A source network load storage control system and method based on multi-agent scheduling

Abstract

The application discloses a source network load storage control system and method based on multi-agent scheduling, comprising the following steps: collecting distributed power output, load demand, energy storage state of charge, line capacity, voltage safety boundary and historical scheduling results, and establishing an operation state sequence; performing segmented linearization and safety margin contraction on power balance, line capacity, voltage safety and energy storage state of charge constraints to obtain a constraint set; constructing a dynamic factor graph, writing the subject adjustment margin, local power deviation and constraint proximity into the composite message of the improved Max-Sum algorithm to obtain a candidate scheduling vector; obtaining a feasible scheduling vector and a classification projection residual through the Dykstra projection algorithm, and performing closed-loop iteration after converting the classification projection residual into a feedback message parameter to output a source network load storage control instruction. The application realizes constraint projection correction and feedback message updating in the multi-agent scheduling process of the source network load storage.

Description

Technical Field

[0001] This invention relates to the field of power network dispatch and control technology, and in particular to a source-grid-load-storage control system and method based on multi-agent dispatch. Background Technology

[0002] With the increasing integration of distributed power sources, adjustable loads, and energy storage devices into distribution networks, coordinated control of power generation, grid, load, and energy storage has become an important technical direction in the automation control of AC power networks. Existing power generation, grid, load, and energy storage control mainly revolves around day-ahead planning, real-time dispatching, and safety verification. It uses data collection on distributed power source output, load demand, energy storage status of charge, line capacity, and voltage boundaries to form the basis for dispatching decisions, and generates control results under the constraints of power balance, line capacity, voltage safety, and energy storage operation boundaries.

[0003] In existing technologies, centralized optimization scheduling methods rely on global state data and a unified solver, enabling the formation of scheduling plans under conditions of data integrity and stable communication. However, in scenarios involving multiple agents, autonomous edge sides, and rapid state changes, they place high demands on communication links, computational scale, and data synchronization. Distributed multi-agent scheduling methods model the source side, grid side, load side, and energy storage side as scheduling agents, forming local decisions through information interaction between agents. This approach is suitable for multi-agent collaborative control scenarios. However, in grid environments with strong constraint coupling, the results of local message transmission are easily affected by fluctuations in distributed power generation output, sudden load changes, and changes in energy storage state boundaries.

[0004] Existing methods for handling constraints in source-grid-load-storage coordinated control often employ fixed penalty coefficients, static safety margins, centralized constraint corrections, or simple out-of-bounds truncation to pull candidate scheduling results back to a feasible range. These approaches fail to adequately characterize the coupling relationships between power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state-of-charge constraints. They also struggle to promptly feed back constraint deviations to the next round of multi-agent scheduling, leading to discrepancies between scheduling decisions and the actual feasible operational boundaries.

[0005] Therefore, how to provide a source-grid-load-storage control system and method based on multi-agent scheduling is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a source-grid-load-storage control system and method based on multi-agent scheduling. This invention realizes constraint projection correction and feedback message closed-loop update in multi-agent scheduling.

[0007] A source-grid-load-storage control method based on multi-agent scheduling according to an embodiment of the present invention includes:

[0008] Collect data on distributed power generation output, load demand, energy storage status of charge, line capacity, voltage safety boundary, and historical scheduling results, and establish an operational status sequence according to the scheduling period.

[0009] Based on the operating state sequence, piecewise linearization and safety margin shrinkage are performed on the power balance constraint, line capacity constraint, voltage safety constraint and energy storage state of charge constraint respectively, and constraint type labels are configured to obtain the constraint set;

[0010] Based on the operating status sequence, determine the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body and energy storage-side main body, and construct a dynamic factor diagram;

[0011] The adjustment margin, local power deviation, and constraint proximity of each subject in the dynamic factor graph are written into the composite message of the improved Max-Sum algorithm, and message iteration is performed in the dynamic factor graph according to the composite message to obtain the candidate scheduling vector.

[0012] The candidate scheduling vector is input into the Dykstra projection algorithm, which performs alternating projections between the constraint sets to obtain the feasible scheduling vector. The classification projection residual is then calculated based on the deviation between the feasible scheduling vector and the candidate scheduling vector.

[0013] According to the constraint type of the classification projection residual, the classification projection residual is converted into the factor node penalty weight increment, the adjacent subject edge weight attenuation, the network-side node voltage constraint weight increment, and the energy storage node message damping, respectively, to obtain the feedback message parameters.

[0014] The feasible scheduling vector is used as the basis for the next round of scheduling. The improved Max-Sum algorithm is re-executed based on the feedback message parameters. When the composite message change and the classification projection residual both meet the preset convergence conditions, the feasible scheduling vector of the last round is converted into source-grid-load-storage control instructions.

[0015] Optionally, establishing the running status sequence according to the scheduling period specifically involves:

[0016] Establish a scheduling period index sequence based on the scheduling period number and execution time boundary in the historical scheduling results, and write the output of distributed power sources, load demand and energy storage charge status into the corresponding scheduling period index to obtain the main status index record;

[0017] Extract the distributed power output and load demand within the same scheduling period from the main status index record, calculate the net power deviation value, and generate output surplus markers, load gap markers, and balance markers based on the comparison result of the net power deviation value with 0. Then, mark them with the energy storage charge status within the same scheduling period to obtain the time period power-energy storage coupling record.

[0018] The line capacity and voltage safety boundary are written into the time period power storage coupling record according to the same scheduling time period index, and the network boundary change record is generated according to the absolute value of the difference between the line capacity and the absolute value of the difference between the voltage safety boundary in adjacent scheduling time periods.

[0019] The baseline value of boundary change is calculated based on the scheduling adjustment range of adjacent scheduling periods in the historical scheduling results, and compared and filtered with the network boundary change records to obtain the scheduling period of boundary change.

[0020] The power storage coupling records of the previous and next scheduling periods of the boundary mutation scheduling period are called and corrected to obtain the corrected period state records.

[0021] The status records of the correction period are rearranged according to the scheduling period index sequence, and the correspondence between the corresponding scheduling period number, execution time boundary and historical scheduling result is retained to obtain the running status sequence.

[0022] Optionally, the step of performing piecewise linearization and safety margin shrinkage on power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints respectively is as follows:

[0023] Extract time-period power storage coupling records, network boundary change records, boundary mutation markers, line capacity, voltage safety boundary, and energy storage charge state from the operating state sequence, and establish constraint state records according to the scheduling time period index;

[0024] Based on the constraint state record, generate the initial power balance constraint segment, the initial line capacity constraint segment, the initial voltage safety constraint segment, and the initial energy storage state of charge constraint segment respectively.

[0025] The absolute values ​​of the line capacity difference and voltage safety boundary difference in the network boundary change record are compared with the boundary change benchmark value, and the line capacity contraction level and voltage safety boundary contraction level are determined based on the comparison results.

[0026] Based on the boundary mutation marker, the energy storage charge status of the corresponding scheduling period is invoked, and it is jointly judged with the output surplus marker, load gap marker and balance marker to determine the energy storage charge status contraction level.

[0027] According to the line capacity contraction level, voltage safety boundary contraction level, and energy storage state of charge contraction level, the boundary values ​​of the initial line capacity constraint segment, the initial voltage safety constraint segment, and the initial energy storage state of charge constraint segment are reduced in stages to obtain the contraction constraint segment.

[0028] The initial power balance constraint fragment and the contraction constraint fragment are reorganized according to the scheduling period index, and the constraint type labels corresponding to the power balance constraint, line capacity constraint, voltage safety constraint and energy storage state of charge constraint are configured respectively to obtain the constraint set.

[0029] Optionally, the step of determining the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body, and energy storage-side main body based on the operating state sequence, and constructing a dynamic factor graph, specifically involves:

[0030] Establish the main mapping record based on the operating status sequence;

[0031] Based on the distributed power output, load demand, and energy storage charge status in the main mapping record, the node variables of the source-side main body, load-side main body, and energy storage-side main body are determined respectively, and the node variables of the grid-side main body are determined based on the line capacity and voltage safety boundary.

[0032] Based on the power surplus marker, load gap marker, and balance marker, establish the power regulation correspondence for the nodal variables of the source-side main body, load-side main body, and energy storage-side main body to obtain the power coupling edge;

[0033] Based on the line capacity and voltage safety boundaries, the nodal variables of the grid-side main body are respectively linked to the nodal variables of the source-side main body, load-side main body and energy storage-side main body to establish boundary constraint correspondence, thus obtaining the network constraint edges;

[0034] Calculate the absolute value of the difference between node variables for the same subject according to adjacent scheduling periods, and generate node variable increase / decrease flags and consecutive repetition counts based on the comparison result of the absolute value of the difference between node variables and 0. Write the absolute value of the difference between node variables, node variable increase / decrease flags and consecutive repetition counts into the time sequence connection edge.

[0035] The power coupling edges, network constraint edges, and time-series connection edges are reorganized according to the scheduling period index to obtain the main coupling relationship. Then, the connection relationship between node variables and factor nodes is established according to the main coupling relationship to obtain the dynamic factor graph.

[0036] Optionally, the improved Max-Sum algorithm is specifically as follows:

[0037] The node variables of the source-side main body, network-side main body, load-side main body and energy storage-side main body are extracted from the dynamic factor graph, and the power coupling edge, network constraint edge, time-series connection edge and constraint type label are extracted to obtain the message iteration base graph;

[0038] The first round of scheduling baseline value is determined based on historical scheduling results, and the feasible scheduling vector of the previous round is used as the scheduling baseline value of the current round in subsequent rounds.

[0039] The main adjustment margin is obtained by comparing the difference between the current round scheduling baseline value and the corresponding node variable;

[0040] Based on the power surplus marker, load gap marker, and balance marker, candidate adjustment values ​​for nodal variables of the source-side main body, grid-side main body, load-side main body, and energy storage-side main body are calculated to obtain the main body candidate value set;

[0041] Write the main candidate value set into the power coupling edge, and calculate the absolute value of the difference under the same candidate combination to obtain the local power deviation;

[0042] The candidate adjustment values ​​in the main candidate value set are compared with the boundary values ​​in the contraction constraint segment to obtain the constraint boundary distance. The order of the constraint boundary distance in the corresponding contraction constraint segment is determined as the constraint proximity.

[0043] Based on power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints, constraint evaluation items are configured for the factor nodes in the message iteration base graph, and the feedback message parameters are written into the corresponding constraint evaluation items to obtain residual correction factor nodes.

[0044] Based on the main adjustment margin, local power deviation and constraint proximity, a node-side composite message is generated to be sent from the node variables to the adjacent residual correction factor nodes. The absolute value of the difference between the node variables in the time-series connection edge, the node variable increase / decrease flag and the number of consecutive repetitions are written into the node-side composite message.

[0045] Input the node-side composite message into the residual correction factor node, enumerate the main candidate value set of the adjacent node variable according to the power coupling edge, network constraint edge and time sequence connection edge, and filter the factor-side composite message sent by the residual correction factor node to the target node variable according to the constraint evaluation term.

[0046] Collect factor-side composite messages received from the same node variable and sort the candidate adjustment values;

[0047] The candidate adjustment value at the top of the sort is determined as the scheduling value of the corresponding node variable in this round;

[0048] The scheduling values ​​for this round are combined in the order of node variables of the source-side subject, grid-side subject, load-side subject and energy storage subject to obtain the candidate scheduling vector. The composite message of this round is compared with the composite message of the previous round by comparing the same candidate values ​​on the same side to obtain the composite message change.

[0049] Optionally, inputting the candidate scheduling vector into the Dykstra projection algorithm specifically involves:

[0050] Extract power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints from the constraint set according to constraint type, and generate a projected constraint sequence according to the scheduling period index;

[0051] Use the candidate scheduling vector as the current projection vector, and set the residual compensation amount for each constraint type in the projection constraint sequence;

[0052] The current projection vector is superimposed with the residual compensation amount corresponding to the current constraint type to obtain the current projection input vector, and then projected onto the constraint segment corresponding to the current constraint type to obtain the current projection output vector;

[0053] Update the residual compensation amount corresponding to the current constraint type based on the difference between the current projection input vector and the current projection output vector, and pass the current projection output vector to the next constraint type;

[0054] Repeatedly execute alternating projection. When the output vector of the last projection in two adjacent rounds no longer changes, determine the output vector of the last projection as the feasible scheduling vector.

[0055] The feasible scheduling vector and the candidate scheduling vector are compared by performing the same position difference calculation according to the node variable order to obtain the vector deviation record;

[0056] The deviations corresponding to the source-side main body, load-side main body, and energy storage-side main body are extracted from the vector deviation record and aggregated according to the node variable relationship corresponding to the power balance constraint to obtain the power balance residual.

[0057] The deviations corresponding to line capacity, voltage safety boundary, and energy storage state of charge in the grid-side main body and the energy storage side main body are extracted from the vector deviation record to obtain the line capacity residual, voltage safety residual, and energy storage boundary residual, respectively.

[0058] The power balance residual, line capacity residual, voltage safety residual, and energy storage boundary residual are combined according to the constraint type to obtain the classified projection residual.

[0059] Optionally, the parameters for obtaining the feedback message are specifically:

[0060] Power balance residuals, line capacity residuals, voltage safety residuals, and energy storage boundary residuals are extracted from the classified projection residuals according to the constraint type label, and residual mapping records are established according to the scheduling period index.

[0061] Extract the residuals of the same constraint type across all scheduling periods based on the residual mapping records, and determine the residual baseline value;

[0062] The residual benchmark value is calculated with the power balance residual, line capacity residual, voltage safety residual and energy storage boundary residual respectively to obtain the factor node penalty weight increment, adjacent subject edge weight attenuation, grid-side node voltage constraint weight increment and energy storage node message damping.

[0063] The feedback message parameters are obtained by combining the incremental penalty weight of factor nodes, the attenuation of adjacent subject edge weights, the incremental voltage constraint weight of grid-side nodes, and the message damping of energy storage nodes according to the constraint type.

[0064] Optionally, the step of using the feasible scheduling vector as the scheduling benchmark for the next round and re-executing the improved Max-Sum algorithm iteration based on the feedback message parameters specifically involves:

[0065] Write the current round's feasible scheduling vector into the next round's scheduling baseline according to the node variable order, and extract the factor node penalty weight increment, adjacent subject edge weight attenuation, grid-side node voltage constraint weight increment, and energy storage node message damping from the feedback message parameters to obtain the current round's feedback write record.

[0066] Based on the current round of feedback write records, the message iteration base graph is corrected by feedback to obtain the feedback-corrected message iteration base graph;

[0067] Based on the next round of scheduling benchmark and the feedback-corrected message iteration base map, the improved Max-Sum algorithm iteration is re-executed to obtain the next round of candidate scheduling vectors and the next round of composite message changes.

[0068] The candidate scheduling vector for the next round is input into the Dykstra projection algorithm, and alternating projection is performed between the constraint sets to obtain the feasible scheduling vector for the next round. The classification projection residual for the next round is calculated based on the deviation between the feasible scheduling vector and the candidate scheduling vector for the next round.

[0069] The change in the next round of composite messages is compared with the preset message change threshold, and the residuals of each constraint type in the next round of classification projection residuals are compared with the corresponding residual benchmark values ​​to obtain the convergence determination record.

[0070] When the convergence determination record does not meet the preset convergence conditions, the feedback message parameters are regenerated based on the classification projection residual of the next round, and the feasible scheduling vector of the next round is written into the scheduling benchmark of the next round.

[0071] When the convergence determination record meets the preset convergence conditions, the next round of feasible scheduling vector is determined as the last round of feasible scheduling vector, and converted into source-grid-load-storage control instructions according to the node variable order.

[0072] Optionally, a source-grid-load-storage control system based on multi-agent scheduling includes:

[0073] The operation status sequence establishment module is used to collect distributed power output, load demand, energy storage status of charge, line capacity, voltage safety boundary and historical scheduling results, and establish an operation status sequence according to the scheduling period.

[0074] The constraint set generation module is used to perform piecewise linearization and safety margin shrinkage on power balance constraints, line capacity constraints, voltage safety constraints and energy storage state of charge constraints according to the operating state sequence, and configure constraint type tags to obtain constraint sets;

[0075] The dynamic factor graph construction module is used to determine the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body and energy storage-side main body based on the operating state sequence, and to construct the dynamic factor graph.

[0076] An improved Max-Sum iteration module is used to write the adjustment margin, local power deviation and constraint proximity of each subject in the dynamic factor graph into a composite message, and perform message iteration to obtain candidate scheduling vectors.

[0077] The Dykstra projection processing module is used to input candidate scheduling vectors into the Dykstra projection algorithm, perform alternating projections between constraint sets to obtain feasible scheduling vectors, and calculate classification projection residuals.

[0078] The feedback message generation module is used to establish feedback message parameters based on the constraint type of the categorical projection residual.

[0079] The closed-loop scheduling output module is used to take the feasible scheduling vector as the basis for the next round of scheduling, re-execute the improved Max-Sum algorithm iteration according to the feedback message parameters, and convert the last round of feasible scheduling vector into source-grid-load-storage control instructions when the preset convergence conditions are met.

[0080] The beneficial effects of this invention are:

[0081] (1) By establishing an operating state sequence and performing piecewise linearization and safety margin contraction on power balance constraints, line capacity constraints, voltage safety constraints and energy storage charge state constraints, this invention realizes the unified expression of constraints of source-side subjects, grid-side subjects, load-side subjects and energy storage subjects under the same scheduling period index, so that the multi-subject scheduling process can form a correspondence between candidate scheduling vectors and feasible scheduling vectors within a clear constraint set.

[0082] (2) This invention inputs candidate scheduling vectors into the Dykstra projection algorithm, performs alternating projections between constraint sets, and calculates power balance residuals, line capacity residuals, voltage safety residuals and energy storage boundary residuals based on the deviation between feasible scheduling vectors and candidate scheduling vectors. This enables the classification and extraction of constraint deviation states, thereby enabling the source-grid-load-storage control process to generate corresponding feedback message parameters for different constraint types.

[0083] (3) The improved Max-Sum algorithm in this invention, in the multi-subject source-grid-load-storage scheduling scenario, writes the subject adjustment margin, local power deviation, constraint proximity degree and feedback message parameters into the composite message iteration process, realizing the closed-loop connection between candidate scheduling vector generation, Dykstra projection correction, classification projection residual feedback and next round scheduling benchmark update. Attached Figure Description

[0084] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0085] Figure 1 This is a flowchart of a source-grid-load-storage control method based on multi-agent scheduling proposed in this invention.

[0086] Figure 2 This is a system structure block diagram of a source-grid-load-storage control system based on multi-agent scheduling proposed in this invention;

[0087] Figure 3 This is a flowchart of the improved Max-Sum algorithm for a source-grid-load-storage control method based on multi-agent scheduling proposed in this invention. Detailed Implementation

[0088] Example 1: Reference Figures 1-3 A source-grid-load-storage control method based on multi-agent scheduling, comprising:

[0089] Collect data on distributed power generation output, load demand, energy storage status of charge, line capacity, voltage safety boundary, and historical scheduling results, and establish an operational status sequence according to the scheduling period.

[0090] Based on the operating state sequence, piecewise linearization and safety margin shrinkage are performed on the power balance constraint, line capacity constraint, voltage safety constraint and energy storage state of charge constraint respectively, and constraint type labels are configured to obtain the constraint set;

[0091] Based on the operating status sequence, determine the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body and energy storage-side main body, and construct a dynamic factor diagram;

[0092] The adjustment margin, local power deviation, and constraint proximity of each subject in the dynamic factor graph are written into the composite message of the improved Max-Sum algorithm, and message iteration is performed in the dynamic factor graph according to the composite message to obtain the candidate scheduling vector.

[0093] The candidate scheduling vector is input into the Dykstra projection algorithm, and alternating projection is performed between the constraint sets to obtain the feasible scheduling vector. The classification projection residual is calculated based on the deviation between the feasible scheduling vector and the candidate scheduling vector. The classification projection residual includes power balance residual, line capacity residual, voltage safety residual and energy storage boundary residual.

[0094] According to the constraint type of the classification projection residual, the classification projection residual is converted into the factor node penalty weight increment, the adjacent subject edge weight attenuation, the grid-side node voltage constraint weight increment, and the energy storage node message damping amount to obtain the feedback message parameters. Among them, the power balance residual is converted into the factor node penalty weight increment, the line capacity residual is converted into the adjacent subject edge weight attenuation, the voltage safety residual is converted into the grid-side node voltage constraint weight increment, and the energy storage boundary residual is converted into the energy storage node message damping amount.

[0095] The feasible scheduling vector is used as the basis for the next round of scheduling. The improved Max-Sum algorithm is re-executed based on the feedback message parameters. When the composite message change and the classification projection residual both meet the preset convergence conditions, the feasible scheduling vector of the last round is converted into source-grid-load-storage control instructions.

[0096] In this embodiment, establishing the operating status sequence according to the scheduling period specifically means:

[0097] Establish a scheduling period index sequence based on the scheduling period number and execution time boundary in the historical scheduling results, and write the output of distributed power sources, load demand and energy storage charge status into the corresponding scheduling period index to obtain the main status index record;

[0098] Extract the distributed power output and load demand within the same scheduling period from the main status index record, calculate the net power deviation value, and generate output surplus markers, load gap markers, and balance markers based on the comparison result of the net power deviation value with 0. Then, mark them with the energy storage charge status within the same scheduling period to obtain the power storage coupling record for the time period. The net power deviation value is obtained by subtracting the load demand from the distributed power output.

[0099] In Example 1, the distributed power output value and load demand value within the same scheduling period are extracted from the main status index record according to the scheduling period index, and the net power deviation value is obtained by subtracting the load demand value from the distributed power output value. When the net power deviation value is greater than 0, a power surplus mark is written in the corresponding scheduling period; when the net power deviation value is less than 0, a load deficit mark is written in the corresponding scheduling period; when the net power deviation value is 0, a balance mark is written in the corresponding scheduling period. The written marks are associated with the energy storage charge state within the same scheduling period to form a time period power-energy storage coupling record.

[0100] The line capacity and voltage safety boundary are written into the time period power storage coupling record according to the same scheduling time period index, and the network boundary change record is generated according to the absolute value of the difference between the line capacity and the absolute value of the difference between the voltage safety boundary in adjacent scheduling time periods.

[0101] In Example 1, according to the scheduling period index sequence, the power storage coupling records of the previous and subsequent scheduling periods are extracted one by one; the line capacity of the previous scheduling period is subtracted from the line capacity of the subsequent scheduling period, and the absolute value of the difference is obtained to obtain the absolute value of the line capacity difference; the voltage safety boundary of the previous scheduling period is subtracted from the voltage safety boundary of the subsequent scheduling period, and the absolute value of the difference is obtained to obtain the absolute value of the voltage safety boundary difference; the scheduling period number, execution time boundary, absolute value of line capacity difference, and absolute value of voltage safety boundary difference are written into the same record item to obtain the network boundary change record;

[0102] The baseline value of boundary change is calculated based on the scheduling adjustment range of adjacent scheduling periods in the historical scheduling results, and compared and filtered with the network boundary change records to obtain the scheduling period of boundary change.

[0103] In Example 1, the source-side scheduling quantity, load-side scheduling quantity, and energy storage-side scheduling quantity of adjacent scheduling periods are extracted from the historical scheduling results according to the scheduling period index sequence. The absolute value of the change value of the later scheduling period relative to the previous scheduling period is calculated and accumulated to obtain the scheduling adjustment range. The scheduling adjustment ranges corresponding to all adjacent scheduling periods are sorted according to their numerical values, and the median value is taken as the boundary change benchmark value. The absolute values ​​of the line capacity difference and voltage safety boundary difference in the network boundary change record are compared with the boundary change benchmark value. When the absolute value of the line capacity difference is greater than the boundary change benchmark value, and the absolute value of the voltage safety boundary difference is greater than the boundary change benchmark value, the corresponding scheduling period is selected as the boundary change scheduling period.

[0104] The power storage coupling records of the previous and next scheduling periods of the boundary mutation scheduling period are called and corrected to obtain the corrected period status record. The correction process is to merge the boundary mutation scheduling period into the continuous period with the same corresponding mark when the corresponding marks of the records on both sides are consistent, and to retain the boundary mutation scheduling period and write the boundary mutation mark when the corresponding marks of the records on both sides are inconsistent.

[0105] The status records of the correction period are rearranged according to the scheduling period index sequence, and the correspondence between the corresponding scheduling period number, execution time boundary and historical scheduling result is retained to obtain the running status sequence.

[0106] In this embodiment, the step of performing piecewise linearization and safety margin contraction on power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints respectively is as follows:

[0107] Extract time-period power storage coupling records, network boundary change records, boundary mutation markers, line capacity, voltage safety boundary, and energy storage charge state from the operating state sequence, and establish constraint state records according to the scheduling time period index;

[0108] Based on the distributed power output, load demand, line capacity, voltage safety boundary and energy storage state of charge in the constraint state record, the initial power balance constraint segment, the initial line capacity constraint segment, the initial voltage safety constraint segment and the initial energy storage state of charge constraint segment are generated respectively.

[0109] In Example 1, the distributed power output, load demand, line capacity, voltage safety boundary, and energy storage state of charge for the same scheduling period are extracted from the constraint state record according to the scheduling period index. The net power deviation value is obtained by subtracting the load demand from the distributed power output and written into the corresponding scheduling period to generate an initial power balance constraint segment. The line capacity is used as the upper limit boundary of line transmission for the corresponding scheduling period to generate an initial line capacity constraint segment. The voltage safety boundary is used as the voltage deviation limit boundary for the corresponding scheduling period to generate an initial voltage safety constraint segment. The energy storage state of charge is used as the energy storage state boundary input value for the corresponding scheduling period to generate an initial energy storage state of charge constraint segment.

[0110] The absolute values ​​of the line capacity difference and voltage safety boundary difference in the network boundary change record are compared with the boundary change benchmark value, and the line capacity contraction level and voltage safety boundary contraction level are determined based on the comparison results.

[0111] In Example 1, the absolute value of the line capacity difference in the network boundary change record is compared with the boundary change reference value. When the absolute value of the line capacity difference is not greater than the boundary change reference value, it is recorded as the 0th level of line capacity contraction; when the absolute value of the line capacity difference is greater than the boundary change reference value but not greater than twice the boundary change reference value, it is recorded as the 1st level of line capacity contraction; when the absolute value of the line capacity difference is greater than twice the boundary change reference value, it is recorded as the 2nd level of line capacity contraction. The absolute values ​​of the voltage safety boundary difference are compared according to the same classification rule to obtain the 0th, 1st, and 2nd levels of voltage safety boundary contraction.

[0112] Based on the boundary mutation marker, the energy storage charge status of the corresponding scheduling period is invoked, and it is jointly judged with the output surplus marker, load gap marker and balance marker to determine the energy storage charge status contraction level.

[0113] In Example 1, the scheduling period marked with a boundary mutation is located in the operating state sequence, and the energy storage state of charge, output surplus marker, load gap marker, and balance marker of the corresponding scheduling period are invoked. The upper limit and lower limit values ​​of the energy storage state of charge are extracted from the initial energy storage state of charge constraint segment. The upper limit and lower limit values ​​are summed and divided by 2 to obtain the median value of the energy storage state of charge. When the output surplus marker is established and the energy storage state of charge is greater than the median value, it is recorded as the first contraction level. When the load gap marker is established and the energy storage state of charge is less than the median value, it is recorded as the first contraction level. When the balance marker is established, it is recorded as the 0th contraction level. When the energy storage state of charge reaches the upper limit or the lower limit value, it is recorded as the second contraction level.

[0114] According to the line capacity contraction level, voltage safety boundary contraction level, and energy storage state of charge contraction level, the boundary values ​​of the initial line capacity constraint segment, the initial voltage safety constraint segment, and the initial energy storage state of charge constraint segment are reduced in stages to obtain the contraction constraint segment.

[0115] In Example 1, line capacity boundary values ​​are extracted from the initial line capacity constraint segment. At line capacity contraction level 0, the line capacity boundary values ​​remain unchanged. At line capacity contraction level 1, 5% of the line capacity boundary values ​​are subtracted from the line capacity boundary values. At line capacity contraction level 2, 10% of the line capacity boundary values ​​are subtracted from the line capacity boundary values. Voltage safety boundary values ​​are extracted from the initial voltage safety constraint segment, and the contracted voltage is generated according to the following rules: no subtraction at contraction level 0, 5% subtraction at contraction level 1, and 10% subtraction at contraction level 2. Safety boundary values; extract the upper limit and lower limit of the energy storage state of charge from the initial energy storage state of charge constraint segment, subtract the lower limit from the upper limit to obtain the state of charge boundary span, divide the state of charge boundary span by 10 to obtain the state of charge deduction step size; the 0th contraction level keeps the upper and lower limits unchanged, the 1st contraction level lowers the upper limit by 1 state of charge deduction step size and raises the lower limit by 1 state of charge deduction step size, the 2nd contraction level lowers the upper limit by 2 state of charge deduction step sizes and raises the lower limit by 2 state of charge deduction step sizes to obtain the contraction constraint segment;

[0116] The initial power balance constraint fragment and the contraction constraint fragment are reorganized according to the scheduling period index, and the constraint type labels corresponding to the power balance constraint, line capacity constraint, voltage safety constraint and energy storage state of charge constraint are configured respectively to obtain the constraint set.

[0117] In this embodiment, the step of determining the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body, and energy storage-side main body based on the operating state sequence, and constructing a dynamic factor graph, specifically involves:

[0118] The main mapping record is established based on the operating status sequence. The main mapping record is established by extracting distributed power output, load demand, energy storage charge status, line capacity, voltage safety boundary, output surplus marker, load gap marker and balance marker from the operating status sequence, and establishing the main mapping record according to the scheduling period index.

[0119] Based on the distributed power output, load demand, and energy storage charge status in the main mapping record, the node variables of the source-side main body, load-side main body, and energy storage-side main body are determined respectively, and the node variables of the grid-side main body are determined based on the line capacity and voltage safety boundary.

[0120] In Example 1, the distributed power generation output, load demand, energy storage status of charge, line capacity, and voltage safety boundary for the same scheduling period are extracted from the main mapping record according to the scheduling period index. The distributed power generation output value is written into the source-side main body, and the corresponding scheduling period index is used as the node variable index of the source-side main body. The load demand value is written into the load-side main body, and the corresponding scheduling period index is used as the node variable index of the load-side main body. The energy storage status of charge value is written into the energy storage-side main body, and the corresponding scheduling period index is used as the node variable index of the energy storage-side main body. The line capacity value and voltage safety boundary value are combined according to the same scheduling period index and written into the grid-side main body to obtain the node variables of the grid-side main body.

[0121] Based on the power surplus marker, load gap marker, and balance marker, establish the power regulation correspondence for the nodal variables of the source-side main body, load-side main body, and energy storage-side main body to obtain the power coupling edge;

[0122] In Example 1, according to the scheduling period index, the output surplus marker, load gap marker, balance marker, source-side main node variables, load-side main node variables, and energy storage-side main node variables for the current scheduling period are extracted from the main mapping record. When the output surplus marker is established, the source-side main node variables are used as the regulation output end, and the energy storage-side main node variables are used as the surplus absorption end, establishing a power coupling edge from the source-side main body to the energy storage-side main body. When the load gap marker is established, the energy storage-side main node variables are used as the regulation output end, and the load-side main node variables are used as the gap compensation end, establishing a power coupling edge from the energy storage-side main body to the load-side main body. When the balance marker is established, the source-side main node variables, load-side main node variables, and energy storage-side main node variables are written into the same balance maintenance edge to obtain the power coupling edge.

[0123] Based on the line capacity and voltage safety boundaries, the nodal variables of the grid-side main body are respectively linked to the nodal variables of the source-side main body, load-side main body and energy storage-side main body to establish boundary constraint correspondence, thus obtaining the network constraint edges;

[0124] In Example 1, according to the scheduling period index, the network-side main node variables, source-side main node variables, load-side main node variables, and energy storage-side main node variables for the same scheduling period are extracted from the main mapping records. Line capacity values ​​and voltage safety boundary values ​​are separated from the network-side main node variables. The line capacity values ​​are paired with the source-side, load-side, and energy storage-side main node variables for the same scheduling period to obtain line capacity boundary pairing records. The voltage safety boundary values ​​are paired with the source-side, load-side, and energy storage-side main node variables for the same scheduling period to obtain voltage safety boundary pairing records. The line capacity boundary pairing records and voltage safety boundary pairing records are merged according to the corresponding main node variables to obtain network constraint edges.

[0125] Calculate the absolute value of the difference between node variables for the same subject according to adjacent scheduling periods, and generate node variable increase / decrease flags and consecutive repetition counts based on the comparison result of the absolute value of the difference between node variables and 0. Write the absolute value of the difference between node variables, node variable increase / decrease flags and consecutive repetition counts into the time sequence connection edge.

[0126] The power coupling edges, network constraint edges, and time-series connection edges are reorganized according to the scheduling period index to obtain the main coupling relationship. Then, the connection relationship between node variables and factor nodes is established according to the main coupling relationship to obtain the dynamic factor graph.

[0127] In this embodiment, the improved Max-Sum algorithm is specifically as follows:

[0128] The node variables of the source-side main body, network-side main body, load-side main body and energy storage-side main body are extracted from the dynamic factor graph, and the power coupling edge, network constraint edge, time-series connection edge and constraint type label are extracted to obtain the message iteration base graph;

[0129] The first round of scheduling baseline value is determined based on historical scheduling results, and the feasible scheduling vector of the previous round is used as the scheduling baseline value of the current round in subsequent rounds.

[0130] In Example 1, the source-side main scheduling value, grid-side main scheduling value, load-side main scheduling value, and energy storage-side main scheduling value for the most recently completed scheduling period are extracted from historical scheduling results according to the scheduling period index. The source-side main scheduling value is written into the reference bit corresponding to the source-side main node variable, the grid-side main scheduling value is written into the reference bit corresponding to the grid-side main node variable, the load-side main scheduling value is written into the reference bit corresponding to the load-side main node variable, and the energy storage-side main scheduling value is written into the reference bit corresponding to the energy storage-side main node variable. The reference bit values ​​are combined according to the node variable order of the source-side main, grid-side main, load-side main, and energy storage-side main to obtain the first round of scheduling reference value.

[0131] The main adjustment margin is obtained by comparing the difference between the current round scheduling baseline value and the corresponding node variable;

[0132] In Example 1, the current round scheduling baseline value is split into baseline values ​​for each subject according to the node variables of the source-side subject, grid-side subject, load-side subject, and energy storage subject. The node variable values ​​of the source-side subject, grid-side subject, load-side subject, and energy storage subject are extracted from the message iteration base graph according to the same scheduling period index. The node variable values ​​of the same subject are subtracted from the baseline values ​​of the same subject, and the absolute value of the difference is taken to obtain the single-subject adjustment margin. The single-subject adjustment margin is written into the same margin record according to the subject identifier and scheduling period index to obtain the subject adjustment margin.

[0133] Based on the power surplus marker, load gap marker, and balance marker, calculate the candidate adjustment values ​​of the nodal variables of the source-side main body, grid-side main body, load-side main body, and energy storage-side main body, and delete the candidate adjustment values ​​that are inconsistent with the corresponding marker direction to obtain the main body candidate value set;

[0134] In Example 1, the following parameters are extracted according to the scheduling period index: surplus output marker, load gap marker, balance marker, variable values ​​of each main node, and main regulation margin. When the surplus output marker is established, the corresponding main regulation margin is subtracted from the variable values ​​of the source-side main nodes to obtain the source-side candidate regulation values. The corresponding main regulation margin is added to the variable values ​​of the energy storage-side main nodes to obtain the energy storage-side candidate regulation values. The node variable values ​​of the load-side and grid-side main nodes are kept unchanged to obtain the load-side and grid-side candidate regulation values. When the load gap marker is established, the corresponding main regulation margin is added to the variable values ​​of the source-side main nodes to obtain the source-side candidate regulation values. The corresponding main regulation margin is subtracted from the variable values ​​of the energy storage-side main nodes to obtain the energy storage-side candidate regulation values. The node variable values ​​of the load-side and grid-side main nodes are kept unchanged to obtain the load-side and grid-side candidate regulation values. When the balance marker is established, the variable values ​​of each main node are kept unchanged to obtain the candidate regulation values ​​of each main node.

[0135] The source-side main candidate regulation values, grid-side main candidate regulation values, load-side main candidate regulation values, and energy storage-side main candidate regulation values ​​in the main candidate value set are written into the power coupling edge, and the absolute value of the difference between the source-side main candidate regulation values, the energy storage-side main candidate regulation values, and the load-side main candidate regulation values ​​under the same candidate combination is calculated to obtain the local power deviation.

[0136] The candidate adjustment values ​​in the main candidate value set are compared with the boundary values ​​in the contraction constraint segment by difference, and the constraint boundary distance is obtained by sorting the absolute values ​​of the differences from small to large. The order of the constraint boundary distance in the corresponding contraction constraint segment is determined as the constraint proximity.

[0137] Based on power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints, constraint evaluation items are configured for the factor nodes in the message iteration base graph, and the feedback message parameters are written into the corresponding constraint evaluation items to obtain residual correction factor nodes.

[0138] In Example 1, power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints are extracted from the constraint set according to constraint type. The corresponding factor nodes in the message iteration base graph are located according to the scheduling period index. The absolute values ​​of the differences between the source-side candidate adjustment values, the energy storage-side candidate adjustment values, and the load-side candidate adjustment values ​​are configured as power balance constraint evaluation items. The absolute values ​​of the differences between the grid-side candidate adjustment values ​​and the line capacity boundary values ​​are configured as line capacity constraint evaluation items. The absolute values ​​of the differences between the grid-side candidate adjustment values ​​and the voltage safety boundary values ​​are configured as voltage safety constraint evaluation items. The absolute values ​​of the differences between the energy storage-side candidate adjustment values ​​and the energy storage state of charge boundary values ​​are configured as energy storage state of charge constraint evaluation items.

[0139] Based on the main adjustment margin, local power deviation and constraint proximity, a node-side composite message is generated to be sent from the node variables to the adjacent residual correction factor nodes. The absolute value of the difference between the node variables in the time-series connection edge, the node variable increase / decrease flag and the number of consecutive repetitions are written into the node-side composite message.

[0140] In Example 1, the subject adjustment margin, local power deviation, constraint proximity, and subject candidate value set are extracted according to the scheduling period index and subject identifier. The same subject candidate adjustment value is bound to the same subject adjustment margin to generate subject candidate message items. The local power deviation is generated into a first evaluation rank according to the numerical value from small to large. The constraint proximity is generated into a second evaluation rank according to the rank from low to high. The subject adjustment margin is generated into a third evaluation rank according to the numerical value from large to small. The subject candidate message items, the first evaluation rank, the second evaluation rank, and the third evaluation rank are written into adjacent residual correction factor nodes according to the same subject identifier to form a node-side composite message.

[0141] Input the node-side composite message into the residual correction factor node, enumerate the main candidate value set of the adjacent node variable according to the power coupling edge, network constraint edge and time sequence connection edge, and filter the factor-side composite message sent by the residual correction factor node to the target node variable according to the constraint evaluation term.

[0142] In Example 1, from the node-side composite message received from the residual correction factor node, candidate value combinations for the main body are enumerated for adjacent node variables other than the target node variable; for each candidate value combination, the corresponding constraint evaluation item is called to calculate the power balance evaluation value, line capacity evaluation value, voltage safety evaluation value, and energy storage state of charge evaluation value respectively; the evaluation values ​​are sorted in ascending order of value, and the first candidate value combination in the sort is retained; the candidate adjustment value, main candidate message item, and evaluation position corresponding to the target node variable are extracted from the first candidate value combination in the sort, and written into the incoming edge message record of the target node variable to form the factor-side composite message;

[0143] The factor-side composite messages received by the same node variable are collected and sorted according to the order of local power deviation from small to large, constraint proximity from low to high, and main adjustment margin from large to small.

[0144] The candidate adjustment value at the top of the sort is determined as the scheduling value of the corresponding node variable in this round;

[0145] The scheduling values ​​for this round are combined in the order of node variables of the source-side subject, grid-side subject, load-side subject and energy storage subject to obtain the candidate scheduling vector. The composite message of this round is compared with the composite message of the previous round by comparing the same candidate values ​​on the same side to obtain the composite message change.

[0146] In this embodiment, the step of inputting the candidate scheduling vector into the Dykstra projection algorithm specifically involves:

[0147] Extract power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints from the constraint set according to constraint type, and generate a projected constraint sequence according to the scheduling period index;

[0148] Use the candidate scheduling vector as the current projection vector, and set the residual compensation amount for each constraint type in the projection constraint sequence;

[0149] In Example 1, power balance residual compensation, line capacity residual compensation, voltage safety residual compensation, and energy storage boundary residual compensation are established according to the power balance constraint, line capacity constraint, voltage safety constraint, and energy storage state of charge constraint in the projection constraint sequence. Before the first round of alternating projection, the values ​​of the four types of residual compensation are all set to 0, and compensation vectors with the same length as the candidate scheduling vectors are configured according to the node variable order. After each round of projection, the difference between the current projection input vector and the current projection output vector is written into the residual compensation of the current constraint type.

[0150] The current projection vector is superimposed with the residual compensation amount corresponding to the current constraint type to obtain the current projection input vector, and then projected onto the constraint segment corresponding to the current constraint type to obtain the current projection output vector;

[0151] Update the residual compensation amount corresponding to the current constraint type based on the difference between the current projection input vector and the current projection output vector, and pass the current projection output vector to the next constraint type;

[0152] Repeatedly execute alternating projection. When the output vector of the last projection in two adjacent rounds no longer changes, determine the output vector of the last projection as the feasible scheduling vector.

[0153] The feasible scheduling vector and the candidate scheduling vector are compared by performing the same position difference calculation according to the node variable order to obtain the vector deviation record;

[0154] In Example 1, both the feasible scheduling vector and the candidate scheduling vector are expanded into sequences of the same length according to the node variables of the source-side entity, grid-side entity, load-side entity, and energy storage-side entity. Under the same scheduling time period index and the same entity identifier, the value of the corresponding position in the feasible scheduling vector is subtracted from the value of the corresponding position in the candidate scheduling vector to obtain the single position deviation. The single position deviation, scheduling time period index, entity identifier, and corresponding node variable name are written into the same record item. After traversing all positions, a vector deviation record is formed.

[0155] The deviations corresponding to the source-side main body, load-side main body, and energy storage-side main body are extracted from the vector deviation record and aggregated according to the node variable relationship corresponding to the power balance constraint to obtain the power balance residual.

[0156] In Example 1, the source-side main body deviation, load-side main body deviation, and energy storage-side main body deviation are filtered from the vector deviation record according to the scheduling period index; the power coupling edge corresponding to the power balance constraint is called to determine that the source-side main body and energy storage-side main body are supply-side node variables and the load-side main body is demand-side node variable; the source-side main body deviation and the energy storage-side main body deviation are added to obtain the total supply-side deviation; the load-side main body deviation is subtracted from the total supply-side deviation, and the absolute value of the difference is taken to obtain the power balance residual for the current scheduling period.

[0157] The deviations corresponding to line capacity, voltage safety boundary, and energy storage state of charge in the grid-side main body and the energy storage side main body are extracted from the vector deviation record to obtain the line capacity residual, voltage safety residual, and energy storage boundary residual, respectively.

[0158] In Example 1, the deviations of the grid-side main line capacity, the grid-side main voltage safety boundary, and the energy storage main energy storage state of charge are filtered from the vector deviation records according to the scheduling period index and node variable name. The absolute value of the grid-side main line capacity deviation is taken and written into the line capacity residual record of the corresponding scheduling period. The absolute value of the grid-side main voltage safety boundary deviation is taken and written into the voltage safety residual record of the corresponding scheduling period. The absolute value of the energy storage main energy storage state of charge deviation is taken and written into the energy storage boundary residual record of the corresponding scheduling period. The line capacity residual record, voltage safety residual record, and energy storage boundary residual record are summarized according to the scheduling period index to obtain the line capacity residual, voltage safety residual, and energy storage boundary residual, respectively.

[0159] The power balance residual, line capacity residual, voltage safety residual, and energy storage boundary residual are combined according to the constraint type to obtain the classified projection residual.

[0160] In this embodiment, obtaining the feedback message parameters specifically refers to:

[0161] Power balance residuals, line capacity residuals, voltage safety residuals, and energy storage boundary residuals are extracted from the classified projection residuals according to the constraint type label, and residual mapping records are established according to the scheduling period index.

[0162] Extract the residuals of the same constraint type in all scheduling periods according to the residual mapping records, determine the residual benchmark value, arrange the residuals of the same constraint type in ascending order of value, take the median of the arrangement as the residual benchmark value, and take the smallest non-zero residual as the residual benchmark value when the residual benchmark value is 0 and there are non-zero residuals.

[0163] The residual benchmark value is calculated with the power balance residual, line capacity residual, voltage safety residual and energy storage boundary residual respectively to obtain the factor node penalty weight increment, adjacent subject edge weight attenuation, grid-side node voltage constraint weight increment and energy storage node message damping.

[0164] In Example 1, the power balance residual is divided by the corresponding residual benchmark value to obtain the power balance residual multiple, and the current factor node penalty weight corresponding to the power balance constraint is multiplied by the power balance residual multiple to obtain the factor node penalty weight increment; the line capacity residual is divided by the corresponding residual benchmark value to obtain the line capacity residual multiple, and the current adjacent subject edge weight corresponding to the network constraint edge is multiplied by the line capacity residual multiple to obtain the edge weight to be attenuated, and the edge weight to be attenuated is restricted within the current adjacent subject edge weight to obtain the adjacent subject edge weight attenuation; the voltage safety residual is divided by the corresponding residual benchmark value to obtain the voltage safety residual multiple, and the current voltage constraint weight corresponding to the node variable of the network-side subject is multiplied by the voltage safety residual multiple to obtain the grid-side node voltage constraint weight increment; the energy storage boundary residual is divided by the corresponding residual benchmark value to obtain the energy storage boundary residual multiple, and the current node-side composite message amplitude corresponding to the node variable of the energy storage-side subject is multiplied by the energy storage boundary residual multiple to obtain the energy storage node message damping amount.

[0165] The feedback message parameters are obtained by combining the incremental penalty weight of factor nodes, the attenuation of adjacent subject edge weights, the incremental voltage constraint weight of grid-side nodes, and the message damping of energy storage nodes according to the constraint type.

[0166] In this embodiment, the step of using the feasible scheduling vector as the scheduling benchmark for the next round and re-executing the improved Max-Sum algorithm iteration based on the feedback message parameters specifically involves:

[0167] Write the current round's feasible scheduling vector into the next round's scheduling baseline according to the node variable order, and extract the factor node penalty weight increment, adjacent subject edge weight attenuation, grid-side node voltage constraint weight increment, and energy storage node message damping from the feedback message parameters to obtain the current round's feedback write record.

[0168] Based on the current round of feedback write records, the message iteration base graph is corrected by feedback to obtain the feedback-corrected message iteration base graph;

[0169] In Example 1, the following steps are taken: First, the incremental penalty weight of factor nodes, the attenuation of adjacent subject edge weights, the incremental voltage constraint weight of grid-side nodes, and the message damping of energy storage nodes are extracted from the current round of feedback write records. Second, the residual correction factor node corresponding to the power balance constraint is located in the message iteration base graph. The incremental penalty weight of the current factor node is added to the incremental penalty weight of the factor node to obtain the corrected power balance constraint evaluation term. Third, adjacent subject edges are located in the network constraint edges. The attenuation of the adjacent subject edge weight is subtracted from the current adjacent subject edge weight. If the subtraction result is less than 0, it is recorded as 0 to obtain the corrected network constraint edge. Fourth, the voltage safety constraint evaluation term is located in the residual correction factor node. The incremental voltage constraint weight of the current voltage constraint is added to the incremental voltage constraint weight of the grid-side node to obtain the corrected voltage safety constraint evaluation term. Fifth, the message damping of energy storage nodes is subtracted from the composite message of the node side adjacent to the energy storage subject node variable. If the subtraction result is less than 0, it is recorded as 0 to obtain the corrected node side composite message. Sixth, the corrected power balance constraint evaluation term, the corrected network constraint edge, the corrected voltage safety constraint evaluation term, and the corrected node side composite message are written back to the message iteration base graph to obtain the message iteration base graph after feedback correction.

[0170] Based on the next round of scheduling benchmark and the feedback-corrected message iteration base map, the improved Max-Sum algorithm iteration is re-executed to obtain the next round of candidate scheduling vectors and the next round of composite message changes.

[0171] The candidate scheduling vector for the next round is input into the Dykstra projection algorithm, and alternating projection is performed between the constraint sets to obtain the feasible scheduling vector for the next round. The classification projection residual for the next round is calculated based on the deviation between the feasible scheduling vector and the candidate scheduling vector for the next round.

[0172] The change amount of the next round of composite messages is compared with the preset message change threshold, and the residuals of each constraint type in the next round of classification projection residuals are compared with the corresponding residual benchmark values ​​to obtain the convergence judgment record. Among them, 1% of the change amount of the first round of composite messages is set as the preset message change threshold, and when the change amount of the first round of composite messages is 0, 0.001 is set as the preset message change threshold.

[0173] When the convergence determination record does not meet the preset convergence condition, the feedback message parameters are regenerated based on the next round of classification projection residuals, and the next round of feasible scheduling vectors are written into the next round of scheduling benchmarks. The preset convergence condition means that the change in the next round of composite messages is not greater than the preset message change threshold, and the residuals of each constraint type in the next round of classification projection residuals are not greater than the corresponding residual benchmark values.

[0174] When the convergence determination record meets the preset convergence conditions, the next round of feasible scheduling vector is determined as the last round of feasible scheduling vector, and converted into source-grid-load-storage control instructions according to the node variable order.

[0175] In Example 1, when the convergence determination record meets the preset convergence conditions, the next round of feasible scheduling vectors is marked as the last round of feasible scheduling vectors. The last round of feasible scheduling vectors is split according to the node variables of the source-side subject, grid-side subject, load-side subject, and energy storage subject to obtain the source-side scheduling value, grid-side scheduling value, load-side scheduling value, and energy storage-side scheduling value. The source-side scheduling value is written into the distributed power generation output control field, the grid-side scheduling value is written into the network boundary control field, the load-side scheduling value is written into the load regulation control field, and the energy storage-side scheduling value is written into the energy storage load status control field. The control fields are combined according to the scheduling period index and subject identifier to obtain the source-grid-load-storage control command.

[0176] In this embodiment, a source-grid-load-storage control system based on multi-agent scheduling includes:

[0177] The operation status sequence establishment module is used to collect distributed power output, load demand, energy storage status of charge, line capacity, voltage safety boundary and historical scheduling results, and establish an operation status sequence according to the scheduling period.

[0178] The constraint set generation module is used to perform piecewise linearization and safety margin shrinkage on power balance constraints, line capacity constraints, voltage safety constraints and energy storage state of charge constraints according to the operating state sequence, and configure constraint type tags to obtain constraint sets;

[0179] The dynamic factor graph construction module is used to determine the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body and energy storage-side main body based on the operating state sequence, and to construct the dynamic factor graph.

[0180] An improved Max-Sum iteration module is used to write the adjustment margin, local power deviation and constraint proximity of each subject in the dynamic factor graph into a composite message, and perform message iteration to obtain candidate scheduling vectors.

[0181] The Dykstra projection processing module is used to input candidate scheduling vectors into the Dykstra projection algorithm, perform alternating projections between constraint sets to obtain feasible scheduling vectors, and calculate classification projection residuals.

[0182] The feedback message generation module is used to establish feedback message parameters based on the constraint type of the categorical projection residual.

[0183] The closed-loop scheduling output module is used to take the feasible scheduling vector as the basis for the next round of scheduling, re-execute the improved Max-Sum algorithm iteration according to the feedback message parameters, and convert the last round of feasible scheduling vector into source-grid-load-storage control instructions when the preset convergence conditions are met.

[0184] Example 2:

[0185] To verify the feasibility of this invention in practice, it was applied to a source-grid-load-storage coordinated control scenario, specifically an AC power distribution operation unit comprising distributed power sources, adjustable loads, and energy storage devices. In this scenario, the output of distributed power sources fluctuates, load demand rises and falls during continuous scheduling periods, and the state of charge of energy storage is affected by preceding charging and discharging behaviors and subject to boundary constraints. Existing multi-agent scheduling methods struggle to promptly feed back constraint deviations to the next round of message iteration when candidate scheduling results deviate from power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints, leading to discrepancies between candidate scheduling vectors and feasible scheduling vectors.

[0186] In this embodiment, the sampling period is set to 5 minutes, collecting distributed power output, load demand, energy storage state of charge, line capacity, voltage safety boundary, and historical scheduling results for 96 scheduling periods. The system establishes a scheduling period index sequence according to the scheduling period number and execution time boundary, and writes the distributed power output, load demand, and energy storage state of charge into the corresponding scheduling period index to obtain the main state index record. The system calculates the net power deviation value and writes the output surplus mark, load gap mark, and balance mark according to the comparison result of the net power deviation value with 0, and then matches them with the corresponding energy storage state of charge mark to obtain the time period power-energy storage coupling record. The system generates a network boundary change record based on the absolute value of the line capacity difference and the absolute value of the voltage safety boundary difference between adjacent scheduling periods, and calculates the boundary change benchmark value according to the scheduling adjustment amplitude in the historical scheduling results to filter scheduling periods with boundary abrupt changes. For scheduling periods with boundary abrupt changes, the system calls the time period power-energy storage coupling records of the previous and subsequent scheduling periods for correction processing to obtain the corrected time period state record, and rearranges it into an operating state sequence.

[0187] The system establishes constraint state records from the operating state sequence and generates initial power balance constraint segments, initial line capacity constraint segments, initial voltage safety constraint segments, and initial energy storage state of charge constraint segments. The system compares the absolute values ​​of the line capacity difference and voltage safety boundary difference with the boundary change benchmark values ​​to determine the line capacity contraction level and voltage safety boundary contraction level. Based on boundary abrupt change markers, energy storage state of charge, output surplus markers, load gap markers, and balance markers, the system determines the energy storage state of charge contraction level. The first contraction level corresponds to a 5% reduction in boundary values, and the second contraction level corresponds to a 10% reduction. The energy storage state of charge constraint segments are calculated by forming a deduction step size based on the boundary span of the upper and lower limits, resulting in contraction constraint segments. The system reassembles the initial power balance constraint segments and contraction constraint segments according to the scheduling period index and configures constraint type markers to obtain a constraint set.

[0188] The system establishes a main body mapping record based on the operating state sequence, determines the node variables of the source-side main body, load-side main body, and energy storage main body based on distributed power output, load demand, and energy storage charge status, and determines the node variables of the grid-side main body based on line capacity and voltage safety boundaries. The system establishes power coupling edges based on output surplus markers, load gap markers, and balance markers, establishes network constraint edges based on line capacity and voltage safety boundaries, and calculates the absolute value of the difference between node variables of the same main body according to adjacent scheduling periods, generates node variable increase / decrease markers and consecutive repetition counts, and writes them into the time-series connection edges to obtain the main body coupling relationship and dynamic factor graph.

[0189] In the improved Max-Sum algorithm execution process, the system extracts node variables, power coupling edges, network constraint edges, temporal connection edges, and constraint type labels for four types of subjects from the dynamic factor graph to obtain the message iteration base graph. The first round of scheduling baseline values ​​is determined by the scheduling values ​​of each subject in the most recently completed scheduling period from the historical scheduling results. Subsequent rounds call the feasible scheduling vector of the previous round as the current round's scheduling baseline value. The system compares the current round's scheduling baseline value with the corresponding node variables to obtain the subject's adjustment margin, and calculates candidate adjustment values ​​based on the output surplus label, load gap label, and balance label. Candidate adjustment values ​​with inconsistent directions are deleted to obtain the subject's candidate value set. The system writes the subject's candidate value set into the power coupling edge, calculates the local power deviation, and compares the candidate adjustment values ​​with the boundary values ​​in the contraction constraint segment to obtain the constraint proximity. The system generates node-side composite messages based on the main control margin, local power deviation, and constraint proximity. Factor-side composite messages are obtained by filtering the residual correction factor nodes. The scheduling values ​​for this round are determined in the order of local power deviation from small to large, constraint proximity from low to high, and main control margin from large to small. The candidate scheduling vector is obtained by combining these values.

[0190] After candidate scheduling vectors are generated, the system extracts power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints from the constraint set and generates a projection constraint sequence. The system uses the candidate scheduling vectors as the current projection vectors, sets residual compensation amounts for each constraint type, and projects the current projection vector onto the corresponding constraint segment after superimposing the current projection vector with the residual compensation amount corresponding to the current constraint type, obtaining the current projection output vector. The system updates the residual compensation amount based on the difference between the current projection input vector and the current projection output vector, and passes the current projection output vector to the next constraint type. When the last projection output vector in two adjacent rounds no longer changes, the last projection output vector is determined as the feasible scheduling vector. The system performs positional difference calculations on the feasible scheduling vector and the candidate scheduling vector according to the node variable order to obtain vector deviation records, and obtains the power balance residual, line capacity residual, voltage safety residual, and energy storage boundary residual respectively. These are then combined according to the constraint type label to form a categorized projection residual.

[0191] Table 1. Verification Results of Source-Grid-Load-Storage Dispatch Implementation

[0192] Scheduling period Distributed power output / kW Load demand / kW Energy storage state of charge / % Candidate scheduling deviation / kW Feasible dispatch deviation / kW Power balance residual / kW Line capacity residual / kW Voltage safety residual Energy storage boundary residual / % T01 186.4 204.7 58.2 15.6 4.8 3.1 1.6 0.012 1.4 T02 193.8 215.2 55.9 18.3 5.2 3.6 1.9 0.015 1.6 T03 221.5 208.6 62.4 14.1 3.9 2.7 1.3 0.010 1.1 T04 247.2 232.5 68.7 16.8 4.5 2.9 1.5 0.011 1.3 T05 205.6 226.9 51.3 19.7 5.6 3.8 2.1 0.017 1.8

[0193] As shown in Table 1, the candidate scheduling deviations for T01 to T05 were 15.6kW, 18.3kW, 14.1kW, 16.8kW, and 19.7kW, respectively. After alternating projection, the feasible scheduling deviations decreased to 4.8kW, 5.2kW, 3.9kW, 4.5kW, and 5.6kW, respectively. The power balance residuals were all below 4.0kW, the line capacity residuals were all below 2.2kW, the voltage safety residuals were all below 0.018, and the energy storage boundary residuals were all below 1.9%, indicating that the classification projection residuals can distinguish the deviation states of different constraint types. The operation log shows that in the first iteration, the composite message change was 0.184, and the maximum constraint type residual in the classification projection residuals was 5.9; in the sixth iteration, the composite message change decreased to 0.009, and the convergence judgment record met the preset convergence conditions.

[0194] The system extracts four types of residuals from the classification projection residuals and establishes residual mapping records. Residuals of the same constraint type are arranged in ascending order of value, and the median of the arrangement is used as the residual baseline value. The system also calculates the incremental weight of factor nodes' penalty weights, the attenuation of adjacent subject edge weights, the incremental weight of grid-side node voltage constraints, and the message damping of energy storage nodes, combining these into feedback message parameters. After the current round's feasible scheduling vector is written into the next round's scheduling baseline according to the node variable order, the system performs feedback correction on the message iteration baseline based on the current round's feedback writing record. If the convergence judgment record does not meet the preset convergence conditions, the system regenerates the feedback message parameters based on the next round's classification projection residuals and writes the next round's feasible scheduling vector into the subsequent round's scheduling baseline. If the preset convergence conditions are met, the system splits the last round's feasible scheduling vector according to the node variable order of the four subject types and writes it into the distributed power generation output control field, network boundary control field, load regulation control field, and energy storage state of charge control field, forming source-grid-load-storage control instructions. The results of this embodiment show that the present invention can form a closed-loop processing chain between the running state sequence, constraint set, dynamic factor graph, candidate scheduling vector, feasible scheduling vector, classification projection residual and feedback message parameters.

[0195] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A source-grid-load-storage control method based on multi-agent scheduling, characterized in that, include: Collect data on distributed power generation output, load demand, energy storage status of charge, line capacity, voltage safety boundary, and historical scheduling results, and establish an operating status sequence according to the scheduling period. Based on the operating state sequence, piecewise linearization and safety margin shrinkage are performed on the power balance constraint, line capacity constraint, voltage safety constraint and energy storage state of charge constraint respectively, and constraint type labels are configured to obtain the constraint set; Based on the operating status sequence, determine the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body and energy storage-side main body, and construct a dynamic factor diagram; The adjustment margin, local power deviation, and constraint proximity of each subject in the dynamic factor graph are written into the composite message of the improved Max-Sum algorithm, and message iteration is performed in the dynamic factor graph according to the composite message to obtain the candidate scheduling vector. The candidate scheduling vector is input into the Dykstra projection algorithm, which performs alternating projections between the constraint sets to obtain the feasible scheduling vector. The classification projection residual is then calculated based on the deviation between the feasible scheduling vector and the candidate scheduling vector. According to the constraint type of the classification projection residual, the classification projection residual is converted into the factor node penalty weight increment, the adjacent subject edge weight attenuation, the network-side node voltage constraint weight increment, and the energy storage node message damping, respectively, to obtain the feedback message parameters. The feasible scheduling vector is used as the basis for the next round of scheduling. The improved Max-Sum algorithm is re-executed based on the feedback message parameters. When the composite message change and the classification projection residual both meet the preset convergence conditions, the feasible scheduling vector of the last round is converted into source-grid-load-storage control instructions.

2. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The establishment of the operating status sequence according to the scheduling period is specifically as follows: Establish a scheduling period index sequence based on the scheduling period number and execution time boundary in the historical scheduling results, and write the output of distributed power sources, load demand and energy storage charge status into the corresponding scheduling period index to obtain the main status index record; Extract the distributed power output and load demand within the same scheduling period from the main status index record, calculate the net power deviation value, and generate output surplus markers, load gap markers, and balance markers based on the comparison result of the net power deviation value with 0. Then, mark them with the energy storage charge status within the same scheduling period to obtain the time period power-energy storage coupling record. The line capacity and voltage safety boundary are written into the time period power storage coupling record according to the same scheduling time period index, and the network boundary change record is generated according to the absolute value of the difference between the line capacity and the absolute value of the difference between the voltage safety boundary in adjacent scheduling time periods. The baseline value of boundary change is calculated based on the scheduling adjustment range of adjacent scheduling periods in the historical scheduling results, and compared and filtered with the network boundary change records to obtain the scheduling period of boundary change. The power storage coupling records of the previous and next scheduling periods of the boundary mutation scheduling period are called and corrected to obtain the corrected period state records. The status records of the correction period are rearranged according to the scheduling period index sequence, and the correspondence between the corresponding scheduling period number, execution time boundary and historical scheduling result is retained to obtain the running status sequence.

3. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The piecewise linearization and safety margin contraction are applied to the power balance constraint, line capacity constraint, voltage safety constraint, and energy storage state of charge constraint, respectively, as follows: Extract time-period power storage coupling records, network boundary change records, boundary mutation markers, line capacity, voltage safety boundary, and energy storage charge state from the operating state sequence, and establish constraint state records according to the scheduling time period index; Based on the constraint state record, generate the initial power balance constraint segment, the initial line capacity constraint segment, the initial voltage safety constraint segment, and the initial energy storage state of charge constraint segment respectively. The absolute values ​​of the line capacity difference and voltage safety boundary difference in the network boundary change record are compared with the boundary change benchmark value, and the line capacity contraction level and voltage safety boundary contraction level are determined based on the comparison results. Based on the boundary mutation marker, the energy storage charge status of the corresponding scheduling period is invoked, and it is jointly judged with the output surplus marker, load gap marker and balance marker to determine the energy storage charge status contraction level. According to the line capacity contraction level, voltage safety boundary contraction level, and energy storage state of charge contraction level, the boundary values ​​of the initial line capacity constraint segment, the initial voltage safety constraint segment, and the initial energy storage state of charge constraint segment are reduced in stages to obtain the contraction constraint segment. The initial power balance constraint fragment and the contraction constraint fragment are reorganized according to the scheduling period index, and the constraint type labels corresponding to the power balance constraint, line capacity constraint, voltage safety constraint and energy storage state of charge constraint are configured respectively to obtain the constraint set.

4. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The process involves determining the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body, and energy storage-side main body based on the operating state sequence, and constructing a dynamic factor graph, specifically as follows: Establish the main mapping record based on the operating status sequence; Based on the distributed power output, load demand, and energy storage charge status in the main mapping record, the node variables of the source-side main body, load-side main body, and energy storage-side main body are determined respectively, and the node variables of the grid-side main body are determined based on the line capacity and voltage safety boundary. Based on the power surplus marker, load gap marker, and balance marker, establish the power regulation correspondence for the nodal variables of the source-side main body, load-side main body, and energy storage-side main body to obtain the power coupling edge; Based on the line capacity and voltage safety boundaries, the nodal variables of the grid-side main body are respectively linked to the nodal variables of the source-side main body, load-side main body and energy storage-side main body to establish boundary constraint correspondence, thus obtaining the network constraint edges; Calculate the absolute value of the difference between node variables for the same subject according to adjacent scheduling periods, and generate node variable increase / decrease flags and consecutive repetition counts based on the comparison result of the absolute value of the difference between node variables and 0. Write the absolute value of the difference between node variables, node variable increase / decrease flags and consecutive repetition counts into the time sequence connection edge. The power coupling edges, network constraint edges, and time-series connection edges are reorganized according to the scheduling period index to obtain the main coupling relationship. Then, the connection relationship between node variables and factor nodes is established according to the main coupling relationship to obtain the dynamic factor graph.

5. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The improved Max-Sum algorithm is as follows: The node variables of the source-side main body, network-side main body, load-side main body and energy storage-side main body are extracted from the dynamic factor graph, and the power coupling edge, network constraint edge, time-series connection edge and constraint type label are extracted to obtain the message iteration base graph; The first round of scheduling baseline value is determined based on historical scheduling results, and the feasible scheduling vector of the previous round is used as the scheduling baseline value of the current round in subsequent rounds. The main adjustment margin is obtained by comparing the difference between the current round scheduling baseline value and the corresponding node variable; Based on the power surplus marker, load gap marker, and balance marker, candidate adjustment values ​​for nodal variables of the source-side main body, grid-side main body, load-side main body, and energy storage-side main body are calculated to obtain the main body candidate value set; Write the main candidate value set into the power coupling edge, and calculate the absolute value of the difference under the same candidate combination to obtain the local power deviation; The candidate adjustment values ​​in the main candidate value set are compared with the boundary values ​​in the contraction constraint segment to obtain the constraint boundary distance. The order of the constraint boundary distance in the corresponding contraction constraint segment is determined as the constraint proximity. Based on power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints, constraint evaluation items are configured for the factor nodes in the message iteration base graph, and the feedback message parameters are written into the corresponding constraint evaluation items to obtain residual correction factor nodes. Based on the main adjustment margin, local power deviation and constraint proximity, a node-side composite message is generated to be sent from the node variables to the adjacent residual correction factor nodes. The absolute value of the difference between the node variables in the time-series connection edge, the node variable increase / decrease flag and the number of consecutive repetitions are written into the node-side composite message. Input the node-side composite message into the residual correction factor node, enumerate the main candidate value set of the adjacent node variable according to the power coupling edge, network constraint edge and timing connection edge, and filter the factor-side composite message sent by the residual correction factor node to the target node variable according to the constraint evaluation term. Collect factor-side composite messages received from the same node variable and sort the candidate adjustment values; The candidate adjustment value at the top of the sort is determined as the scheduling value of the corresponding node variable in this round; The scheduling values ​​for this round are combined in the order of node variables of the source-side subject, grid-side subject, load-side subject and energy storage subject to obtain the candidate scheduling vector. The composite message of this round is compared with the composite message of the previous round by comparing the same candidate values ​​on the same side to obtain the composite message change.

6. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The step of inputting the candidate scheduling vector into the Dykstra projection algorithm is as follows: Extract power balance constraints, line capacity constraints, voltage safety constraints, and energy storage state of charge constraints from the constraint set according to constraint type, and generate a projected constraint sequence according to the scheduling period index; Use the candidate scheduling vector as the current projection vector, and set the residual compensation amount for each constraint type in the projection constraint sequence; The current projection vector is superimposed with the residual compensation amount corresponding to the current constraint type to obtain the current projection input vector, and then projected onto the constraint segment corresponding to the current constraint type to obtain the current projection output vector; Update the residual compensation amount corresponding to the current constraint type based on the difference between the current projection input vector and the current projection output vector, and pass the current projection output vector to the next constraint type; Repeatedly execute alternating projection. When the output vector of the last projection in two adjacent rounds no longer changes, determine the output vector of the last projection as the feasible scheduling vector. The feasible scheduling vector and the candidate scheduling vector are compared by performing the same position difference calculation according to the node variable order to obtain the vector deviation record; The deviations corresponding to the source-side main body, load-side main body, and energy storage-side main body are extracted from the vector deviation record and aggregated according to the node variable relationship corresponding to the power balance constraint to obtain the power balance residual. The deviations corresponding to line capacity, voltage safety boundary, and energy storage state of charge in the grid-side main body and the energy storage side main body are extracted from the vector deviation record to obtain the line capacity residual, voltage safety residual, and energy storage boundary residual, respectively. The power balance residual, line capacity residual, voltage safety residual, and energy storage boundary residual are combined according to the constraint type to obtain the classified projection residual.

7. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The obtained feedback message parameters are specifically as follows: Power balance residuals, line capacity residuals, voltage safety residuals, and energy storage boundary residuals are extracted from the classified projection residuals according to the constraint type label, and residual mapping records are established according to the scheduling period index. Extract the residuals of the same constraint type across all scheduling periods based on the residual mapping records, and determine the residual baseline value; The residual benchmark value is calculated with the power balance residual, line capacity residual, voltage safety residual and energy storage boundary residual respectively to obtain the factor node penalty weight increment, adjacent subject edge weight attenuation, grid-side node voltage constraint weight increment and energy storage node message damping. The feedback message parameters are obtained by combining the incremental penalty weight of factor nodes, the attenuation of adjacent subject edge weights, the incremental voltage constraint weight of grid-side nodes, and the message damping of energy storage nodes according to the constraint type.

8. The source-grid-load-storage control method based on multi-agent scheduling according to claim 1, characterized in that, The step of using the feasible scheduling vector as the scheduling benchmark for the next round and re-executing the improved Max-Sum algorithm iteration based on the feedback message parameters is as follows: Write the current round's feasible scheduling vector into the next round's scheduling baseline according to the node variable order, and extract the factor node penalty weight increment, adjacent subject edge weight attenuation, grid-side node voltage constraint weight increment, and energy storage node message damping from the feedback message parameters to obtain the current round's feedback write record. Based on the current round of feedback write records, the message iteration base graph is corrected by feedback to obtain the feedback-corrected message iteration base graph; Based on the next round of scheduling benchmark and the feedback-corrected message iteration base map, the improved Max-Sum algorithm iteration is re-executed to obtain the next round of candidate scheduling vectors and the next round of composite message changes. The candidate scheduling vector for the next round is input into the Dykstra projection algorithm, and alternating projection is performed between the constraint sets to obtain the feasible scheduling vector for the next round. The classification projection residual for the next round is calculated based on the deviation between the feasible scheduling vector and the candidate scheduling vector for the next round. The change in the next round of composite messages is compared with the preset message change threshold, and the residuals of each constraint type in the next round of classification projection residuals are compared with the corresponding residual benchmark values ​​to obtain the convergence determination record. When the convergence determination record does not meet the preset convergence conditions, the feedback message parameters are regenerated based on the classification projection residual of the next round, and the feasible scheduling vector of the next round is written into the scheduling benchmark of the next round. When the convergence determination record meets the preset convergence conditions, the next round of feasible scheduling vector is determined as the last round of feasible scheduling vector, and converted into source-grid-load-storage control instructions according to the node variable order.

9. A source-grid-load-storage control system based on multi-agent scheduling, comprising the source-grid-load-storage control method based on multi-agent scheduling as described in any one of claims 1 to 8, characterized in that, include: The operation status sequence establishment module is used to collect distributed power output, load demand, energy storage status of charge, line capacity, voltage safety boundary and historical scheduling results, and establish an operation status sequence according to the scheduling period. The constraint set generation module is used to perform piecewise linearization and safety margin shrinkage on power balance constraints, line capacity constraints, voltage safety constraints and energy storage state of charge constraints according to the operating state sequence, and configure constraint type tags to obtain constraint sets; The dynamic factor graph construction module is used to determine the node variables and coupling relationships of the source-side main body, grid-side main body, load-side main body and energy storage-side main body based on the operating state sequence, and to construct the dynamic factor graph. An improved Max-Sum iteration module is used to write the adjustment margin, local power deviation and constraint proximity of each subject in the dynamic factor graph into a composite message, and perform message iteration to obtain candidate scheduling vectors. The Dykstra projection processing module is used to input candidate scheduling vectors into the Dykstra projection algorithm, perform alternating projections between constraint sets to obtain feasible scheduling vectors, and calculate classification projection residuals. The feedback message generation module is used to establish feedback message parameters based on the constraint type of the categorical projection residual. The closed-loop scheduling output module is used to take the feasible scheduling vector as the basis for the next round of scheduling, re-execute the improved Max-Sum algorithm iteration according to the feedback message parameters, and convert the last round of feasible scheduling vector into source-grid-load-storage control instructions when the preset convergence conditions are met.

Images