar power. Through simulation statistics of thousands of wind and solar grid-connected dispatching examples, critical judgment criteria for grid dispatching constraint tension, and sensitivity analysis of flexible regulation resource response, combined with the general judgment standard for constraint activity in two-stage robust optimization dispatching in the power industry, when the Lagrange multiplier estimate is greater than 0.3, the constraint will have a significant rigid restriction on the power regulation rate, dispatching command execution, and grid power flow security, directly affecting dispatching decisions and system stable operation; when the Lagrange multiplier estimate is less than or equal to 0.3, the constraint is in a relaxed state and will not have a substantial restriction on dispatching operations. Based on the aforementioned engineering and theoretical foundations, the preset threshold for the Lagrange multiplier is set to 0.3. Constraints are classified according to this threshold: constraints with Lagrange multiplier estimates greater than 0.3 are marked as active constraints; these are rigid and urgent constraints under the current scheduling period and are the primary restrictions that power regulation and scheduling decisions must satisfy. Constraints with Lagrange multiplier estimates less than or equal to 0.3 are marked as inactive constraints; these constraints are loose and redundant and do not need to be core constraints for scheduling control. All active constraints are sorted according to a descending order of Lagrange multiplier estimates, with larger estimates indicating higher urgency and higher ranking. This results in a constraint tightness ranking list.

[0034] Step 225: Based on the tightness ranking list, assign the tightness values ​​of active constraints to the relevant nodes or coupled potential energy edges, while setting the tightness values ​​of inactive constraints to 0; perform spatial smoothing on all tightness values, and superimpose the smoothed tightness distribution onto the basic topology graph composed of nodes and coupled potential energy edges to obtain the final constraint force field topology graph; specifically, this includes: according to the constraint tightness ranking list, assigning tightness values ​​to all grid nodes and coupled potential energy edges; for each active constraint, first locate the corresponding grid node or coupled potential energy edge associated with the constraint, then directly use the estimated Lagrange multiplier value corresponding to the active constraint as the tightness value of the associated node or coupled potential energy edge to complete the assignment; for all grid nodes and coupled potential energy edges associated with inactive constraints, uniformly set their tightness values ​​to zero, thus completing the initial tightness assignment of all nodes and coupled potential energy edges; after completing the initial assignment, perform spatial smoothing on the tightness values ​​of all grid nodes and coupled potential energy edges. Taking each grid node or coupled potential energy edge to be processed as the center, a corresponding spatial neighborhood is defined. The original density values ​​of all nodes and coupled potential energy edges in this neighborhood are used as the basis for calculation. The values ​​in the neighborhood are weighted, with higher weights assigned to the values ​​at the center and progressively lower weights assigned to the values ​​at the edges. The weighted result is used as the smoothed density value of the central node or coupled potential energy edge. This smoothing process is repeated for all nodes and coupled potential energy edges to obtain a stable and continuous density spatial distribution. Taking all operating nodes in the grid as vertices of the topology graph and all coupled potential energy edges constructed in step 223 as connecting edges of the topology graph, a basic topology graph is constructed to characterize the distribution and coupling connection relationship of grid nodes. The density distribution result after spatial smoothing is then superimposed on this basic topology graph. The density values ​​of coupled potential energy edges are intuitively represented by the edge weights, and the density values ​​of grid nodes are intuitively represented by the node's own attributes, ultimately forming a constraint force field topology graph.

[0035] By analyzing node interaction features, quantifying and mapping coupling strength, and dynamically evaluating constraint tightness, a constraint force field topology diagram that fits the actual operating state of the power grid is constructed. This diagram depicts the coupling relationship between flexible regulation resources and source-load fluctuation nodes, as well as the tightness of scheduling constraints, thereby improving the adaptability of two-stage robust scheduling and the stability of power grid operation.

[0036] In a preferred embodiment of the present invention, step 3 above involves optimizing the power regulation rate of the flexible regulation resource node based on the constraint force field topology diagram. During the optimization process, the rate trajectory is discretized into a time series using the upper and lower limits of the jerk of the power regulation rate. A power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is then dynamically matched with a preset power grid flow response inertia coefficient. The step size change rate and attenuation factor of the two-stage correction scheduling command are calculated to generate a robust scheduling command sequence, which may include: In this embodiment of the invention, step 330 involves determining the upper and lower limits of jerk and rate constraints for the power regulation rate of each flexible adjustment resource node within the scheduling cycle, based on the density weight of each coupled potential energy edge in the constraint force field topology diagram and the power regulation capability boundary of each flexible adjustment resource node itself. This process, along with the node power balance boundary conditions and branch power flow boundary conditions defined by the constraint force field topology diagram, forms the initial feasible region for the power regulation rate of each node. Specifically, this includes: first, reading the constraint force field topology diagram and extracting the density weight of each coupled potential energy edge. The density weight directly reflects the tightness of the scheduling constraints between nodes; a higher weight corresponds to a stricter power regulation constraint on the flexible adjustment resource node. Simultaneously, retrieving the inherent operating parameters of each flexible adjustment resource node, including the rated power regulation range of the equipment, the allowable power regulation rate limit, the allowable jerk limit, and the jerk... The power regulation rate is the rate of change of the power regulation rate, a core parameter that limits rate abrupt changes and avoids equipment impact. Combined with the tightness weight, the inherent parameters are dynamically corrected. The higher the tightness weight, the more conservative the upper and lower limits of the power regulation rate and jerk. This determines the upper limit of the power regulation rate, the lower limit of the power regulation rate, the upper limit of the jerk, and the lower limit of the jerk for each flexible regulation resource node throughout the entire scheduling cycle. Then, based on the constraint distribution of nodes and coupling potential energy edges in the constraint force field topology diagram, the power balance boundary conditions that each node must satisfy are extracted, namely, the sum of the node's input and output power must remain constant, and the power flow boundary conditions that each branch must satisfy, namely, the branch's transmission power must not exceed the safe operating limit. Integrating all the above constraints, the allowable range of power regulation rate and the legal operating boundary of each flexible regulation resource node are defined, ultimately forming the initial feasible region of the node's power regulation rate.

[0037] Step 331: Based on the initial feasible region of the power regulation rate, the scheduling period is discretized into multiple continuous time segments at equal intervals, and the power regulation rate values ​​in each time segment are arranged in chronological order to form the discrete time sequence of the power regulation rate to be solved. Specifically, this includes: the preset scheduling period is divided into two stages: intraday rolling scheduling and real-time scheduling. The intraday rolling scheduling stage lasts for 4 hours with a time resolution of 5 minutes, and the real-time scheduling stage lasts for 15 minutes with a time resolution of 1 minute. Using the initial feasible region as the legal boundary for rate values, the continuous scheduling period is uniformly and equally discretely divided according to the time resolution of the two stages. The intraday rolling scheduling stage is divided into independent time segments of 5 minutes each. The real-time scheduling phase divides the time into independent time segments of 1 minute each, breaking down the complete continuous scheduling cycle into multiple discrete time segments with a fixed number, continuous time sequence, and no overlap or omissions. For each flexible adjustment resource node, a power adjustment rate value to be solved is matched for each discrete time segment, with its own exclusive initial feasible domain as a value constraint. This value can only be taken within the boundary range of the initial feasible domain to ensure the legality and feasibility of the rate value. The power adjustment rate values ​​to be solved for all discrete time segments are arranged in the order of the scheduling cycle time, finally forming a discrete time sequence of power adjustment rates with a length that perfectly matches the total number of discrete time segments and a complete and continuous time sequence.

[0038] Step 332 transforms the jerk upper and lower limit constraints into second-order difference inequalities between rate values ​​in three adjacent time segments of the power regulation rate discrete-time series; transforms the rate upper and lower limit constraints into boundary inequalities for rate sequence values ​​in each time segment; and transforms the node power balance boundary conditions and branch power flow boundary conditions into a set of inequalities that the power regulation rate discrete-time series must satisfy. Specifically, this includes: determining the core transformation principle; the jerk upper and lower limit constraints, power regulation rate upper and lower limit constraints, node power balance boundary conditions, and branch power flow boundary conditions determined in step 330 are all continuous approximations. The jerk constraint cannot be directly applied to the optimization solution of discrete time series. It needs to be discretized and transformed into inequalities or a system of inequalities. The upper and lower limits of the jerk are transformed into discretized inequalities. The jerk is the rate of change of the power regulation rate. Its core function is to limit abrupt rate changes and avoid equipment shock. In continuous scenarios, it is represented by the second derivative of the rate with respect to time. In discrete time series scenarios, the second difference is calculated using the rate values ​​of three adjacent discrete time segments to approximate the continuous jerk. Specifically, let the rate values ​​of the t-th, t-1-th, and t-2-th time segments in the discrete time series be respectively... , , The second-order difference is calculated as follows: This value is proportional to the jerk in a continuous scene. Limiting this second-order difference result to the upper and lower limits of the jerk transforms it into a second-order difference inequality. ,in For the upper and lower limits of jerk, The duration of the time segment is defined. The upper and lower limits of the power adjustment rate are transformed into boundary inequalities. The upper and lower limits of the power adjustment rate are rigid constraints on the flexible adjustment resource nodes. The original constraint is that the continuous rate always stays within the set upper and lower limits. Since each time segment in the discrete time series corresponds to an independent rate value... Therefore, boundary inequalities are directly constructed for each time segment, requiring that the rate value of each segment must strictly fall between the upper and lower limits of the power regulation rate determined in step 330, ensuring that the rate value at each discrete moment meets the equipment operation requirements and does not exceed the constraint boundaries. The node power balance boundary conditions are transformed into a system of linear inequalities. The core requirement of node power balance is that the sum of the node input power, output power, and regulation power remains constant, which is a rigid constraint of steady-state grid operation. Since the power regulation rate corresponds to the power regulation increment per unit time, combined with the duration of each discrete time segment, the rate value can be transformed into the power regulation increment of the corresponding segment. Then, combined with the node initial power and source-load fluctuation power, a linear equation about the discrete time series of the power regulation rate is constructed. A set of inequalities ensures that the sum of the rate-related adjustment increments across all time segments, in conjunction with the initial power of nodes and the fluctuating power of source loads, meets the power balance requirements and avoids power mismatch. The branch power flow boundary conditions are transformed into a set of linear inequalities. These conditions require that the branch transmission power not exceed its safe operating limit. Changes in the node power adjustment rate directly trigger corresponding changes in the branch power flow. Combining the node relationships in the constraint force field topology diagram, the impact of the rate adjustment of each flexible adjustment resource node on the associated branch power flow is determined. The branch power flow safety limit constraint is transformed into a set of linear inequalities concerning the discrete-time series of power adjustment rates, ensuring that the adjustment actions corresponding to the rate series do not cause the branch power flow to exceed the safety limit, thus guaranteeing the safe operation of the power grid branches.

[0039] Step 333: Based on the discrete-time series of power regulation rates and the various inequalities and equality constraints after transformation, a quadratic programming problem is constructed with the joint optimization objectives of minimizing the sum of the acceleration changes in the rate series over all time segments and minimizing the total regulation energy consumption. The second-order difference inequality, rate boundary inequality, and all sets of inequalities are used as constraints for the quadratic programming problem. Solving the quadratic programming problem yields the final power regulation rate value for each flexible regulation resource node in each discrete time segment. Specifically, the first optimization objective is to minimize the sum of the acceleration changes in the power regulation rate series over all discrete time segments. The core purpose is to ensure a smooth and abrupt power regulation process, avoiding electrical or mechanical shocks to flexible regulation resource equipment due to excessively rapid rate changes, extending equipment lifespan, and preventing grid power oscillations. The second optimization objective is to minimize the total regulation energy consumption of flexible regulation resources. The core purpose is to improve the operational economy of grid dispatch, reduce energy losses during flexible resource regulation, and control dispatch costs. To achieve the above dual-objective optimization, a corresponding objective function is constructed. To unify the dimensions, a time reference is introduced. =1 minute, and all rate-related variables are dimensionless, the specific expression is as follows: ,in It is a dimensionless energy consumption coefficient (the energy required to adjust a unit power per unit). The weighting coefficient for the change in jerk is set according to the power grid dispatch safety priority. It is used to balance the stability target and is usually set to 0.65 to prioritize the stability of regulation. To adjust the energy consumption weighting coefficient, a value of 0.35 is set, which is consistent with... The sum of these is 1, used to balance the weights of economic objectives; For the first The second-order difference of the power regulation rate for each time segment is calculated from the rate values ​​of three adjacent time segments, and characterizes the drasticness of the rate change. For the first The power adjustment rate of each time segment is the core object of this optimization. For the first The energy consumption coefficient of flexible regulation resources in each time segment is extracted from the operating parameters of the flexible regulation resource equipment, reflecting the energy consumption level of the equipment's unit power regulation in different time segments. During the optimization process, all kinds of constraints transformed in step 332 must be used as optimization restrictions to ensure that the optimization results are legal and feasible. Specifically, these include the second-order difference inequality corresponding to the jerk constraint, the boundary inequality corresponding to the power regulation rate constraint, the set of linear inequalities corresponding to the node power balance constraint, and the set of linear inequalities corresponding to the branch power flow constraint. All constraints work together to limit the legal range of the discrete-time series of power regulation rates, ensuring that the optimized rate value meets both the equipment operation constraints and the requirements for safe operation of the power grid. After the constraints and objective function are determined, the quadratic programming optimization scheme is solved. During the solution process, the value of the discrete-time series of power regulation rates is continuously adjusted by iteratively approaching the final solution to ensure that the result of each iteration meets all constraints. At the same time, the objective function value is gradually reduced until the objective function reaches its minimum value and the iteration converges. The power regulation rate value of each flexible regulation resource node in each discrete time segment obtained at this time is the final power regulation rate value.

[0040] Step 334: Based on the final power regulation rate value, the sequence is spliced ​​in ascending order of time to obtain the complete discrete rate trajectory; cubic spline smoothing filtering is performed on the discrete rate trajectory to finally obtain the power regulation rate curve; specifically, this includes: performing a time-series splicing operation of discrete rate values. In step 333, for each flexible regulation resource node, the final power regulation rate value corresponding to all discrete time segments within the scheduling period has been solved. Each discrete time segment corresponds to a unique rate value, and all rate values ​​satisfy the various constraints transformed in step 332, meeting the requirements of equipment operation and grid safety. During the splicing process, for each flexible regulation resource node, the final power regulation rate value of the first discrete time segment is strictly spliced ​​in ascending order of time according to the preset scheduling period. Starting from the trajectory, the rate value of each subsequent discrete time segment is sequentially connected to the rate value of the previous time segment, forming a complete discrete rate trajectory. This discrete rate trajectory fully covers the entire scheduling cycle and is completely consistent with the chronological order of the discrete time segments. Each trajectory point corresponds to a specific time segment and a corresponding rate value, accurately reflecting the optimized power regulation rate target at each moment. However, since there are independent divisions between discrete time segments, the rate values ​​of two adjacent time segments may change suddenly, i.e., a numerical jump phenomenon occurs. This jump can cause electrical or mechanical shocks when the flexible regulation resources perform power regulation actions, making it impossible to directly use for continuous power regulation control of the power grid. Therefore, the discrete rate trajectory must be smoothed.

[0041] For the spliced ​​discrete velocity trajectories, a cubic spline smoothing filter is applied. The specific process is as follows: All interpolation nodes of the discrete velocity trajectory are determined. Each spliced ​​discrete velocity value and its corresponding time node are used as the basic nodes for cubic spline interpolation. Each node contains a time coordinate (corresponding to the start time of the discrete time segment) and a velocity coordinate (corresponding to the final power adjustment rate value of that time segment). All nodes are arranged sequentially in chronological order, forming an interpolation node sequence. For two adjacent nodes in the interpolation node sequence, an independent cubic polynomial fitting relationship is constructed. This relationship is used to fit the velocity change trend between the two nodes, ensuring that the fitting curve smoothly connects the velocity values ​​of the two nodes. When constructing the cubic polynomial, the principle of continuity and differentiability is strictly followed, meaning that the first and second derivatives of two adjacent cubic polynomials at the connection node remain continuous, avoiding new numerical jumps or... Curve abrupt changes ensure the smoothness and continuity of the entire fitted curve. Simultaneously, curvature adjustment using a cubic polynomial allows the fitted curve to closely match the overall trend of the discrete rate trajectory, ensuring it doesn't deviate from the final power regulation rate value obtained through optimization while effectively eliminating rate jumps between adjacent nodes. The cubic polynomial fitting relationships of all adjacent nodes are integrated, seamlessly connecting the fitted curves of each segment to form a continuous rate curve covering the entire scheduling cycle. During integration, the connection points of each segment curve are verified one by one to ensure no discontinuities or abrupt changes, and a smooth transition in rate changes. Simultaneously, the curve is validated as a whole to confirm that all values ​​of the smoothed rate curve do not exceed the upper and lower limits of the power regulation rate and jerk determined in step 330, still satisfying various constraints and avoiding constraint violations due to smoothing. After completing the cubic spline smoothing filtering process, the resulting continuous rate curve is the power regulation rate curve.

[0042] Step 335: Based on the final power regulation rate value of each flexible regulation resource node in each discrete time segment of the power regulation rate curve, calculate the corresponding power regulation increment sequence by integration; convolve the power regulation increment sequence with the preset power flow response inertia coefficient to obtain the power flow response delay distribution of each flexible regulation resource node after the power regulation action; specifically, this includes: performing numerical integration on the power regulation rate curve obtained in step 334. Since the power regulation rate curve is a continuously differentiable smooth curve, and the duration of each discrete time segment is fixed (the duration of each time segment for intraday rolling scheduling is 5 minutes, and the duration of each time segment for real-time scheduling is 1 minute), the interval of numerical integration strictly corresponds to each time segment divided in step 331. Discrete time segments refer to the process of integrating the power regulation rate curve over each discrete time segment, using the start and end times of that segment as the upper and lower limits of integration. The specific integration process is as follows: For each flexible regulation resource node, each discrete time segment within the scheduling cycle is processed one by one. The power regulation rate curve segment corresponding to that time segment is used as the integration object. The area enclosed by the rate curve and the time axis within that time segment is calculated using numerical integration methods. This area represents the power regulation increment of the flexible regulation resource node within that discrete time segment. Since the power regulation rate is the amount of power regulation per unit time, the integral of the rate over time is essentially the total power regulation increment within that time segment, reflecting the actual power regulation amplitude performed by the node within each time segment.

[0043] After completing the numerical integration of all discrete time segments, the power adjustment increments corresponding to each time segment are arranged sequentially according to the time increment of the scheduling cycle to form a complete power adjustment increment sequence. This sequence corresponds one-to-one with the discrete time segments divided in step 331, with each sequence element corresponding to a specific time segment and a corresponding power adjustment increment. The preset grid power flow response inertia coefficient is retrieved, which is defined as the time constant of the grid power flow response. (Unit: minutes) represents the time required for the power flow response to reach 63.2% of the change after a power regulation action. A larger coefficient indicates a slower response and longer lag time in the power flow to the regulation action; a smaller coefficient indicates a faster power flow response. In the convolution operation, the power regulation increment sequence is concatenated with an exponential decay kernel. Discrete convolution is performed, and the specific operation process is as follows: each incremental value in the power regulation increment sequence is weighted and superimposed with the power flow response inertia coefficient. The weight allocation is based on the time sequence of the regulation actions, that is, the influence of the regulation actions executed earlier on the subsequent power flow response gradually decreases, and the influence of the regulation actions executed later on the current power flow response is the most significant. Through this weighted superposition, the entire process of power regulation actions from triggering to the power flow reaching a steady state is completely restored, including key features such as response lag time, response amplitude change, and stabilization time. After the convolution operation is completed, the result is the power flow response delay distribution corresponding to each flexible regulation resource node after executing the power regulation action.

[0044] Step 336: Based on the power flow response delay distribution, extract the response lag time constant and overshoot characteristic parameters for each flexible regulation resource node; input the response lag time constant and overshoot characteristic parameters into a preset step size change rate calculation function to obtain the step size change rate between the first-stage prediction step size and the second-stage correction step size for each flexible regulation resource node in the two-stage correction scheduling; based on the power flow response delay distribution, with time as the horizontal axis and power flow response amplitude as the vertical axis, extract two core characteristic parameters (response lag time constant and overshoot characteristic parameters). The first extracted characteristic parameter is the response lag time constant, which is primarily used to reflect how quickly the power flow response reaches a stable value, and its magnitude directly corresponds to the lag characteristics of the power flow response. The specific extraction process is as follows: First, determine the stable value of the power flow response from the power flow response time delay distribution. The stable value is a constant value that tends to stabilize after the power flow response decays. It can be determined by taking the average value of the power flow response amplitude in the later stage of the time delay distribution (the stage when the power flow has no obvious fluctuations). Then, find the time point corresponding to the power flow response amplitude rising to 63.2% of the stable value (the standard threshold for characterizing the response speed of a first-order inertial system in engineering). The difference between this time point and the trigger time of the power regulation action is the response lag time constant. The larger the value of this parameter, the slower the power flow response reaches the steady state and the more obvious the lag characteristic. The smaller the value, the faster the power flow response speed and the weaker the lag characteristic. The extraction result directly reflects the response sensitivity of the power flow to the power regulation action.

[0045] The second extracted feature parameter is the overshoot feature parameter. This parameter primarily reflects the magnitude of the power flow response exceeding the stable value. Its value directly corresponds to the oscillation characteristics of the power flow response and is a key parameter for avoiding dispatch command overshoot. The specific extraction process is as follows: First, determine the stable value of the power flow response (consistent with the stable value extracted when extracting the response lag time constant); then find the peak value of the power flow response (i.e., the maximum value of the power flow response amplitude) from the time delay distribution; by calculating the difference between the peak value and the stable value, and then dividing by the stable value, obtain the overshoot (expressed as a percentage), i.e., overshoot feature parameter = (power flow response peak value - power flow response stable value) / power flow response stable value × 100%. The larger the value of this parameter, the greater the magnitude of the power flow response exceeding the stable value, and the higher the risk of system oscillation; the smaller the value, the more stable the power flow response, and the lower the risk of oscillation. If the overshoot is 0, it indicates that the power flow response has no overshoot and has smoothly reached a stable state. After extracting the two core feature parameters, they are input into the preset step size change rate calculation logic. The specific calculation process is as follows: [Preset step size change rate calculation logic]. A nonlinear mapping method is used to comprehensively quantify the response lag time constant and overshoot characteristic parameters. Combined with the safety priority of power grid dispatch, the weights of the two characteristic parameters are set, with the response lag time constant having a higher weight than the overshoot characteristic parameter. This prioritizes adapting to the lag characteristics of the power flow response. When the response lag time constant is larger (slower power flow response) and the overshoot characteristic parameter is larger (higher system oscillation risk), the calculated step size change rate is smaller. This slows down the transition speed from the first-stage prediction step size to the second-stage correction step size, preventing abrupt changes in the correction step size from causing further overshoot and increased lag in the power flow response. Conversely, when the response lag time constant is smaller (faster power flow response) and the overshoot characteristic parameter is smaller (lower system oscillation risk), the calculated step size change rate is larger, accelerating the step size transition speed and improving the response efficiency of dispatch correction. After calculation, the result is the step size change rate for each flexible regulation resource node in the two-stage correction dispatch. This change rate is a value between 0 and 1, used to clarify the transition ratio between the first-stage prediction step size and the second-stage correction step size, achieving a smooth connection between the two step sizes.

[0046] Step 337: Based on the overshoot characteristic parameters and the peak time characteristics in the power flow response delay distribution, determine the attenuation factor of each flexible regulation resource node in the two-stage correction scheduling. Specifically, this includes extracting the peak time characteristics from the power flow response delay distribution. Peak time is a core auxiliary parameter characterizing the speed and oscillation characteristics of the power flow response. Together with the overshoot characteristic parameters, it determines the value of the attenuation factor. The power flow response delay distribution obtained in Step 335 clearly presents the complete process of the power flow response from triggering, rising, reaching its peak, and gradually attenuating to stability. Peak time is the time interval from the moment the power regulation action is triggered until the power flow response amplitude reaches its maximum value (peak). The specific extraction process is as follows: First, locate the maximum value of the power flow response amplitude from the delay distribution, then record the time coordinate corresponding to this peak value. Subtract this time coordinate from the time coordinate of the moment the power regulation action is triggered; the resulting time difference is the peak time. The core significance of peak time is reflecting the power flow response. The faster the peak time reaches the maximum amplitude, the faster the power flow response rises, and the greater the risk of rapid overshoot and oscillation in the system. Conversely, the longer the peak time, the smoother the power flow response rises, the more stable the system operation, and the lower the risk of overshoot. The linkage between the overshoot characteristic parameter and the peak time characteristic is clearly defined. Both serve as the basis for determining the attenuation factor, and both are based on the calculation results of previous steps to ensure data consistency and correlation. The overshoot characteristic parameter (extracted in step 336) reflects the magnitude of the power flow response exceeding the stable value and is directly related to the risk of system oscillation. The peak time characteristic reflects the speed at which the power flow response reaches its peak and is directly related to the system response speed. The combination of these two parameters comprehensively characterizes the power flow response characteristics of the power grid, providing support for the selection of the attenuation factor. The core function of the attenuation factor is to regulate the attenuation rate of real-time correction commands in two-stage correction scheduling, avoiding power flow overshoot caused by sudden command changes, while ensuring the response efficiency of correction commands. Its value must be highly adapted to the power flow response characteristics.

[0047] Based on the overshoot characteristic parameters and peak time characteristics, a two-stage correction attenuation factor is determined for each flexible regulation resource node through a preset quantization adaptation logic. The specific value rules are as follows: When the overshoot characteristic parameter is large (greater than the preset threshold, usually set to 10%) and the peak time is short (less than the preset threshold, preset to 3 minutes for intraday rolling dispatch and 1 minute for real-time dispatch), it indicates that the power flow response is prone to rapid overshoot and has a high risk of oscillation. In this case, the attenuation factor value is relatively small (usually between 0.1 and 0.3). A smaller attenuation factor can slow down the attenuation rate of real-time correction commands, suppress rapid changes in command amplitude, and avoid further aggravation of power flow overshoot due to excessively strong or rapidly changing correction commands, thus ensuring stable system operation. When the overshoot characteristic parameter is moderate (between 5% and 10%) and the peak time is moderate (3-5 minutes for intraday rolling dispatch and 1-2 minutes for real-time dispatch), it indicates that the power flow response is stable. At the critical state of oscillation, the attenuation factor is moderately valued (usually between 0.3 and 0.7), balancing oscillation suppression and response efficiency. This avoids exacerbating overshoot without causing untimely correction due to excessive attenuation. When the overshoot characteristic parameter is small (less than 5%) and the peak time is long (more than 5 minutes for intraday rolling dispatch and more than 2 minutes for real-time dispatch), it indicates that the power flow response is stable and the overshoot risk is low. In this case, the attenuation factor is larger (usually between 0.7 and 0.9). A larger attenuation factor can accelerate the attenuation speed of real-time correction commands. It should be noted that the value of the attenuation factor is always strictly limited to between 0 and 1, excluding the two endpoints of 0 and 1. When the value is close to 0, the command attenuation speed is extremely slow, focusing on suppressing overshoot. When the value is close to 1, the command attenuation speed is extremely fast, focusing on improving response efficiency. When the value is 0 or 1, it will cause no attenuation or instantaneous attenuation of commands, causing system fluctuations. Therefore, the value range must be strictly controlled.

[0048] Step 338 involves multiplying the step size change rate and attenuation factor by the corresponding basic dispatch command amplitude to obtain the predicted correction command sequence and real-time correction command sequence for each flexible regulation resource node at each moment within the dispatch cycle. Specifically, this includes retrieving the amplitude parameter of the grid's basic dispatch command. This amplitude parameter is a core parameter preset by the grid dispatch center based on source-load prediction data, system power balance requirements, and grid safety operation constraints. It serves as the basic benchmark for constructing the two types of correction command sequences and is not arbitrarily set. Its value is determined comprehensively based on different stages of intraday rolling dispatch and real-time dispatch, combined with factors such as the rated regulation capacity of the flexible regulation resource node, the current operating load level of the grid, and the predicted value of source-load fluctuations. It is stored in the grid dispatch command parameter library and can be directly retrieved and used. This amplitude parameter corresponds to a unique benchmark amplitude for each flexible regulation resource node and each discrete time segment, and is consistent with the amplitude parameter defined in step 331. The discrete time segments are mapped one-to-one, and the predicted correction command sequence for each moment within the scheduling cycle is calculated. This sequence corresponds to the first stage of the two-stage robust scheduling, the predicted command stage. Its core function is to pre-correct the basic scheduling command based on the power flow response lag characteristics. The specific calculation process is as follows: For each flexible adjustment resource node, each discrete time segment within the scheduling cycle is matched one by one. The step size change rate corresponding to the node (calculated in step 336, which corresponds one-to-one with the node and reflects the transition ratio between the predicted step size and the correction step size) is multiplied with the amplitude of the power grid basic scheduling command corresponding to the time segment. Since the step size change rate is a value between 0 and 1, the result of the multiplication is the predicted correction command value of the node within the time segment. After completing the calculation of all discrete time segments, the predicted correction command values ​​corresponding to each time segment are arranged in ascending order of time to form a complete predicted correction command sequence.

[0049] Simultaneously, the real-time correction command sequence for each moment within the scheduling cycle is calculated synchronously. This sequence corresponds to the second stage of the two-stage robust scheduling, the real-time correction stage. Its core function is based on the overshoot and peak time characteristics of the power flow response. The specific calculation process is as follows: For each flexible regulation resource node and each discrete time segment, the attenuation factor corresponding to the node (determined in step 337, corresponding one-to-one with the node, reflecting the attenuation rate of the real-time correction command) is multiplied with the amplitude of the power grid basic scheduling command corresponding to the time segment. The attenuation factor is also between 0 and 1. The result of the multiplication is the real-time correction command value of the node within the time segment. After completing the calculation of all time segments, the real-time correction command values ​​corresponding to each time segment are arranged in ascending order of time to form a complete real-time correction command sequence.

[0050] Step 339 involves concatenating the predicted correction command sequence and the real-time correction command sequence according to the scheduling time sequence, and smoothing the command values ​​at the concatenation boundary to obtain a robust scheduling command sequence. Specifically, this includes strictly adhering to the timing arrangement rules of the two-stage power grid scheduling to complete the sequence concatenation. The predicted correction command sequence corresponds to the pre-decision control stage in the early part of the scheduling cycle, while the real-time correction command sequence corresponds to the dynamic correction stage in the later part of the scheduling cycle. Using time increments as the arrangement logic, the predicted correction command sequence is arranged as the first part of the overall command, and the real-time correction command sequence is arranged as the second part. The two sequences are connected end-to-end, completely covering the entire scheduling duration, ensuring that the timing of each discrete time segment is corresponding and the data is complete and without missing values, thus completing the basic command combination. Since the two command sequences are based on different criteria, there will be significant differences in the command values ​​before and after the concatenation point. Direct connection can easily cause sudden changes in commands, leading to equipment regulation shocks and power grid power flow fluctuations. Therefore, it is necessary to perform a smooth transition at the concatenation boundary. A smooth transition process is implemented at the boundary position. Linear interpolation is used to calculate the intermediate transition command values ​​to eliminate numerical jumps and locate the connection point between the two sequences. The last set of predicted correction command values ​​before the boundary and the first set of real-time correction command values ​​after the boundary are extracted. These two values ​​serve as the start and end references for the transition interval. During the connection period between the two command segments, several transition calculation nodes are evenly divided. Following a linear and uniformly changing calculation method, the intermediate command values ​​corresponding to each transition node are solved sequentially. This allows the command values ​​to change smoothly and gradually from the end of the first segment to the beginning of the second segment, completing the value switching step by step and avoiding sudden increases or decreases in commands. Through the above interpolation filling process, the entire integrated command sequence is made continuous and connected without data breakpoints or numerical abrupt changes. After the overall processing is completed, the change trend and value range of the entire command sequence are uniformly verified to ensure that the command change rate meets the equipment operation limits and power grid constraints. Finally, the integrated and optimized sequence forms a complete robust dispatch command sequence.

[0051] Based on the constraint force field topology, the flexible resource power regulation rate is optimized. Taking into full account the jerk constraint and the inertia of the power flow response of the grid, the dual optimization of smooth regulation and low energy consumption is achieved through quadratic programming. Two-stage robust scheduling instructions are generated to improve the execution accuracy of wind and solar grid-connected scheduling and the stability of system operation.

[0052] In a preferred embodiment of the present invention, step 4 above involves sending the robust scheduling command sequence to the grid edge control node, collecting real-time wind and solar power output deviations, using the real-time wind and solar power output deviations to perform feedback correction and compensation on the robust scheduling command sequence, updating the power imbalance gradient distribution in the constraint force field topology diagram, and obtaining the final two-stage robust optimization scheduling strategy for wind and solar power grid integration. This strategy may include: In this embodiment of the invention, step 440 involves sending a robust scheduling instruction sequence to the grid edge control node to perform node-level instruction parsing and local action mapping to obtain the initial power setting trajectory. Specifically, this includes: using an edge communication protocol adapted to real-time grid scheduling, sending the robust scheduling instruction sequence in batches to the corresponding grid edge control nodes. Each edge control node manages one or more flexible regulation resource nodes. The edge control node performs node-level instruction parsing on the received robust scheduling instruction sequence, breaking down the instruction content of each discrete time segment, eliminating invalid and redundant information, and extracting key parameters such as power regulation targets, regulation timing, and safety constraints corresponding to each time segment. Combined with the type of flexible regulation resource node it manages, it adapts to the operating control logic of different devices. For example, for edge nodes managing energy storage systems, the focus is on parsing charging and discharging power instructions; for edge nodes managing adjustable loads, the focus is on parsing load regulation amplitude instructions, ensuring that the parsed instruction information matches the actual operating characteristics of the flexible regulation resource nodes. Based on the parsed instruction parameters, a local action mapping operation is performed to transform the abstract scheduling instructions into specific power setting values ​​for each flexible regulation resource node in each discrete time segment, arranged sequentially in ascending order of the scheduling cycle time to form the initial power setting trajectory.

[0053] Step 441: A real-time differential comparison is performed between the initial power setting trajectory and the measured power output data collected from the wind and solar power plants on-site to obtain the real-time wind and solar power output deviation. Specifically, this includes: establishing a real-time power output acquisition system for wind and solar power plants; deploying power sensors, data acquisition terminals, and other equipment at each grid-connected wind and solar power plant; synchronously acquiring measured power output data for each discrete time segment at a time resolution consistent with the scheduling cycle; covering all wind and solar power generation units to ensure comprehensive data acquisition; preprocessing the measured data during acquisition to remove abrupt changes and missing values ​​caused by sensor failures or abnormal data transmission; supplementing missing data using linear interpolation; and re-interpolating abrupt changes using the 3σ criterion to ensure the continuity and accuracy of the measured power output data. The preprocessed real-time wind and solar measured power output data sequence is as follows: ,in This is a sequence of real-time measured power output data for wind and solar power plants; Let be the measured wind and solar power output value for the t-th discrete time segment, in kW. This value corresponds one-to-one with the time segments of the initial power setting trajectory to ensure complete time synchronization and avoid distortion of the comparison results due to time sequence deviations. Strictly following the time sequence correspondence principle, the preprocessed real-time measured wind and solar power output data sequence is compared in real-time with the initial power setting trajectory obtained in step 440. That is, the difference between the measured power output value of each discrete time segment and the corresponding power setting value is calculated to obtain the real-time wind and solar power output deviation for each time segment. The calculation uses the following formula: ,in Let be the real-time wind and solar power output deviation for the t-th discrete time segment, in kW; when When the value is >0, it indicates that the measured power output of wind and solar power exceeds the set value, indicating excess power that needs to be absorbed through flexible resource adjustment; when When <0, it indicates that the measured output of wind and solar power is less than the set value, indicating insufficient output, and the power gap needs to be supplemented through flexible resource adjustment; when When the value is 0, it indicates that the wind and solar power output is completely consistent with the set value, with no deviation, and no correction or compensation is required. The real-time wind and solar power output deviation of each discrete time segment is arranged in ascending order of the scheduling cycle to form a complete real-time wind and solar power output deviation sequence. .

[0054] Step 442: Perform feedback correction compensation on the initial power setting trajectory based on the real-time wind and solar power output deviation. Calculate the dynamic power compensation amount for each scheduling period and superimpose it onto the corresponding trajectory nodes to obtain a dynamic correction command sequence. Specifically, this includes: determining the core logic of feedback correction compensation; for the real-time wind and solar power output deviation of each discrete time segment, combined with the regulation capability of flexible regulation resource nodes and the power flow response characteristics of the grid, calculating the corresponding dynamic power compensation amount; superimposing the compensation amount onto the corresponding nodes of the initial power setting trajectory; the corrected power setting value can offset the wind and solar power output deviation, ensuring node power balance and safe grid operation. To ensure the accuracy of the compensation amount and avoid grid fluctuations caused by excessive compensation, the dynamic power compensation amount is calculated using a proportional-integral regulation algorithm. The specific calculation formula is as follows: ,in The power dynamic compensation amount for the t-th discrete time segment is expressed in kW. This is the proportional adjustment coefficient, set according to the adjustment response speed of the flexible adjustment resources. The value range is 0.3-0.7. The larger the coefficient, the faster the compensation response speed, and the faster it can quickly offset sudden deviations in wind and solar power output. This is the integral adjustment coefficient, set according to the cumulative degree of wind and solar power output deviation, with a value range of 0.1-0.3. It is used to eliminate steady-state deviation and avoid grid power imbalance caused by long-term small deviations. The wind and solar power output deviation for the k-th discrete time segment is expressed in kW. The duration of each discrete time segment, intraday rolling scheduling =300s, during real-time scheduling =60s, ensuring the integral and proportional terms have consistent dimensions to avoid dimensional deviation; after calculating the dynamic power compensation for each discrete time segment, it is superimposed on the corresponding node of the initial power setting trajectory obtained in step 440, and the power setting value for each time segment is corrected using the following correction formula. ,in The corrected power control value for the t-th discrete time segment, in kW; synchronous verification is required during the correction process. Whether the upper and lower limits of power regulation of the flexible adjustment resource node determined in step 330 are met; if the corrected power control value exceeds the constraint range, then... The power control values ​​are trimmed and adjusted to fit within constraints to ensure they are legal and feasible, preventing equipment failures or grid fluctuations caused by exceeding constraints. After correcting all discrete time segments, the corrected power control values ​​for each time segment are arranged in ascending order of the scheduling cycle time to form a complete dynamic correction command sequence. .

[0055] Step 443: Perform spatial field recalculation on the constraint force field topology graph according to the dynamic correction command sequence, iteratively reconstruct the power flow offset vector of the entire network branches, obtain the updated power imbalance gradient distribution, and perform closed-loop security verification on the dynamic correction command sequence based on the updated power imbalance gradient distribution. After the verification passes, the final two-stage robust optimization scheduling strategy for wind and solar power grid connection is obtained. Specifically, this includes: performing spatial field recalculation on the constraint force field topology graph according to the dynamic correction command sequence. The core is to recalculate the tightness weight of each coupled potential energy edge in the constraint force field topology graph and the power imbalance of each node based on the corrected power control value, and then iteratively reconstruct the power flow offset vector of the entire network branches. The specific process is as follows: calculate the node power imbalance based on the dynamic correction command sequence. Calculate the actual power adjustment amount for each flexible adjustment resource node, and recalculate the power imbalance amount for each node based on the node power balance constraints. (i is the node number), which is the difference between the actual output of the node and the corrected power control value, reflecting the degree of power imbalance of each node, in kW; update the coupling potential energy edge tightness weight, combined with the node power imbalance. Recalculate the tightness weight of each coupled potential energy edge in the constraint force field topology graph. (i and j are the neighboring node numbers), the weight update formula is: ,in The updated coupling potential edge compactness weight is dimensionless and is used to characterize the tightness of the scheduling constraints between node i and node j. The compactness weight in the original constraint force field topology diagram is dimensionless. This is the weighting adjustment coefficient, with a value ranging from 0.2 to 0.5. It is used to control the degree of influence of power imbalance on the weight. The larger the coefficient, the more sensitive the weight is to power imbalance. The node's rated power, in kW, is used to normalize the node power imbalance, avoid dimensional bias, and ensure reasonable weight calculation. The entire network branch power flow offset vector is iteratively reconstructed based on the updated density weights. With node power imbalance Iteratively reconstruct the power flow offset vector of the entire network branches. This vector reflects the degree to which the power flow deviates from the rated value for each branch. The iterative process is as follows: First, initialize the power flow offset vector by calculating the initial offset value based on the nodal power imbalance. Then, adjust the offset of each branch in conjunction with the density weight to ensure that the offset matches the density of the nodal constraints. Repeat the iteration until the offset vector converges. The convergence condition is that the difference between the offsets of two adjacent iterations is less than 1. kW, the converged power flow offset vector is Based on the converged power flow offset vector of the entire network branches The power imbalance gradient distribution in the constraint force field topology graph is updated. The power imbalance gradient distribution characterizes the spatial distribution of power imbalance across the entire network; a larger gradient value indicates a more severe power imbalance in that region. During the update process, the power imbalance gradient is calculated for each node, with the node as the core. The formula is as follows: ,in is the power imbalance gradient of the i-th node, in kW / unit distance, used to quantify the degree of power imbalance around the node; Let i be the set of neighboring nodes of the i-th node; The power flow offset of the branch between node i and node j is expressed in kW. The updated power imbalance gradient distribution fully reflects the spatial distribution of power imbalance across the entire network, providing a core basis for the security verification of the dynamic correction command sequence.

[0056] Based on the updated power imbalance gradient distribution, a closed-loop security check is performed on the dynamic correction command sequence obtained in step 442. The core of the check is to ensure that the power adjustment actions corresponding to the dynamic correction command sequence will not cause the node power imbalance gradient to exceed the safety threshold or the branch power flow offset to exceed the safety limit. The specific check includes three items: First, check that the power imbalance gradient of each node does not exceed the preset safety threshold, which is set according to the power grid safety operation procedures to ensure that the node power imbalance is within a controllable range; second, check that the power flow offset of each branch does not exceed 10% of the branch's rated transmission capacity to avoid safety issues such as branch overload and thermal stability exceeding limits; third, check the power adjustment speed corresponding to the dynamic correction command sequence. If the rate and accelerator still meet the constraints determined in step 330, the equipment will be protected from adjustment shocks. If all three checks pass, it means that the dynamic correction command sequence is adapted to the real-time wind and solar power output fluctuations and grid safety constraints. At this time, the dynamic correction command sequence, together with the updated constraint force field topology and power imbalance gradient distribution, constitute the final two-stage robust optimization scheduling strategy for wind and solar power grid access. If the check fails, return to step 442, adjust the proportional-integral coefficient of the power dynamic compensation, recalculate the dynamic correction command sequence, and repeat the spatial field recalculation and closed-loop check in step 443 until the check passes, ensuring that the final scheduling strategy has practical execution value and can guarantee the safe and stable operation of the grid.

[0057] By deeply integrating robust dispatch commands with real-time wind and solar power output deviations, dynamic compensation and correction are used to offset the power imbalance caused by wind and solar power output fluctuations. Combined with constraint force field topology recalculation and closed-loop security verification, a two-stage robust optimization dispatch strategy is finally generated to improve the dispatch accuracy and anti-interference capability of wind and solar power grid connection.

[0058] like Figure 2 As shown, embodiments of the present invention also provide a wind power and solar power grid access dispatch system based on two-stage robust optimization, comprising: The distributed construction module is used to initialize the power grid dispatch state space based on a high-dimensional set of uncertain disturbances, and obtain the initial dispatch potential energy distribution tensor. The constraint generation module is used to analyze the interaction characteristics between nodes of the initial scheduling potential energy distribution tensor, and to map the data correlation between flexible adjustment resource nodes and source load fluctuation nodes to equivalent node coupling strength; based on the equivalent node coupling strength, the tightness distribution of scheduling constraints is calculated to obtain the constraint force field topology map. The dispatching instruction module is used to optimize the power regulation rate of flexible regulation resource nodes based on the constraint force field topology diagram. During the optimization process, the rate trajectory is discretized into a time series by using the upper and lower limits of the jerk of the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient, and the step size change rate and attenuation factor of the two-stage correction dispatching instruction are calculated to generate a robust dispatching instruction sequence. The feedback update module is used to send the robust scheduling command sequence to the grid edge control node, collect the real-time wind and solar power output deviation, use the real-time wind and solar power output deviation to perform feedback correction and compensation on the robust scheduling command sequence, update the power imbalance gradient distribution in the constraint force field topology diagram, and obtain the final two-stage robust optimization scheduling strategy for wind and solar power grid connection.

[0059] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0060] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A wind power and solar power grid access dispatching method based on two-stage robust optimization, characterized in that, The method includes: Step 1: Initialize the power grid dispatch state space based on the set of high-dimensional uncertain disturbances to obtain the initial dispatch potential energy distribution tensor; Step 2: Analyze the interaction characteristics between nodes on the initial scheduling potential energy distribution tensor, and map the data correlation between the flexible adjustment resource nodes and the source load fluctuation nodes to the equivalent node coupling strength; calculate the tightness distribution of scheduling constraints based on the equivalent node coupling strength to obtain the constraint force field topology map; Step 3: Based on the constraint force field topology diagram, the power regulation rate of the flexible regulation resource node is optimized. During the optimization process, the rate trajectory is discretized into a time series using the upper and lower limits of the jerk of the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient, and the step size change rate and attenuation factor of the two-stage correction dispatch command are calculated to generate a robust dispatch command sequence. Step 4: Send the robust scheduling command sequence to the grid edge control node, collect the real-time wind and solar power output deviation, use the real-time wind and solar power output deviation to perform feedback correction and compensation on the robust scheduling command sequence, update the power imbalance gradient distribution in the constraint force field topology diagram, and obtain the final two-stage robust optimization scheduling strategy for wind and solar power grid connection.

2. The wind power and solar power grid access dispatching method based on two-stage robust optimization according to claim 1, characterized in that, Before step 1, obtain the wind and solar power output time-series prediction sequence, multi-source heterogeneous meteorological disturbance parameters and network topology constraint data of the target power grid within the preset scheduling cycle, and construct a high-dimensional uncertainty disturbance set.

3. The wind power and photovoltaic grid connection dispatching method based on two-stage robust optimization according to claim 2, characterized in that, The power grid dispatch state space is initialized based on a set of high-dimensional uncertain disturbances, resulting in an initial dispatch potential energy distribution tensor, including: The temporal and spatial prediction sequences of wind and solar power output in the high-dimensional uncertainty disturbance set are aligned with the parameters of multi-source heterogeneous meteorological disturbances and normalized in terms of dimensions. The source load fluctuation characteristics and meteorological disturbance coupling vectors of each node in the preset scheduling period are extracted to construct a standardized temporal and spatial disturbance feature sequence. Based on the standardized spatiotemporal disturbance feature sequence and combined with network topology constraint data, the initial power transmission sensitivity matrix between each node is calculated, and the initial power transmission sensitivity matrix is ​​mapped to the grid steady-state operation boundary space to obtain the grid dispatch initial state space feature matrix. Based on the initial state space feature matrix of the power grid dispatch, the power mismatch risk degree and voltage over-limit probability of each spatiotemporal node under the action of uncertain disturbance are calculated, and the power mismatch risk degree and voltage over-limit probability are fused by multimodal weighting to form a primary potential energy distribution matrix that characterizes the dispatch pressure situation of the entire network. Multidimensional tensor reconstruction and orthogonal basis projection transformation are performed on the primary potential energy distribution matrix. Tensor product operation and feature dimensionality reduction are performed along the node topology dimension, temporal evolution dimension and perturbation confidence dimension to finally obtain the initial scheduling potential energy distribution tensor.

4. The wind power and solar power grid access dispatching method based on two-stage robust optimization according to claim 3, characterized in that, The initial scheduling potential energy distribution tensor is analyzed for inter-node interaction characteristics. The data correlation between flexible adjustment resource nodes and source load fluctuation nodes is mapped to the equivalent node coupling strength, including: Based on the initial scheduling potential energy distribution tensor, extract the power fluctuation feature sequence in the time dimension and the adjacency association feature vector in the topological dimension for each flexible adjustment resource node and each source load fluctuation node; concatenate the power fluctuation feature sequence and the adjacency association feature vector, and calculate the normalized dynamic time curvature distance of the feature sequence between the concatenated nodes to obtain the initial association difference between the nodes. Based on the initial correlation difference, combined with the adjustable capacity and response dead zone threshold of each flexible adjustment resource node, and the fluctuation amplitude level of each source load fluctuation node, the data correlation coefficient of each node under different scheduling periods is calculated; the data correlation coefficients of all periods are then weighted by time series to obtain the comprehensive data correlation between nodes. The comprehensive data correlation degree is converted into a primary coupling strength value within a preset interval according to a preset nonlinear mapping function. The primary coupling strength value is then subjected to neighborhood iterative averaging and boundary truncation to finally obtain the equivalent node coupling strength used to characterize the degree of interaction between the flexible adjustment resource node and the source load fluctuation node.

5. The wind power and photovoltaic grid connection dispatching method based on two-stage robust optimization according to claim 4, characterized in that, The tightness distribution of scheduling constraints is calculated based on the equivalent node coupling strength, resulting in a constraint force field topology diagram, including: Based on the equivalent node coupling strength, a coupling potential energy edge is constructed between each pair of flexible adjustment resource nodes and source load fluctuation nodes that have a coupling relationship, and an initial density weight is assigned to each coupling potential energy edge to obtain a set of coupling potential energy edges with initial density weights. Based on the coupled potential energy edge set, and combined with the upper and lower limits of the power regulation and the ramp rate limit of each node as the node constraint boundary value, the power flow constraints of all branches and the power balance constraints of nodes in the power grid are traversed, and the Lagrange multiplier estimate of each constraint under the current scheduling period is calculated; constraints with Lagrange multiplier estimates greater than a preset threshold are marked as active constraints, and the active constraints are sorted in descending order according to the size of the Lagrange multiplier estimates to obtain the constraint tightness ranking list; Based on the density sorting list, the density values ​​of active constraints are assigned to the relevant nodes or coupled potential energy edges, while the density values ​​of inactive constraints are set to 0. All density values ​​are spatially smoothed, and the smoothed density distribution is superimposed on the basic topology graph composed of nodes and coupled potential energy edges to finally obtain the constraint force field topology graph.

6. The wind power and photovoltaic grid connection dispatching method based on two-stage robust optimization according to claim 5, characterized in that, The initial density weight is proportional to the equivalent node coupling strength and inversely proportional to the electrical distance between nodes.

7. The wind power and photovoltaic grid connection dispatching method based on two-stage robust optimization according to claim 5, characterized in that, Based on the constraint force field topology diagram, trajectory optimization of the power regulation rate of flexible regulation resource nodes is performed. During the optimization process, the rate trajectory is discretized into a time series using the upper and lower limits of the jerk constraint on the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. This curve includes: Based on the density weight of each coupled potential energy edge in the constraint force field topology diagram and the power regulation capability boundary of each flexible regulation resource node, the jerk upper and lower limit constraint values ​​of the power regulation rate of each flexible regulation resource node in the scheduling cycle are determined, namely the second time derivative of the power regulation rate, the upper and lower limit constraint values ​​of the rate, and the node power balance boundary conditions and branch power flow boundary conditions defined by the constraint force field topology diagram, forming the initial feasible region of the power regulation rate of each node. Based on the initial feasible region of the power regulation rate, the scheduling period is discretized into multiple continuous time segments at equal intervals, and the power regulation rate values ​​in each time segment are arranged in chronological order to form a discrete time sequence of the power regulation rate to be solved. The upper and lower limits of jerk are transformed into second-order difference inequalities between rate values ​​of three adjacent time segments in the discrete time series of power regulation rate. The upper and lower limits of rate are transformed into boundary inequalities of rate series values ​​in each time segment. The boundary conditions of node power balance and branch power flow are transformed into a set of inequalities that the discrete time series of power regulation rate needs to satisfy. Based on the discrete time series of power regulation rate and the various inequalities and equality constraints after transformation, a quadratic programming problem is constructed with the joint optimization objectives of minimizing the sum of the jerk changes of the rate series over all time segments and minimizing the sum of regulation energy consumption. The second-order difference inequality, rate boundary inequality and the set of inequalities are all used as constraints of the quadratic programming problem. The quadratic programming problem is solved to obtain the final power regulation rate value of each flexible regulation resource node in each discrete time segment. Based on the final power regulation rate value, the sequence is spliced ​​in ascending order of time to obtain the complete discrete rate trajectory; the discrete rate trajectory is then subjected to cubic spline smoothing filtering to finally obtain the power regulation rate curve.

8. The wind power and photovoltaic grid connection dispatching method based on two-stage robust optimization according to claim 7, characterized in that, The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient. The step size change rate and attenuation factor of the two-stage correction dispatch command are calculated, and a robust dispatch command sequence is generated, including: Based on the final power regulation rate value of each flexible regulation resource node in each discrete time segment in the power regulation rate curve, the corresponding power regulation increment sequence is calculated by integration; the power regulation increment sequence is convolved with the preset power flow response inertia coefficient to obtain the power flow response delay distribution of each flexible regulation resource node after the power regulation action; Based on the power flow response delay distribution, the response lag time constant and overshoot characteristic parameters of each flexible regulation resource node are extracted; the response lag time constant and overshoot characteristic parameters are input into the preset step size change rate calculation function to obtain the step size change rate between the first stage prediction step size and the second stage correction step size of each flexible regulation resource node in the two-stage correction scheduling. Based on the overshoot characteristic parameters and the peak time characteristics in the power flow response delay distribution, the attenuation factor of each flexible regulation resource node in the two-stage correction scheduling is determined. The step size change rate and the attenuation factor are multiplied with the corresponding basic scheduling command amplitude to obtain the predicted correction command sequence and the real-time correction command sequence of each flexible adjustment resource node at each time in the scheduling cycle. The predicted correction instruction sequence and the real-time correction instruction sequence are concatenated according to the scheduling time sequence, and the instruction values ​​at the concatenation boundary are smoothed to obtain the robust scheduling instruction sequence.

9. The wind power and photovoltaic grid connection dispatching method based on two-stage robust optimization according to claim 8, characterized in that, The robust scheduling command sequence is sent to the grid edge control node, real-time wind and solar power output deviations are collected, and the robust scheduling command sequence is corrected and compensated using these deviations. The power imbalance gradient distribution in the constraint force field topology is updated, resulting in the final two-stage robust optimization scheduling strategy for wind and solar power grid integration, including: The robust scheduling instruction sequence is sent to the grid edge control node to perform node-level instruction parsing and local action mapping to obtain the initial power setting trajectory; The real-time differential comparison between the initial power setting trajectory and the measured power output data of the wind and solar power stations collected on site is performed to obtain the real-time wind and solar power output deviation. Feedback correction and compensation are performed on the initial power setting trajectory based on the real-time wind and solar power output deviation. The dynamic power compensation amount for each scheduling period is calculated and superimposed on the corresponding trajectory node to obtain the dynamic correction command sequence. Based on the dynamic correction command sequence, the spatial field recalculation of the constraint force field topology is performed, and the power flow offset vector of the entire network branches is iteratively reconstructed to obtain the updated power imbalance gradient distribution. Then, the closed-loop security verification of the dynamic correction command sequence is performed based on the updated power imbalance gradient distribution. After the verification is passed, the final two-stage robust optimization scheduling strategy for wind power and photovoltaic grid connection is obtained.

10. A wind power and solar power grid connection dispatch system based on two-stage robust optimization, wherein the system implements the method as described in any one of claims 1 to 9, characterized in that, include: The distributed construction module is used to initialize the power grid dispatch state space based on a high-dimensional set of uncertain disturbances, and obtain the initial dispatch potential energy distribution tensor. The constraint generation module is used to analyze the inter-node interaction features of the initial scheduling potential energy distribution tensor and map the data correlation between flexible adjustment resource nodes and source load fluctuation nodes to the equivalent node coupling strength. The tightness distribution of scheduling constraints is calculated based on the equivalent node coupling strength, and the constraint force field topology is obtained. The dispatching instruction module is used to optimize the power regulation rate of flexible regulation resource nodes based on the constraint force field topology diagram. During the optimization process, the rate trajectory is discretized into a time series by using the upper and lower limits of the jerk of the power regulation rate. The power regulation rate curve is obtained by solving a quadratic programming problem that satisfies the jerk limit, rate limit, and boundary conditions defined by the constraint force field topology diagram. The power regulation rate curve is dynamically matched with the preset power flow response inertia coefficient, and the step size change rate and attenuation factor of the two-stage correction dispatching instruction are calculated to generate a robust dispatching instruction sequence. The feedback update module is used to send the robust scheduling command sequence to the grid edge control node, collect the real-time wind and solar power output deviation, use the real-time wind and solar power output deviation to perform feedback correction and compensation on the robust scheduling command sequence, update the power imbalance gradient distribution in the constraint force field topology diagram, and obtain the final two-stage robust optimization scheduling strategy for wind and solar power grid connection.

Images