A wind-solar-storage integrated energy management system and dispatching method and system

Abstract

This invention discloses an integrated wind-solar-storage energy management system and scheduling method, relating to the field of big data processing technology. The integrated wind-solar-storage energy management system and scheduling method includes the following steps: S1, collecting multiple types of operational datasets and performing unified time base alignment and normalization processing; S2, constructing a distribution network topology model and generating a set of network feasible domain constraints; S3, constructing a collaborative rolling scheduling mechanism, performing power flow verification and over-limit correction; S4, performing consistency verification of receipts and status readbacks to generate closed-loop verification data frames. This invention effectively improves the collaborative absorption capacity and scheduling feasibility of integrated wind-solar-storage power plants under grid connection constraints, solving the problem of grid connection constraints and lack of unified modeling of internal grid bottlenecks, leading to grid connection compliance but overloaded main transformers and feeders or voltage over-limits, forcing power curtailment.

Description

Technical Field

[0001] This invention relates to the field of big data processing technology, specifically to an integrated wind, solar, and energy storage energy management system and scheduling method and system. Background Technology

[0002] With the continuous increase in the proportion of multi-source access such as wind power, photovoltaics, and energy storage, and the tightening of distribution network operation boundaries, existing technologies often struggle to normalize and reliably encapsulate multi-source measurements and capacity boundaries under a unified time base and scale, thus making it difficult to support real-time closed-loop scheduling, power flow verification, and limit correction.

[0003] For example, invention patent CN118587038B discloses a method and device for real-time verification and reporting of power generation data from an integrated wind-solar-storage power station. The method includes: acquiring the maximum available power and actual power output fed back by the energy management platform; determining the reporting dispatch step size based on the data frequency of the actual power output fed back by the energy management platform; collecting the maximum available power output fed back by the energy management platform according to the reporting dispatch step size; comparing the collected maximum available power output with the actual power output at the same time; updating the current maximum available power output to the actual power output at the same time when the collected maximum available power output is less than the actual power output at the same time; and uploading the updated maximum available power output and actual power output to the power grid dispatch center according to the reporting dispatch step size. This invention overcomes the problem that the maximum available power output of the data transmitted to the power grid is less than the actual power output due to differences in communication links and update time scales, as well as unit inertia, during normal unit operation.

[0004] In existing technologies, under the existing power distribution structure of the collection station or park, the output of wind and solar power is concentrated in certain branches, and the hot spots of the main transformer, the current carrying capacity of the feeder and the node voltage form internal bottlenecks. Existing EMS often treats the active power limit of the grid connection point as the only constraint, and relies solely on post-event alarms or manual adjustment of reactive power, which leads to local voltage exceeding the limit triggering inverter derating, protection malfunctions, and power curtailment and fluctuations occurring simultaneously.

[0005] Therefore, in order to address the above problems, there is an urgent need for an integrated wind, solar, and energy storage energy management system and scheduling method. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an integrated wind, solar, and energy storage energy management system and scheduling method, which solves the problem that grid connection points are limited and internal power grid bottlenecks are not uniformly modeled, resulting in grid connection points meeting the standards but main transformers and feeders being overloaded or voltage exceeding limits, forcing power curtailment and limiting.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a wind-solar-storage integrated energy management and scheduling method, comprising: S1, collecting multiple types of operational datasets, performing unified time base alignment and normalization on the multiple types of operational datasets to form a joint observation data frame; S2, constructing a distribution network topology model, generating a network feasible domain constraint set, and performing sensitivity evaluation to generate a partitioned control domain; S3, constructing a collaborative rolling scheduling based on the feasible domain enhancement model and the partitioned control domain, performing power flow verification and limit overrun correction, and generating safety control quantities; S4, performing consistency verification of receipts and status readbacks to generate a closed-loop verification data frame, and performing anomaly attribution and online backoff replanning.

[0008] Furthermore, the specific process of collecting multiple types of operational datasets is as follows: multiple types of operational datasets are collected through collaborative access via the station edge acquisition gateway. These multiple types of operational datasets include: grid-side datasets, source-storage-side datasets, monitoring datasets, and time-series identifier datasets; operational data that has already been stored under the same substation and park power distribution structure is used as historical operational data; and the collected multiple types of operational datasets are classified, labeled, and prioritized.

[0009] Furthermore, the specific process of performing unified time base alignment and normalization on multiple types of operational datasets to construct a joint observation data frame is as follows: A unified time service is used to perform periodic time base alignment on the station-end edge gateway and the telemetry and control device to obtain the processed multiple types of operational datasets; the processed multiple types of operational datasets are normalized, and missing measurement detection, duplicate packet detection, spike detection, and rate of change boundary detection are performed, using the grid connection point power closure error and node voltage consistency error as consistency quality indicators; the consistency quality indicator of each record is written to the quality marker field in the joint observation data frame to record missing measurement, spike, backfill segment, time delay anomaly, and closure error exceeding limits; the processed multiple types of operational datasets within each time slice are encapsulated in a unified data structure and their fields are bound to construct the joint observation data frame.

[0010] Furthermore, the specific process of constructing the distribution network topology model and generating the network feasible region constraint set is as follows: Input the grid-side dataset and, combined with the node, branch, and partition identifiers in the joint observation data frame, construct the distribution network topology model through a graph reconstruction and consistency verification algorithm based on topology prior constraints; use the primary wiring topology version number as the topology context identifier, generate node and branch sets according to node and branch identifiers, perform computability processing on the distribution network topology model, and generate network state variables and constraint expressions; perform piecewise convex approximation on the power flow relationship of the distribution network to form the network feasible region constraint set. Domain constraint set; Under the radial power distribution structure, an approximate linear power flow model is adopted to transform the nonlinear relationship between branch power flow and node voltage into piecewise linear constraints, outputting a feasible domain constraint set for the network; Based on the source-storage side dataset and the device access node identifiers in the joint observation data frame, the capacity boundaries of wind turbines, inverters, energy storage converters, and reactive power devices are embedded into the feasible domain constraint set for the network; Through the convex hull embedding algorithm for the device PQ capacity curve and the energy storage P-SOC capacity boundary, combined with the joint constraint splicing rules of the device capacity boundary, a feasible domain enhancement model with device constraints is constructed.

[0011] Further, the specific process of generating the partitioned control domain through sensitivity assessment is as follows: Based on the feasible domain enhancement model, the node voltage constraint and branch current carrying capacity constraint are bound and determined in each time slice, and a bottleneck list is generated; sensitivity assessment is performed on the bottleneck location to generate the mapping relationship between the bottleneck and controllable resources and determine the partitioned control domain; the excess amount of the bottleneck constraint that needs to be repaired under the current observation state is calculated, and only the positive part is retained as the positive part of the amount to be repaired; then, a set of executable control action vectors is obtained by traversing the executable control set corresponding to the bottleneck as the constraint space increment; for each set of candidate control actions, the L2 norm of the positive part of the amount to be repaired and the L2 norm of the constraint space increment are calculated respectively, and the L2 norm of the positive part of the amount to be repaired is divided by the L2 norm of the constraint space increment and a minimum term is added to obtain the sensitivity assessment value; the sensitivity assessment value is compared with the sensitivity threshold in real time to trigger the determination of the partitioned control domain and the selection of the disposal path; when the bottleneck sensitivity assessment value is less than or equal to the sensitivity threshold, the bottleneck is determined to be a remediable bottleneck, and when the bottleneck sensitivity assessment value is greater than the sensitivity threshold, the bottleneck is determined to be an unremediable bottleneck, and the partitioned control domain and bottleneck disposal suggestions are output.

[0012] Furthermore, the specific process of constructing collaborative rolling scheduling based on the feasible domain enhancement model and the partitioned control domain is as follows: Based on the feasible domain enhancement model, a rolling scheduling window is constructed and the partitioned combined control quantity sequence is obtained by solving; the wind power availability prediction, photovoltaic availability prediction and load disturbance estimation of each future time slice are used as available output priors, and the confidence level label is transformed into a prediction corridor constraint; a hierarchical optimization objective solution strategy is adopted to generate the combined control quantity sequence and output the partitioned combined control quantity sequence.

[0013] Further, the specific process of performing power flow verification and limit correction to generate safety control quantities is as follows: Input the partitioned combined control quantity sequence, perform power flow verification and limit correction on the control quantities of each time slice, and generate a scheduling plan data frame; Multiply the constraint sensitivity matrix with the transpose matrix to obtain a matrix product; Multiply the damping stability coefficient with the identity matrix to obtain the damping matrix; Add the matrix product and the damping matrix to obtain the matrix sum; Perform an invertibility check on the matrix sum, and determine non-singularity by calculating the condition number and eigenvalues ​​of the matrix. When the invertibility condition is met, invert the matrix sum to obtain the inverse matrix; Multiply the inverse matrix with the vector to be repaired to obtain the intermediate vector; Use the online linearized power flow model and hotspot approximation model to apply the current... The constraint sensitivity matrix is ​​calculated under the observation state. The transpose of the constraint sensitivity matrix is ​​multiplied by the intermediate vector to obtain the correction direction vector. The correction direction vector is substituted into the projection function to obtain the projected vector. The negative value of the projected vector is taken to obtain the over-limit correction control increment. The over-limit correction control increment is written as the safety control quantity to obtain the corrected safety control quantity. The power flow is re-verified on the corrected safety control quantity to obtain the remaining over-limit quantity. The remaining over-limit quantity is compared with the safety corridor threshold in real time. When the remaining over-limit quantity is less than or equal to the safety corridor threshold, it is determined that the over-limit correction has met the standard and the safety control quantity is output. When the remaining over-limit quantity is greater than the safety corridor threshold, it is determined that the correction has not met the standard and the correction result is written back as the safety control quantity sequence.

[0014] Furthermore, the specific process for generating a closed-loop verification data frame through consistency verification of receipts and status readbacks is as follows: Input the scheduling plan data frame, perform pre-readback confirmation according to the time slice index to read the current operating mode and status, and issue the setting instruction after the pre-conditions are met. Perform receipt collection and status readback collection for each instruction transaction to construct an execution proof package set; perform consistency verification on the execution proof package set to generate an execution deviation vector and output a closed-loop verifiable value; divide the setting consistency deviation by the setting consistency tolerance, the effective consistency deviation by the effective consistency tolerance, and the timeliness consistency deviation by the timeliness. Five deviation ratios are obtained by dividing the tolerance, evidence completeness deviation by the evidence completeness tolerance, and control strategy consistency deviation by the control strategy tolerance. The maximum value among these five deviation ratios is taken as the worst deviation ratio. The worst deviation ratio is multiplied by the attenuation coefficient, negativeened, and then exponentially calculated to obtain the closed-loop verifiable value. The closed-loop verifiable value is compared with the verification threshold in real time. When the closed-loop verifiable value is higher than or equal to the verification threshold, the execution of the time-slice instruction is deemed trustworthy and the closed-loop confirmation is completed. When the closed-loop verifiable value is lower than the verification threshold, the execution of the time-slice instruction is deemed untrustworthy and a rollback and retry strategy is triggered, and a closed-loop verification data frame is output.

[0015] Furthermore, the specific process of anomaly attribution and online rollback replanning is as follows: for verification failure time slices, anomaly attribution is performed according to the dimensions of receipt timeliness, readback consistency, equipment availability, and capacity boundary contraction, triggering rollback and replanning; the online update of the feasible domain enhancement model is driven by the closed-loop verification data frame; the effective boundary calibration of the equipment capacity curve is performed based on the readback and actual output; when the same type of current limiting occurs, the linearized power flow sensitivity matrix is ​​adjusted by residual drive based on the measured node voltage, branch current, and hotspot response after execution, and the online rollback replanning result is output.

[0016] Furthermore, a second aspect of the present invention provides an integrated wind-solar-storage energy management and scheduling system, applied to an integrated wind-solar-storage energy management and scheduling method, comprising: a multi-source data acquisition module, used to coordinate access to various grid-side, source-storage-side, and contextual time-series data at the substation, perform unified time base alignment, caliber normalization, and quality detection, and encapsulate them into joint observation data frames and historical operation data according to time slices; a feasible region modeling module, used to construct a distribution network topology model, perform piecewise convex approximation of power flow relationships to form a set of feasible region constraints, and embed consistency constraints into the feasible region to form a feasible region enhancement model; a collaborative rolling scheduling module, used to generate a partitioned combined control quantity sequence using a hierarchical objective solution strategy, and perform power flow verification and limit violation correction on candidate control quantities, outputting safe control quantities and scheduling plan data frames; and a scheduling execution closed-loop module, used to collect multi-level receipts and status readbacks to construct an execution proof package set, trigger backoff or retry according to thresholds, and simultaneously output closed-loop verification data frames.

[0017] The present invention has the following beneficial effects: (1) In this invention, a topology model of the substation distribution network is constructed by graph reconstruction and consistency verification algorithm based on topology prior constraints, and the primary topology version number is used as a context key for versioned constraint expression, so that network state variables, node voltage constraints and branch current carrying constraints can be computed in the same topology context, thereby reducing the verification distortion and omission risk caused by topology changes.

[0018] (2) In this invention, the capacity boundaries, caliber mutual exclusion rules and state-triggered boundary contraction rules of wind turbines, inverters, energy storage converters and reactive power devices are explicitly embedded into the feasible region through convex hull embedding and joint constraint splicing algorithms, forming a feasible region enhancement model with equipment constraints, ensuring that the scheduling solution results are executable under the constraints of equipment capacity, mode caliber and availability.

[0019] (3) This invention, through a collaborative rolling scheduling strategy of hierarchical targets or lexicographical targets, prioritizes the improvement of new energy consumption and suppresses the risk of exceeding limits under the premise of satisfying network constraints and equipment constraints. At the same time, it constrains the jump of set values ​​between time slices, mode switching and energy storage direction reversal, reduces control jitter and operation oscillation, and improves the stability and sustainable execution of scheduling output.

[0020] (4) In this invention, through power flow verification and over-limit correction mechanism, when the voltage, current or hot spot constraints are found to be over-limit, the control quantity is projected and corrected by the idea of ​​minimum modification, and the ramp, mutual exclusion and availability are not destroyed, so that the corrected control quantity is as close as possible to the original combined control quantity and the time series is kept smooth, avoiding the introduction of new over-limit and shock in order to repair a certain constraint.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] Figure 1 This is a flowchart of an integrated wind, solar, and energy storage energy management and dispatching method according to the present invention; Figure 2 This is a diagram illustrating the architecture of an integrated wind, solar, and energy storage management system according to the present invention. Figure 3 This is a time-series change diagram of the operating data of this invention; Figure 4 This is a visualization of the bottleneck sensitivity evaluation data of the present invention; Figure 5 The flowchart for generating closed-loop verification data frames for consistency verification of this invention is shown below. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figures 1-5This invention provides a technical solution: a wind-solar-storage integrated energy management and scheduling method, comprising: S1, collecting multiple types of operational datasets, performing unified time base alignment and normalization on the multiple types of operational datasets to form a joint observation data frame; S2, constructing a distribution network topology model, generating a network feasible domain constraint set, and performing sensitivity evaluation to generate a partitioned control domain; S3, constructing a collaborative rolling scheduling based on the feasible domain enhancement model and the partitioned control domain, performing power flow verification and exceeding limit correction, and generating safety control quantities; S4, performing consistency verification of receipts and status readbacks to generate a closed-loop verification data frame, and performing anomaly attribution and online backoff replanning.

[0025] Specifically, the process of collecting multiple types of operational datasets is as follows: Multiple types of operational datasets are collaboratively accessed and collected through a station-side edge acquisition gateway. These datasets include: grid-side datasets, source-storage-side datasets, monitoring datasets, and time-series identifier datasets. The grid-side datasets include: a subset of grid-connected point status, a subset of station-internal grid status, and a subset of grid parameter constraints. The grid-connected point status subset includes: active power, reactive power, three-phase voltage, three-phase current, frequency, power factor, circuit breaker and disconnector status, and demand statistics window values ​​at the grid-connected point, measured by the grid-connected point energy meter or synchronous phasor measurement unit at a second-level sampling period. The data, obtained from previous submissions, is used to characterize the upper limit of grid connection point, ramp-up, and PQ compliance constraints. The substation grid status subset includes the 35kV bus voltage, 10kV bus voltage, voltage at the beginning and end nodes of key feeders, active power, reactive power, current, load factor, direction markers, voltage over-limit markers, and voltage change rate. This data is collected through substation protection and control devices, voltage transformer measurement channels, feeder terminal units, feeder control devices, or sectionalizing switch control units at a sampling period of 1s-10s. It is used to characterize node voltage constraints and current carrying capacity constraints in internal bottlenecks. The grid parameter constraint subset... This includes: primary wiring topology version number, node identifier, branch identifier, zone identifier, branch impedance parameters, feeder rated current and thermal stability threshold, main transformer rated capacity and hotspot upper limit, voltage upper and lower limit thresholds, grid connection point power upper limit, and ramping constraints. These are imported from the station-side configuration library during the commissioning phase and updated when change orders take effect, and are used for network feasible domain modeling and fast power flow verification. The source-storage side dataset includes: wind power side operation and capacity subsets, photovoltaic side operation and capacity subsets, energy storage side operation and capacity subsets, and reactive power resource pool subsets. The wind power side operation and capacity subset includes active and reactive power data from units or unit groups. Output, available power, power limiting status, ramp-up capability, grid connection status, and fault alarm summary are collected by wind turbine SCADA or power aggregation unit at a sampling period of 2s-10s, and are used to generate adjustable output and available reactive power on the wind power side; the photovoltaic side operation and capability subset, including inverter active and reactive power output, available active power, temperature rationing flag, current limiting flag, power factor setting, reactive power setting, Volt-Var curve level and start / stop flag, and current effective parameter summary, are collected by inverter communication interface at a sampling period of 1s-5s, and are used to generate photovoltaic side PQ collaborative control input;The energy storage side operation and capacity subset includes PCS active and reactive power output, charging and discharging direction, operation mode marking, current limiting reason code, grid connection support start / stop status and response delay statistics, as well as BMS output SOC, SOH, maximum cell temperature, alarm status, upper limit of chargeable power, upper limit of dischargeable power, charging and discharging current limit, and estimated available capacity. This is obtained through collaborative sampling via the PCS control interface at a 1-2s sampling period and the BMS interface at a 2-10s sampling period. This data forms the executable control inputs for energy storage buffering and reactive power support and constrains the energy storage capacity boundary. The reactive power resource pool subset includes SVG and SV. The reactive power output, reactive power availability margin, switching status of electrical and reactance components, control mode, and fault alarm summary are collected through the reactive power compensation device control interface at a sampling period of 1s-5s. These data are used to construct the reactive power capacity pool within the station and serve voltage constraints. In addition, the equipment capacity curves and control caliber subsets, including the PQ capacity boundary point set or segmented parameters of the inverter and PCS, power factor allowable range, current limiting boundary, reactive power priority, active power priority caliber markings and symbol conventions, are imported through the equipment capacity declaration file during the commissioning and testing phase and updated when firmware and parameters change. These data are used to explicitly inject the equipment capacity boundaries for optimization and verification. The monitoring dataset includes: a subset of station meteorological observations and short-term forecasts, and a subset of operational modes and scheduling contexts. The station meteorological observations and short-term forecasts subset includes wind speed, wind direction, irradiance, ambient temperature, and short-term output forecast sequences and confidence labels at a 5-15 minute granularity. These are obtained through the station meteorological station's interface with the forecast service and are used to provide available output priors for rolling optimization and to form uncertainty boundaries. The operational mode and scheduling context subset includes runtime segment labels, limited power generation, maintenance and grid connection mode labels, key equipment commissioning / decommissioning status, and safety interlocking labels. These are read through the station control system's interface with the equipment status and are used to constrain the executability of scheduling actions. The time-series identifier dataset includes: equipment local sampling timestamps, edge gateway receiving timestamps, station-end aggregation timestamps, backfill write labels, equipment and channel numbers, data type labels, sampling period labels, and data quality labels. These are appended by the edge acquisition gateway when receiving various measurements and statuses. These are used for unified time base sorting, backfill rearrangement, and quality downweighting, and to provide time and data reliability labels for fast power flow verification.

[0026] Operational data already stored within the same substation and industrial park power distribution structure is designated as historical operational data. This historical operational data includes: historical grid connection point P and Q trajectories, historical node voltage trajectories, feeder load rate trajectories, main transformer hotspot temperature rise trajectories, inverter temperature rationing records, PCS current limiting records, SOC and SOH evolution trajectories, and summaries of historical power curtailment events. These are used to statistically analyze bottleneck frequency, constraint-bound patterns, and data quality baselines. The collected multi-type operational datasets are categorized, labeled, and prioritized. Grid connection point metering, key bus and node voltages, main transformer hotspot temperature rise, and tightly constrained feeder current carrying capacity are labeled as key constraint measurement point frames, and these frames are given high priority in the cache queue. The output and capacity boundaries of wind turbines, inverters, PCS, and BMS are labeled as equipment capacity frames, and topology and threshold parameters are labeled as constraint parameter frames. When acquisition link bandwidth is limited or there is a risk of cache overflow, key constraint measurement point frames and equipment capacity frames are retained first, while non-key measurement point frames are delayed or downsampled.

[0027] This implementation plan significantly improves the real-time visibility and location speed of voltage over-limit, current tight binding, and hotspot approaching bottlenecks, enhances the input reliability and stability of subsequent feasible domain modeling, rolling scheduling, and over-limit correction, reduces the risk of misjudgment and unexecutable control caused by missing measurements, spikes, delays, and inconsistent standards, and provides reusable data support for historical pattern statistics, extraction of constraint tight binding patterns, and solidification of data quality baselines.

[0028] Specifically, the process of performing unified time base alignment and normalization on multiple types of operational datasets to form a joint observation data frame is as follows: A unified time service is used to perform periodic time base alignment on the station-end edge gateway and the measurement and control device to obtain the processed multi-type operational datasets; PTP or NTP time synchronization is used to ensure that the edge gateway and the station clock are consistent; for devices that cannot be precisely synchronized, the fixed deviation and drift term are estimated using the device's local timestamp and the gateway's received timestamp, and the device sampling time is remapped to a continuous unified time axis; data from different sampling periods are aligned according to a unified time slice granularity to generate a time slice index, and a unique sorting time is written for each record to ensure that grid connection points P and Q, node voltage, feeder current carrying capacity, main transformer hotspots, and device PQ outputs share a unified time index at the same physical moment.

[0029] The processed multi-type operational datasets are normalized, unifying the units and directional symbols for active power, reactive power, voltage, and current. Voltage quantities are standardized to rated voltage, and power quantities are scaled dimensionlessly to rated capacity, ensuring data from different voltage levels and equipment capacities are comparable on the same scale. Active and reactive power at grid connection points, feeders, branches, wind turbines, inverters, and energy storage converters are scaled dimensionlessly to their corresponding rated apparent capacity or station rated capacity to obtain normalized active and reactive power values. The rates of change of active and reactive power are normalized to the allowable ramp-up limit to obtain ramp-up normalized values. Feeder and branch currents are normalized to thermal stability thresholds or rated currents to obtain normalized current values, and current carrying margins are calculated for use in... It directly characterizes the tightness of the current carrying capacity constraint; it normalizes the main transformer hot spot temperature, winding temperature, or oil temperature according to the corresponding upper limit to obtain the temperature normalized value, and calculates the hot spot margin to characterize the main transformer heat capacity constraint; it maps the energy storage state of charge to the zero-to-one interval to obtain the state of charge normalized value, and normalizes the upper limit of chargeable power and the upper limit of dischargeable power according to the rated power of the energy storage converter to obtain the normalized values ​​of the upper limit of chargeable power and the upper limit of dischargeable power, and at the same time maps the health state to the available capacity coefficient to correct the available boundary; it expresses the reactive power capacity of the inverter, energy storage converter, and reactive power compensation device in a dimensionless manner according to the rated apparent capacity, and expresses the active and reactive power capacity curves in a dimensionless piecewise point set manner to ensure that the capacity curves of different manufacturers can be compared and constrained in a unified space.

[0030] It performs missing measurement detection, duplicate packet detection, spike detection, and rate of change boundary detection, using the grid connection point power closure error and node voltage consistency error as consistency quality indicators. The grid connection point power closure error is used to quantify the deviation between the active and reactive power at the grid connection point and the feeder aggregation, energy storage, and reactive power resource pool within the station. The critical node voltage consistency error is used to quantify the deviation level of the same node under different measurement channels or adjacent time slices. Based on the consistency quality indicators, quality scores and quality labels are generated. Closure error exceeding limits, inter-channel voltage deviation exceeding limits, rate of change exceeding limits, and spike segments are downweighted or eliminated. The downweighting coefficient is written into the joint observation data frame to constrain the input credibility of power flow verification and the penalty strength of the rolling optimization objective function, avoiding low-credibility measurement-driven error over-limit correction and control oscillation; it is used for power flow verification and optimization downweighting.

[0031] The consistency quality indicators for each record are written to the quality tag field in the joint observation data frame to record missing measurements, spikes, backfill segments, time delay anomalies, and closure error exceeding limits. The processed multi-class operational datasets within each time slice are encapsulated in a unified data structure and their fields are bound to construct the joint observation data frame. The joint observation data frame includes: topology version number, time slice index, device, node, branch and partition identifier, active power, reactive power, voltage, current and frequency at the grid connection point, voltage of key nodes, active power, reactive power, current and load factor of feeders, load factor and hotspot margin of main transformers, active power and reactive power output and available boundary parameters of wind turbines, inverters and energy storage converters, state of charge, health status and upper limit of charge / discharge of the battery management system, output and available margin of reactive power devices, device mode and current limiting, derating reason code, device timestamp and gateway receiving timestamp, unified sorting time, and quality score and quality tag. The grid connection point power closure error is used to characterize power balance closure: taking the grid connection point as the boundary, the active and reactive power measured at the grid connection point are compared with the sum of the power of each feeder and branch in the station, the sum of the source-side output, the charging and discharging power of energy storage, and the output of the reactive power resource pool, after being converted according to a unified direction sign in the same time slice. The difference between the two is taken as the closure error. When the closure error exceeds the error threshold, it is determined that there is a risk of inconsistent metering, missing measurement points, time delay misalignment, or double counting in the convergence of time slices, and a closure error exceeding the limit mark is written. The node voltage consistency error is used to characterize the consistency of multi-channel measurement points: the voltage measurement values ​​of the same key node in different acquisition channels or different sources are compared, or the consistency of the voltage difference between adjacent measurement points upstream and downstream of the same node within a reasonable range after considering line impedance and power flow direction is checked. The channel deviation or check residual is taken as the consistency error. When the consistency error exceeds the limit, it is determined that there are problems such as channel drift, inconsistent transformer range and ratio, asynchronous sampling, or measurement abnormality, and a node voltage consistency exceeding the limit mark is written.

[0032] like Figure 3The time-series charts of the operational data show a typical intraday variation pattern with mutual coupling on the time axis: The first chart shows the time-series curves of active and reactive power at the grid connection point. It can be seen that the grid-connected switching power exhibits obvious peak-valley fluctuations over time, and the reactive power at the grid connection point adjusts synchronously with the changes in active power, reflecting the pressure changes of PQ constraints and ramp-up constraints on the grid connection side; The second chart simultaneously shows the changes in the per-unit voltage and current carrying capacity margin of key nodes. When the grid-connected power or net injection in a zone increases, the voltage and current carrying capacity margin of key nodes will fluctuate in tandem, with a decrease in current carrying capacity margin corresponding to the... The current-carrying constraints of the main transformer are tightening, and the risk of exceeding the limits is increasing. The third figure shows the change of the main transformer hot spot temperature over time. The hot spot temperature rises during periods of high load or concentrated power flow and falls during periods of load decline, which is used to characterize the tightness of the thermal constraints of the main transformer and the influence of thermal inertia. The fourth figure shows the equipment-side operation and output status, including the fluctuation of active power output of wind turbines and photovoltaics, the compensation changes of reactive power output of inverters, and the switching of energy storage charging and discharging status. This shows that by dynamically correcting and smoothing voltage and current constraints, a calculable time-series evidence chain is provided for feasible domain modeling, bottleneck identification, and rolling scheduling.

[0033] In this implementation scheme, the closure error exceeding the limit and the consistency exceeding the limit are directly converted into quality tags, quality scores and weighting coefficients and written into the frame. This enables rolling scheduling and limit correction to use data according to credibility. When the data is reliable, it converges quickly. When the data is suspicious, it automatically suppresses aggressive correction and triggers alarms and remediation. This improves the stability, reproducibility and interpretability of scheduling decisions, and provides a traceable time-series evidence basis for bottleneck identification, sensitivity assessment and closed-loop verification.

[0034] Specifically, the process of constructing a distribution network topology model and generating a set of feasible network constraints is as follows: Input the grid-side dataset and, in conjunction with the node, branch, and partition identifiers in the joint observation data frame, construct the distribution network topology model using a graph reconstruction and consistency check algorithm based on topology prior constraints. The topology prior constraints include: switch-branch connection priors given by the primary wiring configuration library, the allowable state set for bus tie and sectionalizing switches, node numbering and voltage level consistency rules, feeder outgoing direction and transformer-bus-feeder hierarchical constraints, radial or ring network constraints, and branch endpoint nodes. Points must meet the constraints of being connected at the same voltage level or across voltage levels via main transformer branches; consistency verification includes: verification of branch connectivity based on switch remote signaling and disconnector status, topology direction consistency verification based on node voltage and branch power flow direction, zonal injection-feeder summary closure verification based on power conservation, and multi-channel voltage consistency verification based on measurement redundancy; the output of consistency verification includes at least: topology validity flags, a list of branches with suspected erroneous connections, a list of abnormal measurement points that need to be frozen or downweighted, and topology repair actions under the premise of satisfying topology prior constraints.

[0035] The primary wiring topology version number is used as the topology context identifier. Node sets and branch sets are generated based on node and branch identifiers. The branch set includes main transformer branches, bus tie branches, feeders, and branch segments. Impedance parameters and thermal stability thresholds are bound to each branch. Voltage upper and lower limit thresholds and node voltage level markers are bound to each node. The grid connection point is written into the topology model as an equivalent interface node of the external power grid. The main transformer rated capacity, cooling level, hotspot upper limit, and tap position are written into the main transformer branch as main transformer thermal constraint parameters. The feeder rated current and thermal stability threshold are written into the feeder and branch segment constraint parameters. The zoning identifier and equipment mapping relationship are written into the node attributes for determining the scope of zoning power limiting and local reactive power support.

[0036] The power distribution network topology model is made computable to generate network state variables and constraint expressions. Node voltage amplitude variables, branch current variables, branch active and reactive power flow variables, and grid connection point active and reactive power exchange variables are defined. Active and reactive power injection variables of wind turbines, inverters, energy storage converters, and reactive power devices at each access node are also defined. Node voltage upper and lower limits are converted into node voltage constraints, feeder and branch thermal stability thresholds are converted into branch current carrying capacity constraints, main transformer hotspot limits and main transformer capacity are converted into main transformer thermal constraints and capacity constraints, and grid connection point power limits and ramping constraints are converted into grid connection point exchange constraints.

[0037] A piecewise convex approximation of the power flow relationship in the distribution network is used to form a set of constraints for the network feasible region. Under the radial distribution structure, a linearized power flow approximation model is adopted to transform the nonlinear relationship between branch power flow and node voltage into piecewise linear constraints. For the relationship between main transformer hotspots and load factor, a piecewise linear temperature rise model or an equivalent thermal inertia approximation model is adopted to transform the hotspot temperature rise constraint into a time slice constraint that can be used for rolling optimization. For nodes with significant voltage over-limit risks, a voltage sensitivity coefficient matrix is ​​introduced to explicitly embed node voltage constraints into the feasible region. For nodes with different voltage levels, a uniform per-unit variable is adopted to make the feasible region constraints computable on the same scale, and the set of constraints for the network feasible region is output. The set of constraints for the network feasible region includes: node voltage constraints, branch current carrying capacity constraints, main transformer capacity and hotspot constraints, grid connection point switching and ramping constraints, power injection conservation constraints for each node, and optional reactive power resource pool constraints, which are used for PQ collaborative rolling scheduling and fast power flow verification.

[0038] Based on the source-storage side dataset and the device access node identifiers in the joint observation data frames, the capability boundaries of wind turbines, inverters, energy storage converters, and reactive power devices are embedded into the network feasible domain constraint set. The set of controllable variables is defined with time slices as the granularity. For wind turbines and inverters, available power, power limiting status, and ramping capability are transformed into active power adjustable range constraints, and power factor setting, reactive power setting, and volt-ampere characteristic curve levels are transformed into reactive power adjustable range constraints. Through the convex hull embedding algorithm for the device PQ capability curve and the energy storage P-SOC capability boundary, combined with the joint constraint splicing rules of the device capability boundary, a feasible domain enhancement model with device constraints is constructed. The joint constraint splicing rules include: aligning and splicing the network constraint set and the device capacity boundary convex hull constraint set according to the same time slice index and the same node injection mapping relationship, ensuring that each device's control variables only participate in power balance and voltage and current constraints through the access node; using joint constraints of the same capacity convex hull for the active and reactive power of the same device instead of separate constraints, avoiding splicing gaps where P is feasible but Q is not; and simultaneously splicing SOC corridor constraints, charging and discharging direction mutual exclusion constraints, and minimum duration constraints for energy storage, avoiding unrealizable instants in adjacent time slices. The solution is flipped; the minimum switching interval and locking condition are spliced ​​for the reactive resource pool to avoid the solution result being unexecutable under discrete switching constraints; a unique effective selection constraint is established for the mutual exclusion terms of the control caliber, so that the control variables within the same time slice have a unique interpretation; when infeasibility occurs after splicing, feasibility processing is carried out in the order of boundary shrinkage - caliber switching - control domain expansion: first shrink the convex hull boundary of the equipment with current limiting and derating reason codes and write it into the version key, then switch to an executable control caliber, and finally relax non-critical objectives or enable the fallback control domain under the premise of satisfying safety constraints.

[0039] The set of controllable variables includes: active power regulation of wind turbines or wind turbine groups, active power regulation of inverters or photovoltaic clusters, active power regulation of energy storage converters during charging and discharging, reactive power regulation of inverters, reactive power regulation of energy storage converters, and reactive power setting or switching status variables of reactive power devices. Each variable is bound to a device identifier, access node identifier, and partition identifier for injection mapping and scope constraints. For energy storage converters, the upper limit of chargeable power, the upper limit of dischargeable power, the state of charge corridor, and the health state coefficient are transformed into charging and discharging feasible domain constraints, and the reactive power capacity boundary under the grid-connected support mode is written into the reactive power constraints. For reactive power devices, reactive power output and availability margin are transformed into reactive power resource pool constraints, and the switching status and control mode are written into availability constraints. The feasible domain enhancement model includes: a set of decision variables and their device-node mapping relationships, a joint constraint set of network constraint set and device capacity boundary constraint set, control caliber consistency constraints, device availability constraints, and boundary contraction and update rules triggered by state, used for rolling scheduling solutions and limit overrun corrections.

[0040] This implementation plan enables the rapid, accurate, and traceable assessment of key safety boundaries such as node voltage, branch current carrying capacity, main transformer hotspots, and grid-connected switching. Simultaneously, it embeds the capacity boundaries of wind turbines, inverters, energy storage, and reactive power devices into network constraints in a unified feasible domain format. This significantly improves the stability, real-time performance, and successful implementation rate of rolling scheduling and limit correction, reduces the risks of misjudgment, oscillations, and limit violations caused by topology inconsistencies, unreliable measurements, capacity contraction, or conflicting definitions, and enhances continuous safe operation and adaptive optimization capabilities under topology switching and operating condition fluctuations.

[0041] Specifically, the process of generating the zoned control domain for sensitivity assessment is as follows: Based on the feasible region enhancement model, the node voltage constraints and branch current carrying capacity constraints are bound and a bottleneck list is generated in each time slice; the voltage margin of each node, the current carrying capacity margin of each branch, and the hot spot margin of the main transformer are calculated. If the margin is lower than the margin threshold, the corresponding constraint is determined to be in a tightly bound state; the location identifier, current margin value, change trend, and trigger reason code of the tightly bound constraints are recorded, and a bottleneck list is generated according to the scope of influence and risk level. By performing dimensionless and scale normalization processing on the vector of the quantity to be repaired and the incremental vector of the constraint space, the excess quantity and repair effect under different constraint types, different dimensions, and different rated benchmarks are uniformly mapped to a comparable per-unit scale, thereby ensuring that the sensitivity assessment values ​​have consistent comparability and numerical stability in cross-constraint, cross-equipment, and cross-zone scenarios.

[0042] Sensitivity assessment is performed on the bottleneck location to generate a mapping relationship between the bottleneck and controllable resources and determine the partitioned control domain. The excess amount of the bottleneck constraint that needs to be corrected under the current observation state is calculated, and only the positive part is retained as the positive part of the corrected amount. Then, a set of executable control action vectors is obtained by traversing the executable control set corresponding to the bottleneck, serving as the constraint space increment. For each set of candidate control actions, the L2 norm of the positive part of the corrected amount and the L2 norm of the constraint space increment are calculated respectively. The sensitivity assessment value is obtained by dividing the L2 norm of the positive part of the corrected amount by the L2 norm of the constraint space increment and adding a minimum term. The specific formula for calculating the sensitivity assessment value is as follows: ; In the formula, This represents a sensitivity assessment value, used to determine whether the bottleneck can be easily mitigated through existing control measures; It represents the positive part of the amount of bottleneck that needs to be repaired under the current observation state. It is calculated by substituting the voltage, current and hot spot temperature rise state quantities in the joint observation data frame into the bottleneck constraint function. The value range is real number, and it is used to characterize the distance of the bottleneck from the feasible region boundary. This represents the predicted repair effect on the bottleneck constraint function after the control vector is applied to the system under the current observation state, which is also known as the constraint space increment. It is obtained through a linearized power flow model or a piecewise convex approximation model, and its value range is a real number matrix. It is used to measure the actual bottleneck mitigation capability of a certain set of executable controls. The executable control domain corresponding to the bottleneck is constructed by combining the equipment capacity boundary constraints, energy storage state of charge corridor constraints, caliber consistency constraints, availability, latching constraints and partition scope constraints in the feasible domain enhancement model. The control vectors used for constraint search must be executable. It represents a set of candidate control action vectors within the executable control domain, obtained by solving the optimization process within the executable control domain, and is used to describe the joint regulation of active and reactive power by controllable resources within the partition; The bottleneck identifier is obtained through the binding and judgment process and is used to generate a bottleneck list and disposal recommendations. This represents the minimum term, which is obtained by adaptively setting the minimum resolvable action and the measurement noise level. Its value range is a real number greater than zero, used to avoid the evaluation value from diverging due to a zero or extremely small denominator and to improve the stability of real-time calculation.

[0043] The system compares the sensitivity assessment value with the sensitivity threshold in real time to trigger the determination of the zonal control domain and the selection of the handling path. When the bottleneck sensitivity assessment value is less than or equal to the sensitivity threshold, the bottleneck is determined to be a remediable bottleneck. The most sensitive and executable subset of control resources from the corresponding candidate control resource set is prioritized for entry into the zonal control domain, and a handling suggestion prioritizing local reactive power support and energy storage buffering is generated. When the bottleneck sensitivity assessment value is greater than the sensitivity threshold, the bottleneck is determined to be an unremediable bottleneck. This triggers an upgrade in the handling level and prioritizes a combination of handling paths including zonal power limiting, reactive power device switching, and topology adjustment or operation mode switching. Simultaneously, the bottleneck is marked as high-risk and its ranking priority in the bottleneck list is increased. The system outputs the zonal control domain and bottleneck handling suggestions. The zonal control domain includes: bottleneck identifier, corresponding zonal identifier, candidate control resource set, priority sequence, estimated mitigation amount, and constraint recovery target, used for generating combined control quantities for zonal power limiting, local reactive power support, and energy storage buffering.

[0044] As shown in Table 1, the bottleneck sensitivity assessment data is as follows: Bottleneck identifier BOT-001: Bottleneck type is voltage over-limit, current margin is 0.02, positive repair amount is 0.03, sensitivity assessment value is 0.45, and the solution path is local reactive power support (SVG regulation); Bottleneck identifier BOT-002: Bottleneck type is feeder current over-limit, current margin is -0.05, positive repair amount is 0.05, sensitivity assessment value is 0.62, and the solution path is zoned power limiting + energy storage buffer; Bottleneck identifier BOT-003: Bottleneck type is main transformer hotspot margin too low, current margin is 8.5, positive repair amount is 6.5, sensitivity assessment value is 0.38, and the solution path is energy storage. The system can absorb active power and regulate cooling. Bottleneck indicator BOT-004: Bottleneck type is voltage exceeding limits, current margin is 0.01, required repair amount is 0.04, sensitivity assessment value is 0.58, and the solution is reactive power regulation of the photovoltaic inverter. Bottleneck indicator BOT-005: Bottleneck type is feeder current exceeding limits, current margin is -0.03, required repair amount is 0.03, sensitivity assessment value is 0.42, and the solution is local reactive power support + load transfer. Bottleneck indicator BOT-006: Bottleneck type is main transformer hotspot margin too low, current margin is 7.2, required repair amount is 7.8, sensitivity assessment value is 0.65, and the solution is zoned power limiting + emergency cooling.

[0045] Table 1 Bottleneck Sensitivity Assessment Data

[0046] like Figure 4The bottleneck sensitivity assessment data visualization shown uses a horizontal bar chart to sort and display the sensitivity assessment values ​​of each bottleneck. The vertical axis represents the bottleneck identifier (such as BOT-012, BOT-006, etc.), and the horizontal axis represents the bottleneck sensitivity assessment value. The longer the bar, the greater the control increment required to eliminate the bottleneck's overshoot / gap within the current executable control domain; the more "difficult" the bottleneck is to alleviate and the higher the handling cost. Conversely, the smaller the bar, the easier it is to be quickly repaired with existing controllable resources. The diagram uses colors to distinguish bottleneck types (voltage exceeding limits, current exceeding limits, and low transformer hotspot margin). It's clear that bottlenecks with low transformer hotspot margins are more prevalent in high-value ranges (e.g., approximately 0.71 for BOT-012 and 0.65 for BOT-006), indicating higher priority and a greater likelihood of triggering escalation measures. Voltage exceeding limits are mostly in the medium range (e.g., approximately 0.58 for BOT-004 and 0.48 for BOT-007), while current exceeding limits are more dispersed (e.g., approximately 0.62 for BOT-002, 0.51 for BOT-008, and 0.42 for BOT-005). Furthermore, the reference dashed lines in the diagram serve as an auxiliary scale for real-time threshold comparison, dividing bottlenecks into rapidly remedial and difficult-to-remediate zones, thereby driving the determination of zoned control domains and the selection of remediation paths.

[0047] In this implementation plan, the constraint bottleneck can be quickly identified in each time slice, and the bottleneck-resource mapping and partition control domain boundary can be automatically formed. This allows the triggering and sequencing of local reactive power, energy storage buffer and partition limited generation disposal paths to be based on calculable and reproducible indicators, reducing the omissions and misjudgments caused by manual experience in domain selection, and improving the convergence stability of rolling scheduling and the timeliness of over-limit disposal.

[0048] Specifically, the process of constructing a collaborative rolling scheduling system based on the feasible domain enhancement model and the zonal control domain is as follows: Based on the feasible domain enhancement model, a rolling scheduling window is constructed and the zonal combined control variable sequence is obtained; the scheduling decision variable sequence includes: the active power setting or limiting sequence of wind turbines and inverters in each zonal, the reactive power setting or power factor setting sequence of inverters and energy storage converters in each zonal, the charging and discharging power setting sequence of energy storage converters in each zonal, and the reactive power setting or switching state sequence of the reactive power resource pool in each zonal. The scheduling decision variable sequence is then aggregated into a node injection sequence through the device-node mapping relationship; the wind power in each future time slice is then... Available power prediction, photovoltaic available power prediction, and load disturbance estimation are used as priors for available output, and the confidence level is transformed into a prediction corridor constraint. For each time slot, the wind power and photovoltaic prediction values ​​are given as point prediction values ​​and uncertainty characterization quantities, and the corridor width coefficient is obtained by mapping the confidence level. Based on the corridor width coefficient, the upper and lower bounds of the prediction corridor are constructed: the available output corridor is obtained by symmetric or asymmetric expansion of the point prediction values ​​with the upper and lower bounds. The prediction corridor constraint is written into the rolling scheduling problem in the form of inequalities to limit the available output and net injection of each time slot within the corridor range, and serves as a prior boundary for over-limit verification and correction.

[0049] A hierarchical optimization objective solution strategy is adopted to generate a sequence of combined control variables. The first objective is to maximize absorption, prioritizing the minimization of abandoned power and unplanned power generation, so that the active power exchange at the grid connection point is as close to the upper limit as possible without triggering internal limit exceedances. The second objective is to suppress limit exceedance risks, prioritizing the repair of bottlenecks with high risk levels or high sensitivity assessment values ​​in the bottleneck list, so that the node voltage margin, branch current margin, and main transformer hotspot margin are restored to the safe corridor. The third objective is to achieve control smoothness and cost constraints, prioritizing the minimization of inter-time slice setpoint jumps, equipment mode switching frequency, and frequent reversals of energy storage charging and discharging directions, reducing execution jitter and equipment stress. The hierarchical solution is achieved by satisfying the upper-level objectives first and then optimizing the lower-level objectives within the feasible solution set, avoiding simple weighting of multiple objectives. The output is a sequence of combined control variables for each region. The zonal combined control quantity sequence includes: time slice index, zonal identifier, active power setting or limit for wind turbines and inverters, reactive power setting or power factor setting for inverters and energy storage converters, energy storage charging and discharging setting and direction mark, reactive power setting or switching status of reactive power resource pool, active and reactive power planned values ​​at grid connection point, and constraint margin prediction value and risk label for each time slice, used for fast power flow verification and limit correction.

[0050] In this implementation scheme, the feasibility and robustness of the scheduling solution can still be maintained even in the presence of source-side fluctuations and load disturbances. By using the feasible domain enhancement model and the zonal control domain to limit the scope and capacity boundary of scheduling variables, the optimization solution and bottleneck relief can be directly coupled, reducing the spillover of over-limit risks and cross-regional mutual constraints. It effectively reduces the execution stress and control oscillations caused by frequent power rationing, equipment mode jitter and energy storage direction reversal, and improves the stability, executability and friendliness of the zonal combined control quantity sequence for fast power flow verification and over-limit correction.

[0051] Specifically, the process of performing power flow verification and limit correction to generate safe control quantities is as follows: Input the partitioned combined control quantity sequence; perform power flow verification and limit correction on the control quantities for each time slice, and generate a scheduling plan data frame; substitute the node injection sequence of each time slice into a linearized power flow or piecewise convex approximation model to calculate the predicted values ​​of node voltage, branch current, main transformer load rate, and hotspot response, and perform constraint-by-constraint verification with the node voltage upper and lower limits, branch thermal stability threshold, main transformer hotspot upper limit, and grid connection point exchange and ramping constraints, outputting voltage limit exceedance, current limit exceedance, hotspot limit exceedance, and location identifiers; when any constraint exceeds the limit or the margin falls below the safety corridor, the limit correction process is triggered; multiply the constraint sensitivity matrix with the transpose matrix to obtain a matrix product; multiply the damping stability coefficient with the identity matrix to obtain the damping matrix; and then... The product and damping matrix are added to obtain a matrix sum; the invertibility of the matrix sum is checked by calculating the condition number and eigenvalues ​​of the matrix to ensure non-singularity. When the invertibility condition is met, the matrix sum is inverted to obtain the inverse matrix; the inverse matrix is ​​multiplied by the vector to be repaired to obtain the intermediate vector; the constraint sensitivity matrix is ​​calculated using the online linearized power flow model and the hotspot approximation model under the current observation state, and the transpose of the constraint sensitivity matrix is ​​multiplied by the intermediate vector to obtain the correction direction vector; the correction direction vector is substituted into the projection function to obtain the projected vector; the negative value of the projected vector is taken to obtain the over-limit correction control increment; the over-limit correction control increment is written as the safety control quantity to obtain the corrected safety control quantity; the corrected safety control quantity is re-verified by power flow to obtain the remaining over-limit quantity; the specific calculation formula of the over-limit correction control increment is as follows: ; In the formula, This indicates the over-limit correction control increment, which is used to make minimal changes to the control quantity of the current time slice without changing the overall structure of the original rolling scheduling control quantity, so that voltage over-limit, current over-limit, and hot spot over-limit can all retreat to the safety corridor at the same time. The constraint sensitivity matrix is ​​calculated using the online linearized power flow model and the hot spot approximation model under the current observation conditions. It is used to map the over-limit repair requirements of the constraint space back to the control variable space to form the correction direction. The damping stability coefficient is adaptively set by the available control margin of the current time slice, the measurement noise level, and the tightness of the constraint binding. Its value range is a real number greater than zero. It is used to suppress the divergence of the correction amount and improve the numerical stability when the sensitivity matrix is ​​close to singular, the controllable margin is insufficient, or the measurement noise is large. The identity matrix is ​​determined by the model dimension and is used to ensure stable solution of the correction amount even when the constraint terms are highly correlated or the sensitivity matrix is ​​ill-conditioned. This represents the vector of quantities that need to be repaired, which is calculated from the power flow verification results and is used to characterize the over-limit or margin gap quantities of all the constraints of concern in the current time slice. The projection function is obtained by first truncating each type of control variable according to its upper and lower bounds, then performing pattern pruning according to mutual exclusion constraints, correcting the consistency of the energy storage charging and discharging direction and minimum duration, filtering the switching feasibility of the minimum interval for reactive power device switching, and performing smooth projection on the maximum jump constraint between time slices. It is used to ensure that the over-limit correction can be implemented and will not introduce new unexecutable instructions.

[0052] The remaining excess amount is compared with the safety corridor threshold in real time. When the remaining excess amount is less than or equal to the safety corridor threshold, it is determined that the over-limit correction has met the standard and a safety control quantity is output. When the remaining excess amount is greater than the safety corridor threshold, it is determined that the correction has not met the standard and an upgrade correction path is triggered. The callable control domain is expanded step by step in the order of local reactive power support and reactive power resource pool switching, energy storage absorption or release buffer, and zoned limited generation as a fallback, and the over-limit correction control increment is recalculated until the remaining excess amount returns to the safety corridor or a high-risk alarm is triggered. The correction result is written back as a safety control quantity sequence. Each time slice is written with a verification tag, correction reason code, and correction magnitude summary; the safety control quantity sequence is split into partition plan sub-sequences according to the partition identifier, and the partition plan sub-sequences are mapped to equipment-level settings according to the equipment identifier; for different equipment using reactive power setting control, power factor control, and volt-ampere characteristic curve control, the control mode is uniformly selected according to the control caliber subset and uniquely interpreted setting parameters are generated to avoid mutually exclusive mode instructions in the same time slice; charging and discharging direction constraints, state of charge protection information, and minimum duration constraints are added to energy storage control instructions, and minimum switching intervals and interlocking conditions are added to reactive power resource pool switching instructions to ensure that equipment-level instructions can be directly executed; the equipment-level instruction plan and time slice index are bound and encapsulated into a scheduling plan data frame.

[0053] In this implementation plan, the remaining excess amount is obtained by re-verifying the power flow after correction and compared with the safety corridor threshold in real time. When the standard is met, the safety control quantity can be directly output. When the standard is not met, the callable control domain is expanded to recalculate the correction quantity, ensuring that the risk of exceeding the limit can converge step by step under a controllable handling path. The control mode and setting parameters are mapped to the device level and are uniquely interpreted, so that the scheduling plan can be traced and reproduced and can be directly executed by the equipment, reducing the on-site execution deviation and the difficulty of post-event review.

[0054] Specifically, the process of generating a closed-loop verification data frame by performing consistency checks on receipts and status readbacks is as follows: Figure 5 The flowchart shown illustrates the process of generating a closed-loop verification data frame for consistency checks. The process involves inputting a scheduling plan data frame, performing a pre-readback confirmation based on the time slice index to retrieve the current operating mode and status, and issuing setting instructions only after the pre-conditions are met. For each instruction transaction, an instruction transaction identifier, idempotent key, and session identifier are generated. The topology version number, time slice index, device identifier, control mode flag, active power setting or limit, reactive power setting or power factor setting, energy storage charging / discharging setting and direction flag, reactive power resource pool switching status, and a summary of pre-conditions are written into the instruction transaction header. Pre-conditions include: device online status, interlocked status, control mode mutual exclusion conditions, minimum switching interval, minimum duration, ramp limit, and capacity boundary validity. For devices supporting two-stage delivery, acknowledgment collection and status readback collection are performed for each instruction transaction to construct an execution proof package set. Blind issuance in error modes or interlocked states is avoided. For devices not supporting two-stage delivery, a forced readback method after issuance is used to complete the confirmation link. The receipts include: communication layer delivery receipts, device-side reception receipts, device-side parameter activation receipts, or operating mode switching receipts; the status readbacks include: the device's current control mode, current active and reactive power output, current setpoint mirroring, current limiting or derating reason codes, interlocking flags, and critical protection status; each receipt and readback record is written with the device's local timestamp, gateway reception timestamp, and station-side aggregation timestamp, and bound to the instruction transaction identifier to ensure that the execution link is traceable and reproducible.

[0055] The consistency of the execution proof package set is checked, an execution deviation vector is generated, and a closed-loop verifiable value is output. Five deviation ratios are obtained by dividing the setting consistency deviation by the setting consistency tolerance, the effective consistency deviation by the effective consistency tolerance, the timeliness consistency deviation by the timeliness tolerance, the evidence completeness deviation by the evidence completeness tolerance, and the control strategy consistency deviation by the control strategy tolerance. The maximum value among these five deviation ratios is taken as the worst-case deviation ratio. The worst-case deviation ratio is multiplied by a decay coefficient, negativeened, and then exponentially calculated to obtain the closed-loop verifiable value. The specific calculation formula for the closed-loop verifiable value is as follows: ; In the formula, This indicates a closed-loop verifiable value and its purpose. This represents the verification attenuation coefficient, which is obtained by adaptive calibration through statistical analysis of verification values ​​under different topology versions, equipment types, and operating conditions in historical operating data. The value range is a real number greater than zero, and it is used to adjust the sensitivity of the verification value to deviation. This indicates the consistency deviation, which is obtained by comparing the mirrored setting value read back with the setting value in the scheduling plan field by field. It is used to characterize whether the issued setting is consistent with the mirrored setting on the device. This indicates the setting of consistency tolerance, which is determined jointly by device resolution, control dead zone, communication quantization error and service allowable deviation, and can be dynamically updated according to device type and control mode. It is used for scale normalization and to form a comparable deviation ratio. This indicates the consistency deviation, which is obtained by comparing the actual output read back with the expected transition direction, and is used to characterize whether the setting has truly taken effect. The effective consistency tolerance is determined by measuring the upper bound of steady-state and transient errors of noise level, equipment dynamic response capability, control bandwidth and response delay statistics. It is used for scale normalization and to avoid incomparability of deviations between different equipment and different operating conditions. It indicates the timeliness consistency deviation, which is obtained by comparing the receipt arrival time with the time limit threshold, and is used to characterize whether the link meets the time constraints; The timeliness tolerance is determined adaptively by controlling the receipt time limit in the contract, the real-time requirements of the station, and the historical link jitter distribution. It is used to normalize the timeliness deviation to a dimensionless scale. It indicates the completeness deviation of evidence, which is obtained by comparing the reason codes of blocking, flow limiting, and rate reduction with the scheduling context, and is used to characterize whether the evidence chain is complete and traceable. It represents the tolerance for the completeness of evidence, which is determined by the device communication capability declaration, optional receipt level policy and the upper limit of evidence gap allowed by the business, and is used to convert the degree of evidence missing into a comparable deviation ratio; This indicates the consistency deviation of the control strategy. It is obtained by statistically analyzing the missing information of the acknowledgment items and readback items, and is used to characterize whether the explanations for why it did not take effect and how it took effect are consistent. The tolerance of the control strategy is determined adaptively through the strictness level of the caliber rules, the layering of reason code credibility, and the statistics of historical false alarm rates. It is used to normalize the degree of semantic inconsistency to a comparable scale and avoid the incomparability of reason code systems of different devices.

[0056] The system compares the closed-loop verifiable value with the verification threshold in real time. When the closed-loop verifiable value is higher than or equal to the verification threshold, the system determines that the time-slice instruction execution is trustworthy and completes the closed-loop confirmation. The time slice is marked as verified and the verification tag is written back. The receipt arrival time, read-back set image, and read-back actual output are written as execution proofs into the closed-loop verification data frame and archived. The device status image and actual output are written back to the next time slice of the joint observation data frame as the initial status value. The system releases the idempotent key of the instruction transaction corresponding to the time slice and closes the session window, allowing the issuance of control instructions for the next time slice. The system updates the health score and trustworthiness tag for continuously passing devices and links, which are used for control domain selection and threshold adaptive updates. When the closed-loop verifiable value is lower than the verification threshold, the execution of the time-slice instruction is deemed untrustworthy, triggering a rollback and retry strategy. The time slice is marked as verification failed, and the failure reason code and the weakest link deviation type are written back. The control scope of the device is frozen for a certain number of time slices, or its capability boundaries are narrowed to reduce the risk of non-executability. Based on different failure types such as missing receipts, inconsistent settings, inconsistent effectiveness, time limits exceeding limits, and conflicting scopes, retry, downgrade, or rollback to the safe setting's handling path is selected. The handling results and the secondary readback results are continued to be written into the closed-loop verification data frame for review and model update, and the closed-loop verification data frame is output. The closed-loop verification data frame includes: topology version number, time slice index, partition identifier, device identifier, instruction transaction identifier, control mode flag, planned setting, receipt type and receipt arrival time, read-back setting image and read-back actual output, current limiting or derating reason code, lockout and protection status, setting consistency deviation, effectiveness consistency deviation, timeliness consistency deviation, caliber consistency deviation, evidence integrity deviation, closed-loop verifiable value, verification label and handling suggestions, which are used for subsequent anomaly handling, online rollback and model update.

[0057] This implementation plan ensures the accuracy and executability of control commands through a rigorous closed-loop verification process, combined with consistency checks on receipts and status readbacks. Remedial measures are implemented through multiple paths, and the data is further written into the closed-loop verification data frame for review and model updates. This effectively improves execution accuracy and robustness, ensures stable operation of equipment and networks, and provides precise data support for optimization.

[0058] Specifically, the process of anomaly attribution and online rollback replanning is as follows: For time slices that fail verification, anomaly attribution is performed based on the dimensions of receipt timeliness, readback consistency, equipment availability, and capability boundary contraction, triggering rollback and replanning. When receipt loss or timeout dominates, it is determined to be a communication link or equipment offline anomaly, triggering a degraded control mode and a rollback strategy that maintains the previous moment's safety settings, and temporarily removing the device from the executable control domain of the current window. When inconsistent settings dominate, it is determined to be a control caliber conflict or mutual exclusion mode not met, triggering caliber reorganization and mode correction strategies, prioritizing switching to the control caliber with a unique interpretation before re-issuing. When the actual output is inconsistent with the expected transition and is accompanied by a rate limiting or derating reason code, it is determined to be a capability boundary contraction anomaly, triggering a capability boundary online contraction and resolving the closed-loop verifiable value strategy, and writing the anomaly into the reason code for statistical purposes.

[0059] The system drives online updates of the feasible region enhancement model based on closed-loop verification data frames. It calibrates the effective boundary of equipment capacity curves based on readbacks and actual outputs. When similar current limiting occurs, it adjusts the linearized power flow sensitivity matrix using residual-driven adjustments based on measured node voltages, branch currents, and hotspot responses after execution. It automatically shrinks the dimensionless capacity polygon boundary of the corresponding equipment and writes the boundary version number. When the residual is consistently large, it increases the damping stability coefficient and triggers relinearization to avoid verification and correction distortion. When the topology version number changes or the power flow path changes due to bus tie or segment switch status switching, it triggers topology context switching and reconstructs the constraint graph. It outputs online rollback and replanning results, including: capacity boundary version number, boundary shrinkage magnitude summary, sensitivity matrix update marker, damping coefficient update marker, topology context switching marker, anomaly cause code statistical summary, and equipment availability status. These results support collaborative scheduling and continuous self-learning maintenance in the next rolling window.

[0060] The effective boundary calibration specifically includes: using the closed-loop verification data frame as the calibration sample, under the same control mode and the same topology version context, constructing an effective mapping relationship from the set quantity to the actual achievable output; for the P-Q capability boundary of the inverter and PCS, performing ray scanning of the boundary according to several directional vectors or segmenting statistics according to the power factor level and reactive power setting level, taking the maximum achievable output point that can still be stably reached and verified as the effective boundary point under a given setting, and eliminating outliers by using the quantile value of the sliding window or robust upper bound estimation; when the same current limiting and derating reason codes appear in several consecutive time slices and the effective achievement rate in the corresponding direction is lower than the threshold, shrinking the boundary points in the direction inward to form a new capability polygon, and writing the boundary version number and effective time, ensuring that the scheduling solution uses the achievable boundary instead of the nominal boundary.

[0061] The residual-driven adjustment specifically includes: constructing a residual vector based on the predicted values ​​of the power flow verification and the measured values ​​after execution, and weighting the residuals according to the quality score of the measurement points; estimating the sensitivity drift direction by utilizing the correlation between the residuals and the control increment and injection changes, and incrementally correcting the linearized sensitivity matrix by using recursive least squares and online updates with forgetting factors or small step gradient correction, while applying damping regularization to the ill-conditioned direction to avoid matrix divergence; when the residuals are systematically biased within a continuous window and exceed the residual threshold, triggering the relinearization process: recalculating the Jacobian and piecewise coefficients with the current joint observation data frame as the linearization working point and updating the sensitivity matrix version; when a topology context switch occurs, clearing the residual state of the previous topology and reconstructing the constraint graph and sensitivity matrix to avoid distortion of the verification and over-limit correction caused by cross-topology residual migration.

[0062] This implementation avoids repeated failures caused by scheduling solutions relying on unreachable boundaries; at the same time, it uses predicted and measured residuals to recursively update the linearized power flow sensitivity matrix and re-linearize it when necessary, and switches the context to reconstruct the constraint graph when the topology changes, ensuring that the verification and limit correction models always fit the actual operation, thereby continuously reducing the risk of incorrect correction and control oscillation, and improving the feasibility, stability and self-learning operation and maintenance capabilities of the next rolling window collaborative scheduling.

[0063] Specifically, the second aspect of this invention provides an integrated wind-solar-storage energy management system, applied to an integrated wind-solar-storage energy management and scheduling method, comprising: a multi-source data acquisition module, used to collaboratively access various grid-side, source-storage-side, and contextual time-series data at the substation, perform unified time base alignment, caliber normalization, and quality detection, and encapsulate them into joint observation data frames and historical operation data according to time slices; providing reliable input for feasible domain modeling, bottleneck identification, and scheduling solutions. A feasible domain modeling module is used to construct a distribution network topology model, perform piecewise convex approximation of power flow relationships to form a set of feasible domain constraints, and embed consistency constraints into the feasible domain to form a feasible domain enhancement model; simultaneously, within each time slice, it performs constraint tight-binding determination, sensitivity assessment, and partition control domain generation, outputting a bottleneck list and partition handling suggestions. The collaborative rolling scheduling module incorporates predicted output corridors, network constraints, equipment capacity boundary constraints, caliber consistency constraints, and state-triggered boundary contraction rules as hard constraints into the scheduling problem within the rolling scheduling window. It generates a sequence of partitioned combined control variables using a hierarchical objective solution strategy, performs power flow verification and limit overrun correction on candidate control variables, and outputs safety control variables and scheduling plan data frames. The scheduling execution closed-loop module collects multi-level receipts and state readbacks to construct an execution proof package set. It triggers rollback or retry based on thresholds for consistency of settings, consistency of effectiveness, consistency of timeliness, and consistency of caliber, and outputs closed-loop verification data frames for anomaly handling, online backfilling, and model updates.

[0064] This implementation plan avoids the unexecutability on-site caused by optimization at the planning level alone. It can make real-time judgments and handle inconsistencies, ineffectiveness, timeouts, and conflicting standards, triggering rollbacks, retrying, and control domain shrinkage. At the same time, it writes back the verification data to drive online model calibration and adaptive parameter updates, thereby improving the feasibility, operational safety, interpretability and reproducibility of anomalies, and continuous self-learning operation and maintenance efficiency of wind-solar-storage collaborative control.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for integrated wind, solar, and energy storage management and dispatching, characterized in that, Includes the following steps: S1, collect multiple types of running datasets, and perform unified time base alignment and normalization on the multiple types of running datasets to form a joint observation data frame; S2, construct the power distribution network topology model, generate the network feasible domain constraint set, and perform sensitivity evaluation to generate the partition control domain; S3, based on the feasible domain enhancement model and the partition control domain, constructs a collaborative rolling schedule, performs power flow verification and limit violation correction, and generates safety control quantities; S4 performs consistency verification of receipts and status readbacks to generate closed-loop verification data frames, and performs anomaly attribution and online rollback replanning.

2. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 1, characterized in that: The specific process for collecting multiple types of runtime datasets is as follows: The system collaboratively accesses and collects multiple types of operational datasets through the station-side edge acquisition gateway. These datasets include: grid-side datasets, source-storage-side datasets, monitoring datasets, and time-series identifier datasets. Operational data that has already been stored within the same power distribution structure of the same station and park is used as historical operational data. The collected operational datasets are classified, labeled, and prioritized.

3. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 2, characterized in that: The specific process of performing unified time-base alignment and normalization on multiple types of running datasets to form a joint observation data frame is as follows: A unified time service is used to perform periodic time base alignment processing on the station edge gateway and the telemetry and control device to obtain processed multi-class operational datasets. The processed multi-class operational datasets are then normalized, and missing measurement detection, duplicate packet detection, spike detection, and rate of change boundary detection are performed. The grid connection point power closure error and node voltage consistency error are used as consistency quality indicators. The consistency quality indicator of each record is written to the quality tag field in the joint observation data frame to record the results of missing measurement, spike, backfill segment, time delay anomaly, and closure error exceeding the limit. The processed multi-class runtime datasets within each time slice are encapsulated with a unified data structure and their fields are bound to construct a joint observation data frame.

4. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 1, characterized in that: The specific process of constructing the power distribution network topology model and generating the network feasible region constraint set is as follows: Inputting the grid-side dataset and combining it with node, branch, and partition identifiers from joint observation data frames, a distribution network topology model is constructed using a graph reconstruction and consistency check algorithm based on topology prior constraints. The primary wiring topology version number is used as the topology context identifier, and node and branch sets are generated according to node and branch identifiers. The distribution network topology model is then processed to be computable, generating network state variables and constraint expressions. A piecewise convex approximation of the power flow relationship in the distribution network is applied, forming a set of feasible region constraints. Under a radial distribution structure, a linearized power flow approximation model is used to transform the nonlinear relationship between branch power flow and node voltage into piecewise linear constraints, outputting the set of feasible region constraints. Based on the source-storage side dataset and the device access node identifiers in the joint observation data frames, the capability boundaries of wind turbines, inverters, energy storage converters, and reactive power devices are embedded into the network feasible domain constraint set. By using the convex hull embedding algorithm for the device PQ capability curve and the energy storage P-SOC capability boundary, combined with the joint constraint splicing rules of the device capability boundary, a feasible domain enhancement model with device constraints is constructed.

5. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 4, characterized in that: The specific process for performing sensitivity assessment and generating partitioned control domains is as follows: Based on the feasible region enhancement model, the node voltage constraint and branch current carrying capacity constraint are bound and determined in each time slice, and a bottleneck list is generated. Perform sensitivity assessment on bottleneck locations, generate a mapping relationship between bottlenecks and controllable resources, and determine the partition control domain; Calculate the excess amount of the bottleneck constraint that needs to be repaired under the current observation state, and retain only the positive part as the positive part of the amount to be repaired; then traverse the set of executable controls corresponding to the bottleneck to obtain a set of executable control action vectors as the constraint space increment; for each set of candidate control actions, calculate the L2 norm of the positive part of the amount to be repaired and the L2 norm of the constraint space increment respectively, divide the L2 norm of the positive part of the amount to be repaired by the L2 norm of the constraint space increment and add a minimum term to obtain the sensitivity evaluation value; The sensitivity assessment value is compared with the sensitivity threshold in real time to trigger the determination of the partition control domain and the selection of the treatment path. When the bottleneck sensitivity assessment value is less than or equal to the sensitivity threshold, the bottleneck is determined to be a remediable bottleneck. When the bottleneck sensitivity assessment value is greater than the sensitivity threshold, the bottleneck is determined to be an unremediable bottleneck, and the partition control domain and bottleneck treatment suggestions are output.

6. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 5, characterized in that: The specific process of constructing a collaborative rolling schedule based on the feasible domain enhancement model and the partition control domain is as follows: Based on the feasible region enhancement model, a rolling scheduling window is constructed and the partitioned combined control quantity sequence is obtained by solving it; the wind power availability prediction, photovoltaic power availability prediction and load disturbance estimation of each future time slice are used as available output priors, and the confidence label is transformed into prediction corridor constraints; A hierarchical optimization objective solution strategy is adopted to generate a sequence of combined control variables, and the resulting partitioned combined control variable sequence is output.

7. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 6, characterized in that: The specific process of performing power flow verification and over-limit correction to generate safety control quantities is as follows: Input the partitioned combined control quantity sequence, perform power flow verification and limit overrun correction on the control quantity of each time slice, and generate a scheduling plan data frame; multiply the constraint sensitivity matrix and the transpose matrix to obtain the matrix product; multiply the damping stability coefficient and the identity matrix to obtain the damping matrix; Add the matrix product and the damping matrix to obtain the matrix sum; perform an invertibility check on the matrix sum by calculating the condition number and eigenvalues ​​of the matrix to ensure non-singularity. When the invertibility condition is met, invert the matrix sum to obtain the inverse matrix; multiply the inverse matrix with the vector to be repaired to obtain the intermediate vector; calculate the constraint sensitivity matrix under the current observation state using the online linearized power flow model and the hotspot approximation model; multiply the intermediate vector with the transpose of the constraint sensitivity matrix to obtain the correction direction vector; substitute the correction direction vector into the projection function to obtain the projected vector; take the negative value of the projected vector to obtain the over-limit correction control increment; Write the over-limit correction control increment as the safety control quantity, and obtain the corrected safety control quantity; The power flow is rechecked on the corrected safety control quantity to obtain the remaining over-limit quantity; the remaining over-limit quantity is compared with the safety corridor threshold in real time. When the remaining over-limit quantity is less than or equal to the safety corridor threshold, it is determined that the over-limit correction has met the standard and the safety control quantity is output. When the remaining excess quantity is greater than the safety corridor threshold, the correction is deemed unsatisfactory, and the correction result is written back as a safety control quantity sequence.

8. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 7, characterized in that: The specific process for generating a closed-loop verification data frame by performing consistency verification of receipts and status readbacks is as follows: Input the scheduling plan data frame, perform a pre-readback confirmation according to the time slice index to read the current running mode and status, and issue the setting instruction only after the preconditions are met. For each instruction transaction, perform receipt collection and status readback collection to construct an execution proof package set. Perform consistency verification on the execution proof package set, generate an execution deviation vector and output a closed-loop verifiable value. Divide the setting consistency deviation by the setting consistency tolerance, the effective consistency deviation by the effective consistency tolerance, the timeliness consistency deviation by the timeliness tolerance, the evidence completeness deviation by the evidence completeness tolerance, and the control strategy consistency deviation by the control strategy tolerance to obtain five deviation ratios. Take the maximum value of these five deviation ratios as the worst deviation ratio. Multiply the worst deviation ratio by the attenuation coefficient, take the negative, and perform an exponential operation to obtain the closed-loop verifiable value. The system compares the closed-loop verifiable value with the verification threshold in real time. When the closed-loop verifiable value is higher than or equal to the verification threshold, the execution of the time-slice instruction is deemed trustworthy and the closed-loop confirmation is completed. When the closed-loop verifiable value is lower than the verification threshold, the execution of the time-slice instruction is deemed untrustworthy and a rollback and retry strategy is triggered, and a closed-loop verification data frame is output.

9. The integrated wind, solar, and energy storage energy management and dispatch method according to claim 8, characterized in that: The specific process of anomaly attribution and online rollback replanning is as follows: For time slices that fail verification, anomalies are attributed based on the dimensions of receipt timeliness, readback consistency, equipment availability, and capacity boundary contraction, triggering rollback and replanning. The closed-loop verification data frame drives the online update of the feasible domain enhancement model. Based on the readback and actual output, the effective boundary of the equipment capacity curve is calibrated. When the same current limiting occurs, the linearized power flow sensitivity matrix is ​​adjusted based on the measured node voltage, branch current, and hotspot response after execution, and the online rollback and replanning results are output.

10. A wind-solar-storage integrated energy management and dispatch system, employing the wind-solar-storage integrated energy management and dispatch method as described in any one of claims 1-9, characterized in that, include: The multi-source data acquisition module is used to coordinate various grid-side, source-storage-side and context-based time-series data at the access station, perform unified time base alignment, caliber normalization and quality detection, and encapsulate them into joint observation data frames and historical operation data according to time slices; The feasible region modeling module is used to construct a power distribution network topology model, perform piecewise convex approximation of power flow relationships to form a set of feasible region constraints, and embed consistency constraints into the feasible region to form a feasible region enhancement model. The collaborative rolling scheduling module is used to generate a sequence of partitioned combined control quantities using a hierarchical objective solution strategy, and to perform power flow verification and limit violation correction on candidate control quantities, and output safety control quantities and scheduling plan data frames. The scheduling and execution closed-loop module is used to collect multi-level receipts and status readbacks to construct an execution proof package set, trigger rollback or retry according to thresholds, and output closed-loop verification data frames.

Images