A flexible adjustment device control method and device based on dynamic graph cooperation

Abstract

The application belongs to the field of electric power, and discloses a flexible adjustment device control method and device based on dynamic graph cooperation, which comprises the following steps: constructing a dynamic electrical coupling graph according to the power grid topology and the sensitivity of the action of each subunit in the flexible adjustment device to the target line flow, the communication availability and the operation mode; solving a multi-objective constraint distribution problem on a slow time scale to generate local active / reactive power control reference targets of each subunit and corresponding safe feasible region boundaries; on a fast time scale, each subunit takes the local control reference target as a tracking reference, takes the safe feasible region boundary as a hard constraint, and based on local measurement and neighborhood information, uses an online updated local prediction model to perform distributed rolling optimization to generate candidate control actions; the candidate control actions are corrected to the safe feasible region determined by the device available capacity, the power grid operation limit and the operation mode through physical constraint safety projection to obtain executable actions and issue the executable actions for execution.

Description

Technical Field

[0001] This invention belongs to the field of electric power, and particularly relates to a control method and device for a flexible adjustment device based on dynamic graph coordination. Background Technology

[0002] With the large-scale integration of new energy sources and flexible loads, AC power flow exhibits characteristics such as volatility, bidirectionality, and susceptibility to exceeding limits at local sections. Flexible regulation devices address this by distributing multiple series-connected control sub-units along transmission lines or distribution feeders. These sub-units alter the equivalent impedance of the lines or inject series voltages with controllable amplitude and phase into the lines, thereby enabling active and reactive power flow reconfiguration. Compared to centralized, high-capacity power flow control equipment, flexible regulation devices offer advantages such as modularity, flexible expansion, and a smaller impact range when a unit is deactivated.

[0003] In related technologies, multi-subunit control typically employs equal allocation, fixed capacity weight allocation, centralized optimal power flow calculation, or fixed-parameter PI control. These methods can achieve basic power flow regulation when operating conditions are relatively stable and the number of units is small. However, in scenarios involving rapid fluctuations in renewable energy sources, differences in unit capacity, inconsistent device temperature rise, communication delays, topology switching, and subunit exit, mismatches between control allocation and actual regulation capabilities can easily occur.

[0004] Genetic algorithms and particle swarm optimization can be used to tune control parameters offline, but the optimal parameters offline usually only correspond to a limited number of operating conditions. Centralized model predictive control can explicitly handle constraints, but its optimization variables and communication volume increase with the number of sub-units, and anomalies in the central node will affect the entire device. Summary of the Invention

[0005] In view of this, the present invention discloses a flexible adjustment device control method and device based on dynamic graph coordination, which can solve the shortcomings of related technologies.

[0006] To achieve the above objectives, the present invention discloses the following technical solution: According to a first aspect of the present invention, a flexible adjustment device control method based on dynamic graph coordination is proposed, comprising: Based on the power grid topology and the sensitivity of each sub-unit's actions to the power flow of the target line, communication availability, and operating mode in the flexible adjustment device, a dynamic electrical coupling diagram is constructed. Each sub-unit only exchanges status information with neighboring nodes in the diagram. The edge weights in the dynamic electrical coupling diagram are dynamically updated as the topology changes, sensitivity changes, communication quality changes, or the device exits. On a slow time scale, based on the system-level power flow regulation requirements, the available capacity and operating status of each sub-unit, the multi-objective constraint allocation problem is solved, and the local active / reactive power control reference targets and corresponding safe and feasible domain boundaries of each sub-unit are generated, so that sub-units with high sensitivity and large thermal margin can bear greater regulation weight. On a fast timescale, each subunit uses the local control reference target as the tracking benchmark and the safe and feasible domain boundary as the hard constraint. Based on local measurements and neighborhood information, it performs distributed rolling optimization using an online-updated local prediction model to generate candidate control actions. The candidate control actions are corrected by physical constraint safety projection to the safe and feasible domain determined by the available capacity of the equipment, the power grid operating limits and the operating mode, so as to obtain executable actions and issue them for execution.

[0007] According to a second aspect of the present invention, a flexible adjustment device control device based on dynamic graph coordination is provided, comprising: Construction Unit: Based on the power grid topology and the sensitivity, communication availability and operation mode of each sub-unit in the flexible adjustment device to the power flow of the target line, a dynamic electrical coupling diagram is constructed. Each sub-unit only exchanges status information with the neighboring nodes in the diagram. The edge weights in the dynamic electrical coupling diagram are dynamically updated as the topology changes, sensitivity changes, communication quality changes or device exits. First generation unit: On a slow time scale, based on the system-level power flow regulation requirements, the available capacity and operating status of each sub-unit, solve the multi-objective constraint allocation problem, generate the local active / reactive power control reference target and the corresponding safe and feasible domain boundary of each sub-unit, so that the sub-unit with high sensitivity and large thermal margin can bear a greater regulation weight. The second generation unit: On a fast time scale, each subunit uses the local control reference target as the tracking benchmark and the safe and feasible domain boundary as the hard constraint. Based on local measurements and neighborhood information, it performs distributed rolling optimization using an online-updated local prediction model to generate candidate control actions. Execution unit: Corrects the candidate control action to a safe and feasible domain determined by the available capacity of the equipment, the power grid operating limit and the operating mode through physical constraint safety projection, obtains an executable action and issues it for execution.

[0008] According to a third aspect of the present invention, an electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor implements the steps of the method as described in the first aspect by running the executable instructions.

[0009] According to a fourth aspect of the invention, a computer-readable storage medium is provided having computer instructions stored thereon that, when executed by a processor, implement the steps of the method as described in the first aspect.

[0010] As can be seen from the above technical solutions, the beneficial effects of the flexible adjustment device control method and device based on dynamic graph collaboration disclosed in this invention are as follows: On the one hand, by dynamically pruning the electrical coupling graph according to topology, sensitivity, and communication quality, only neighborhood interactions that have a substantial impact on the target power flow are retained, which avoids the burden of full-connection communication and eliminates the reverse regulation and oscillation caused by weak coupling.

[0011] On the other hand, the slow timescale does not generate step-by-step actions, but outputs local control reference targets and safe feasible domain boundaries. It dynamically weights electrical sensitivity, remaining capacity, and thermal state, with high-margin sub-units outputting more power and near-boundary sub-units automatically shrinking, thereby solving the problems of uneven distribution and local overload. The fast timescale uses the slow-layer target as the tracking benchmark and the safe domain as a hard constraint to perform rolling fine-tuning, taking into account the adaptability to new energy fluctuations and the smoothness of control, avoiding the explosion of centralized computing and the blindness of pure local control.

[0012] In addition, after rolling optimization, the equipment-grid feasible region is forcibly pulled back by physical constraint safety projection, so that the optimization results never exceed the equipment capacity, line thermal stability and voltage red line, thereby eliminating the risk of exceeding the limit caused by model mismatch. Attached Figure Description

[0013] Figure 1 This is a flowchart of a flexible adjustment device control method based on dynamic graph coordination, provided in an exemplary embodiment; Figure 2 This is a schematic diagram of the overall architecture of a flexible adjustment device control system provided in an exemplary embodiment; Figure 3 This is a schematic diagram of an intelligent control parameter tuning and execution structure based on dynamic graph collaboration, provided in an exemplary embodiment. Figure 4 This is a schematic diagram of a physical constraint safety action projection mechanism provided in an exemplary embodiment; Figure 5 This is a schematic diagram of an operational mode monitoring and security degradation state machine provided in an exemplary embodiment; Figure 6 This is a schematic diagram of an online intelligent collaborative control process for a flexible adjustment device provided in an exemplary embodiment; Figure 7 This is a schematic diagram of an intelligent control structure for a single flexible adjustment device control node provided in an exemplary embodiment; Figure 8 This is a schematic diagram of a six-node nine-line system and its sub-unit arrangement for algorithm verification provided in an exemplary embodiment; Figure 9 This is a schematic diagram illustrating an algorithm verification scenario, comparison method, and evaluation metrics provided in an exemplary embodiment. Figure 10 This is a schematic structural diagram of a device provided in an exemplary embodiment; Figure 11 This is a block diagram of a flexible adjustment device control device based on dynamic graph coordination, provided in an exemplary embodiment. Detailed Implementation

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of the present invention as detailed in the appended claims.

[0015] It should be noted that the steps of the corresponding methods in other embodiments are not necessarily performed in the order shown and described in this invention. In some other embodiments, the methods may include more or fewer steps than those described in this invention. Furthermore, a single step described in this invention may be broken down into multiple steps in other embodiments; and multiple steps described in this invention may be combined into a single step in other embodiments.

[0016] To address the shortcomings of related technologies, this invention proposes a flexible adjustment device control method and device based on dynamic graph collaboration.

[0017] Figure 1 This is a flowchart illustrating a flexible adjustment device control method based on dynamic graph coordination, as provided in an exemplary embodiment. Figure 1 As shown, the method may include the following steps: Step 101: Based on the power grid topology and the sensitivity of each sub-unit's actions to the power flow of the target line, communication availability, and operating mode in the flexible adjustment device, construct a dynamic electrical coupling diagram. Each sub-unit only exchanges status information with neighboring nodes in the diagram. The edge weights in the dynamic electrical coupling diagram are dynamically updated as the topology changes, sensitivity changes, communication quality changes, or the device exits.

[0018] This flexible regulation device can be a DPFC (Distributed Power Flow Controller), and each sub-unit in the flexible regulation device can be a DPFC sub-unit in the power flow controller. Before constructing the dynamic electrical coupling diagram, it is necessary to collect grid topology data and the operational data of each DPFC sub-unit in the power flow controller. The operational data of each DPFC sub-unit can include: three-phase voltage, current, P / Q, DC side voltage, energy storage status, device temperature, bypass status, and operational history, etc.

[0019] The collected motion data can generate state vectors for each sub-unit. The i-th flexible adjustment device control node forms a state vector during control period k: x i (k)=[U i ,d, U i ,q, I i ,d, I i ,q, P i Q i , Udc,i, E i , Tj,i, m i , a i (k 1), μ i ] ; Among them, U i ,d、U i ,q and I i ,d、I i q represents the local voltage and current components in the synchronous rotating coordinate system; P i Q i For local line power; Udc, i and E i These represent the DC-side voltage and available energy, respectively; Tj,i represents the junction temperature or estimated temperature of the power device; m i For operating mode; a i (k 1) This refers to the action of the previous cycle; μ i For safety margins, the corresponding states can be removed for DPFC sub-units that do not include energy storage or parallel connections.

[0020] Candidate control actions can be selected based on the device structure as series-injected voltage dq components [Uinj,d,Uinj,q]. The equivalent reactance Xeq, phase shift angle δinj, number of modules switched nsm, or parallel-side reactive current Iq,sh are among the following. This invention employs a unified action vector a. i This indicates that the scope of protection should not be limited to a single topology.

[0021] The construction of the dynamic electrical coupling graph specifically includes: treating each sub-unit as a graph node; establishing a directed edge when the action of node j has a significant impact on the power flow of the target line where node i is located; the edge weight is obtained by weighting the electrical distance between nodes, the sensitivity of the action to the active / reactive power flow of the target line, the communication availability, and the compatibility coefficient of the operating mode; updating or deleting the corresponding edge when the line switch state changes, the sub-unit is bypassed, the sensitivity change exceeds the threshold, the communication times out, or the edge weight is lower than the set threshold; each node only sends compressed status messages to neighboring nodes, and when no communication is triggered, the neighboring node estimates the neighboring state by combining the most recent valid message with local prediction.

[0022] Specifically, each sub-unit is treated as a graph node v. i When the action of node j has a significant impact on the power flow of the target line where node i is located, a directed edge is established between node i and node j. The edge weight calculation formula can be: w ij =Norm{exp( αd ij )·(|S p ij |+β|S ij |)·c ij ·r ij}; Where, d ij S represents the electrical distance between nodes; p ij and S ij These represent the sensitivity of node j's action to the active and reactive power flow of the target line at node i, respectively; c ij For communication availability; r ij α represents the operating mode compatibility coefficient; β represents the weights; and Norm{·} represents the normalization function. Sensitivity can be obtained from the AC power flow Jacobian matrix, online small perturbation identification, or a digital simulation model.

[0023] When the line switch state changes, a sub-unit is bypassed, the sensitivity change exceeds a threshold, communication times out, or the edge weight falls below a threshold, the corresponding edge is updated or deleted. The neighborhood summary can be formed using a weighted summation. z i (k)=Σ j ∈N i w ij ·[ΔP j (k), ΔQ j (k), μ j (k), Δa j (k)] ; Among them, z i (k) represents the sum of neighborhood coupling formed by node i during control period k; N i Let w be the set of neighboring nodes of node i in the current dynamic electrical coupling graph; ij Let ΔP be the dynamic edge weight from node j to node i; j (k), ΔQ j (k) represent the active and reactive power flow increments of node j in period k, respectively; the increments can be calculated relative to the measured values ​​of the previous period or the current reference operating point; μ j (k) represents the safety margin of node j; Δa j(k) is the increment of control action of node j; the superscript T means vector transposition.

[0024] In this way, each node only exchanges compressed states that are significantly related to its own power flow control, which not only retains electrical coupling information but also reduces global communication.

[0025] Step 102: on the slow time scale, according to the system-level power flow regulation requirements, the available capacity and operating status of each sub-unit, solve the multi-objective constraint allocation problem, and generate the local active / reactive power control reference target of each sub-unit and the corresponding safe and feasible region boundary, so that sub-units with high sensitivity and large thermal margin bear greater regulation weight.

[0026] As shown in Figure 2 , the structure of the control method proposed by the present invention includes a slow time scale T2 and a fast time scale T1, and satisfies T1 < T2. In the algorithm verification example, T2 is 0.5 s, which is used for updating the sensitivity matrix, dynamic interaction graph and local safety margin; T1 is 0.05 s, which is used for power flow state updating, rolling optimization and action execution. The slow layer outputs the local power flow target, the central value of the action and the capacity margin, and the fast layer generates the equivalent series phase shift or equivalent impedance increment around the above target. This layer does not output PWM pulses, nor does it limit the inner loop of the converter and the electromagnetic dynamics on the DC side.

[0027] In one embodiment, the safe feasible region boundary at least includes the upper limit of the action amplitude of a single DPFC sub-unit, the upper limit of the action change rate in adjacent cycles and the early warning value of the target line load rate; the multi-objective constraint allocation aims at minimizing the weighted sum of power flow tracking error, sub-unit capacity utilization difference, thermal aging cost and safety margin retention term, and the allocation result is dynamically updated with changes in electrical sensitivity, remaining capacity, device temperature and communication availability, so that the local control reference target of sub-units with high sensitivity and low junction temperature corresponds to a larger action amplitude allocation proportion.

[0028] The slow layer solves the objective function according to the system-level power flow regulation requirements ΔPsys, ΔQsys and the action-power flow sensitivity matrices Sp, Sq. Wherein, ΔPsys and ΔQsys respectively represent the deviation between the active and reactive power target values of the target line or target section and the current measured or estimated values, that is, the system-level active and reactive power flow correction that needs to be jointly undertaken by each sub-unit: ; Wherein, Δa is the vector composed of the control action increments of each sub-unit; Sp and Sq are respectively the sensitivity matrices of the control action increments to the active and reactive power flow changes of the target line or target section; λ1 to λ6 are non-negative weight coefficients of each objective term; Ploss is the network loss estimation; η i is the sub-unit capacity utilization; Var(η i) is used to characterize and suppress the difference in capacity utilization among sub-units; Cthermal,i is the thermal aging cost of the i-th sub-unit; μ i Let represent the safety margin of the i-th sub-unit. Constraints include at least the sub-unit capacity, injection voltage, DC-side voltage, line current, bus voltage, rate of change of operation, and reserve capacity. The solution algorithm can employ quadratic programming, model predictive control, or the distributed alternating direction multiplier method.

[0029] Unlike a fixed equal allocation, this allocation result varies with electrical sensitivity, remaining capacity, device temperature, communication availability, and operating mode. Sub-units with high sensitivity, large margin, and low temperature can handle more regulation; sub-units approaching capacity or temperature rise limits automatically reduce their proportion.

[0030] Step 103: On a fast time scale, each sub-unit uses the local control reference target as the tracking benchmark and the safe and feasible domain boundary as the hard constraint. Based on local measurements and neighborhood information, it performs distributed rolling optimization using an online-updated local prediction model to generate candidate control actions.

[0031] The online updated local prediction model specifically includes: establishing an incremental prediction model containing neighborhood coupling terms, quantifying the impact of neighborhood node actions on local power flow as a neighborhood summary quantity and using it as model input; using recursive least squares with a forgetting factor to update model parameters online; freezing parameter updates and reducing model confidence when the model residuals continuously exceed the threshold, the condition number of the identification matrix is ​​too large, or the measurement quality is insufficient; and reducing the weight of neighborhood coupling terms in rolling optimization.

[0032] The i-th node establishes an incremental prediction model that includes neighborhood coupling terms: Δy i (k+1)=A i (k)Δy i (k)+B i (k)Δa i (k)+G i (k)z i (k)+d i (k); Among them, y i The predicted output for the i-th node can be selected as the local P / Q, the target section P / Q, or the mounting point voltage; Δy i (k) represents the increment of the corresponding output in period k; Δa i (k) represents the local control action increment; z i (k) represents the sum of the aforementioned neighborhood coupling; A i B i G i For online updated model parameters; d iThis represents an unmodeled perturbation. Model parameters can be updated using recursive least squares with a forgetting factor: θ i (k)=θ i (k 1)+K i (k)[y i (k) φ i (k)θ i (k 1)]; K i (k)=Π i (k 1)φ i (k) / [λf+φ i (k)Π i (k 1)φ i (k)]; Where, θ i (k) is the parameter vector of the model to be identified for the i-th node in period k; K i (k) is the recursive gain vector; φ i (k) is a regression vector composed of local historical inputs, outputs, and neighborhood summaries; Π i (k 1) is the parameter estimation covariance matrix of the recursive least squares algorithm in the previous cycle; λf is the forgetting factor, with a value range of 0 < λf ≤ 1.

[0033] When the model residuals continuously exceed the threshold, the condition number of the identification matrix is ​​too large, or the measurement quality is insufficient, the parameter updates are frozen and the model confidence is reduced to prevent erroneous models from entering rolling optimization.

[0034] The cost function of distributed rolling optimization includes at least: a tracking error term based on the local active / reactive control reference target, a smoothing term to suppress sudden action changes, a consistency term to suppress action conflicts of strongly coupled neighboring nodes, and a penalty term to preserve safety margin; each node only executes the first control action of the rolling optimization result sequence, and in the next cycle, it is re-identified and re-optimized using new measurements.

[0035] like Figure 3 As shown, each node solves the problem within the prediction time domain Hp based on the local target, prediction model, neighborhood summation, and safety margin: ; Where t is the discrete prediction step in the prediction time domain, and Hp is the prediction time domain length; P i Q i Let P be the predicted active and reactive power values ​​of the i-th node at the corresponding prediction step.i Q i This serves as a reference target for local active and reactive power control at a slow time scale; Δa i To control the increment of actions; a j For the control action of neighboring node j; w ij The neighborhood edge weights in the dynamic electrical coupling graph; Cthermal,i is the thermal aging cost; μ i For safety margin; ωp, ωq, ωu, ωc, ωT, and ωμ are non-negative weighting coefficients for active power tracking, reactive power tracking, action smoothing, neighborhood consistency, thermal stress, and safety margin, respectively. The first and second terms ensure local power flow tracking; the third term restricts abrupt action changes; the fourth term suppresses action conflicts between strongly coupled neighborhood nodes; the fifth term restricts thermal stress; and the sixth term preserves the safety margin. Each node executes only the first control action of the rolling optimization result sequence, and re-identifies and optimizes using new measurements in the next cycle.

[0036] To meet real-time requirements, any of the following implementations can be adopted: First, linearize the local model and solve the small-scale quadratic programming problem; second, use explicit model predictive control to pre-divide typical feasible regions and look up tables online; third, perform only one neighborhood coordination iteration per cycle, with the rest being corrected in the next cycle; fourth, when the solution time exceeds the upper limit, directly use the action of the previous cycle plus sensitivity correction.

[0037] Step 104: Correct the candidate control action to a safe and feasible domain determined by the available capacity of the equipment, the power grid operating limit and the operating mode through physical constraint safety projection, so as to obtain an executable action and issue it for execution.

[0038] like Figure 4 As shown, the optimization objective is to minimize the weighted deviation between the candidate control action and the action executed in the previous cycle. Three layers of constraints are superimposed: constraints on available equipment capacity and rate of change of action on the equipment side, constraints on line load rate and bus voltage on the grid side, and constraints on feasible domain scaling on the operation mode side. When no feasible solution is found in the safety projection, constraint relaxation is performed according to the priority order: device current and temperature take precedence over DC-side safety, DC-side safety takes precedence over line over-limit, and line over-limit takes precedence over power flow tracking. If this is still not feasible, the system switches to deterministic standby control or bypass. Rolling optimization outputs candidate actions. i Subsequently, instead of applying it directly to the power unit, the final action 'a' is obtained through safe projection. i : a i =arg min a (a ã i ) Wi (a ã i )+ρ(a a i (prev) R i (a a i ,prev); stgdevice(a,x i )≤0, ggrid(a,x)≤0, gmode(a,m i )≤0; Where 'a' represents the desired action in the secure projection optimization; i Candidate actions for optimizing output in distributed scrolling; a i ,prev represents the actual action executed in the previous cycle of the i-th node; a i The final executable action obtained from the secure projection; W i R i ρ is the weighting matrix used to measure the deviation of candidate actions and the magnitude of action changes, respectively; ρ is the non-negative weight coefficient of the action smoothing term; gdevice, ggrid, and gmode represent the constraint functions on the equipment side, the power grid side, and the operation mode side, respectively.

[0039] Device constraints gdevice include control action amplitude, available regulation capacity, action change rate, and node engagement status; grid constraints ggrid include line load rate, bus voltage, and target cross-sectional area P / Q; mode constraints gmode are used to enable different feasible domains in normal, overload, communication degradation, capacity derating, topology change, and recovery modes.

[0040] When there is no feasible solution for the safety projection or the calculation times out, the candidate actions are frozen first, and then constraint relaxation is performed according to the priority of "device current and temperature - DC side safety - line current - bus voltage - power flow tracking"; if it is still not feasible, switch to deterministic standby control, current limiting or bypass.

[0041] On one hand, this invention employs a dynamic electrical coupling graph that is adaptively pruned based on topology, sensitivity, and communication quality, retaining only neighborhood interactions that substantially impact the target power flow. This avoids the burden of full-connectivity communication and eliminates reverse regulation and oscillations caused by weak coupling. On the other hand, the slow timescale does not generate step-by-step actions but instead outputs local control reference targets and safe feasible domain boundaries. It dynamically weights electrical sensitivity, remaining capacity, and thermal state, allowing high-margin sub-units to output more power and near-boundary sub-units to automatically shrink, thereby resolving uneven distribution and local overload issues. The fast timescale uses the slow-layer target as the tracking benchmark and the safe domain as a hard constraint to perform rolling fine-tuning, balancing adaptability to renewable energy fluctuations with control smoothness, and avoiding the explosion of centralized computation and the blindness of purely local control. Furthermore, after rolling optimization, physical constraint safe projection forces the device-grid feasible domain back, ensuring that the optimization results never exceed the device capacity, line thermal stability, and voltage limits, thereby eliminating the risk of exceeding limits caused by model mismatch.

[0042] In one embodiment, an event-triggered communication mechanism is also included.

[0043] Node i stores the state of the last transmission x i (k i s A neighborhood message is sent when any of the following conditions are met: ; Among them, D i x is the state normalization or weighting matrix for node i, used to eliminate differences in the dimensions and scales of different state quantities; i (k) is the current periodic state vector, x i (k i s ) represents the state vector at the time of the most recent transmission of a neighborhood message; k i s εi represents the most recent message sending period; eP,i represents the local active power flow tracking error of node i; εx,i, εP,i, and εμ,i represent the event trigger thresholds for state change, active power tracking error, and safety margin, respectively; Kmax represents the maximum allowed number of silent periods; and ||·|2 represents the L2 norm.

[0044] In addition, it is triggered immediately when the operating mode, topology, bypass status, or capacity derating flag changes. The message includes at least the node number, timestamp, compression status, local target, constraint margin, and mode flag. If not triggered, neighboring nodes estimate the neighborhood state using the most recent valid message and their local prediction model.

[0045] In one embodiment, the method further includes: monitoring communication status, model confidence, capacity margin, and topology changes; switching to the corresponding security degradation mode when an anomaly is detected; wherein when communication is interrupted or the device is derated, the corresponding edge weights in the dynamic graph are frozen and the adjustment tasks are reallocated; after communication is restored and the capacity or topology status returns to normal, the method returns to the cooperative control mode without disturbance after a state consistency check, action slope limitation, and cooperative weight gradual increase process.

[0046] like Figure 5 As shown, the control system is configured with initialization / self-test mode M0, intelligent cooperative power flow control mode M1, local autonomous mode M2, line over-limit constraint mode M3, communication degradation mode M4, sub-unit derating / bypass mode M5, and recovery check mode M6.

[0047] M1: When the online model is valid, communication is normal, and the constraint margins meet the requirements, dynamic collaboration and rolling optimization are performed.

[0048] M2: When the regional coordination module is unavailable or neighbor messages time out continuously, sensitivity droop or PI control is performed only based on the local voltage, P / Q and DC side status.

[0049] M3: When the predicted line load rate is close to or exceeds the set limit, the feasible action range is narrowed and the line safety constraint weight is increased to prioritize the elimination of limit violations.

[0050] M4: When neighbor messages time out continuously or regional coordination information is unavailable, freeze the failed communication edge and perform degraded rolling optimization or sensitivity droop control using the most recent valid message and local measurements.

[0051] M5: When the sub-unit capacity is reduced, self-test is abnormal, bypass is enabled, or the action boundary remains infeasible, update the available capacity of the node and reallocate its adjustment tasks; exit the corresponding node if necessary.

[0052] M6: After communication, capacity or topology status is restored, complete the status consistency check, action slope limit and incremental coordination weight, and then restore M1 without disturbance.

[0053] like Figure 6As shown, the complete process of this invention includes: collecting local voltage, current, P / Q, DC side status, temperature, bypass status, and operation history of each sub-unit; constructing or updating a dynamic interaction graph based on topology, electrical distance, P / Q sensitivity, communication availability, and operating mode; solving system-level multi-objective allocation on a slow time scale to generate local objectives, action center values, and safety margins for each sub-unit; updating the local prediction model using new local measurements on a fast time scale and determining whether to trigger neighborhood message transmission; solving finite-time-domain rolling optimization based on local objectives, online prediction models, and neighborhood aggregate quantities to generate candidate actions; executing safety action projections based on equipment, power grid, and operating mode constraints to obtain executable actions; converting executable actions into modulation or switching commands by the device's inner loop and feeding back the actual execution results; switching to the corresponding safety degradation mode if communication is abnormal, sensitivity model mismatch occurs, capacity is derating, topology changes, or optimization is not feasible; and returning to the cooperative mode without disturbance after the recovery conditions are met.

[0054] like Figure 7 As shown, a single control node consists of a local measurement interface, state preprocessing, neighborhood information fusion, adaptive predictive controller, safety projection, mode supervision, backup controller, and device command interface. The algorithm layer output can be represented as a series phase shift, equivalent impedance, or series injected voltage reference; its conversion to the specific device inner loop is implemented using existing interfaces and is not a necessary component for verifying the algorithm of this invention.

[0055] To ensure smooth mode switching, the intelligent controller and the backup controller can calculate candidate actions in parallel. The unselected controller performs tracking or anti-saturation processing on its integral state; during switching, action interpolation or slope limiting is used to ensure continuity of injected voltage, equivalent impedance, and reactive power commands.

[0056] The method proposed in this invention will be described below based on specific data.

[0057] In AC systems with two or more parallel transmission channels, 4 to 20 sub-units are arranged. Sub-units can be series voltage source converters, controllable series reactors, distributed static synchronous series compensators, or power flow control units with series-parallel power exchange capabilities. The rated capacity of each sub-unit can vary.

[0058] The slow timescale T2 can be set to 0.1–2 s, and the fast timescale T1 can be set to 1–100 ms; in the algorithm verification embodiment, T2 is set to 0.5 s and T1 to 0.05 s. If the specific DPFC device contains a faster current inner loop, its control cycle can be set separately according to the converter bandwidth. The event trigger threshold can be set according to per-unit values, and a maximum silence time Kmax is set to prevent the state from expiring due to prolonged lack of messages. The above values ​​are only implementation examples and are not intended as necessary limitations.

[0059] This invention employs a discrete-time DC power flow sensitivity model to verify the algorithm. Verification includes line over-limit suppression under source load changes, task reallocation after capacity derating, degraded operation during neighborhood communication interruptions, sensitivity and dynamic graph updates after topology changes, and the constraint effect of safety projection on action boundaries and line limits. Verification does not include PWM, DC capacitance, current inner loop, short-circuit current, or other electromagnetic transient processes.

[0060] like Figure 8 As shown, the algorithm verification test system adopts a 6-node, 9-line DC power flow model. Buses 1 and 2 are power supply nodes, buses 5 and 6 are load nodes, and buses 3 and 4 are intermediate nodes. Four sub-units are installed on lines 1, 4, 5, and 8, respectively. Their control actions are represented by a unified per-unit variable using series phase shift or equivalent impedance increment. Lines 5 and 7 together form a parallel channel supplying power to bus 5. Power flow can be redistributed between the two lines through sub-unit actions.

[0061] like Figure 9 As shown, typical operating scenarios include: 2.0–4.0 s: The source load distribution gradually changes, causing line 5 in the target transmission channel to approach and exceed the thermal stability boundary; 4.0–7.0 s: Based on steady-state redistribution, the periodicity and random fluctuations of new energy sources are superimposed; 7.0 s: The usable control range of DPFC-3 drops to 35% of its rated value, and inspection tasks are reassigned; 8.0~9.0 s: DPFC-2 neighborhood communication is interrupted, checking for broken edges in the dynamic graph and local degradation control; 9.5 s: Tie line 9 exits operation, recalculates power flow sensitivity and updates control node interaction graph; Throughout the entire process, constraints are applied to ensure that the DPFC action amplitude, action change rate, and the load rate of all operating lines do not exceed 1.0 pu. Interruption of regional coordination information, loss of neighboring messages, and communication recovery; The thermal stability limits of the line are tightened, and the control action amplitude and rate of change are constrained.

[0062] The comparison methods and evaluation indicators include: Validation model: 6-node, 9-line DC power flow model; 4 DPFCs are arranged on lines 1, 4, 5 and 8 respectively; Comparison methods: (1) No control: All DPFC control actions are zero, and only the natural power flow changes caused by source load disturbances are calculated; (2) Equal allocation: Based on the sensitivity of the most severely overloaded line, the required power flow correction is equally allocated according to the number of available DPFCs, and the electrical coupling relationship is not updated; (3) Fixed sensitivity optimization: The fixed sensitivity matrix under the initial operating condition is used to perform power flow optimization with regularization terms, considering the action amplitude and rate of change, but the model is not updated with capacity derating and topology changes; (4) The method of this invention: The power flow sensitivity and dynamic interaction diagram are updated every 0.5 s, and constraint rolling optimization is performed every 0.05 s; the candidate action is projected through the safe feasible domain composed of the line load rate, action amplitude and rate of change; the communication is triggered by the state change threshold and the maximum silent time. Control period: fast timescale 0.05 s, slow timescale 0.5 s, total verification time 12 s; Constraints: The physical upper limit of line load rate is 1.0 pu; the safety projection control target is 0.998 pu; the amplitude of each DPFC action and the rate of change of adjacent cycles are limited, and the available capacity is updated in real time after derating; Key metrics: peak line load rate, number of over-limit steps, over-limit integral, maximum capacity utilization, utilization standard deviation, total variation of actions, and number of communication messages.

[0063] Table 1

[0064] The final verification results are shown in Table 1. Without control, the peak line load rate was 1.0469 pu. Equal allocation and fixed sensitivity optimization still had 160 and 157 steps exceeding the limit, respectively. The method of this invention limits the peak line load rate to 0.9980 pu under all test conditions, with zero steps exceeding the limit and zero integral exceeding the limit, and zero steps where safe projection is infeasible. After 7.0 s derating, DPFC-3 automatically transfers its task to other nodes. During the DPFC-2 communication interruption, the corresponding communication edge is deleted while control remains feasible. After a 9.5 s topology change, the sensitivity matrix and dynamic graph are reconstructed.

[0065] In terms of communication, based on the benchmark of 960 messages corresponding to the full periodic transmission of data by four DPFC nodes within 240 control cycles, the event-triggered mechanism of this technology actually sends 54 messages, reducing the message count by approximately 94.4%. The maximum capacity utilization rate of the method of this invention is 0.654525, the mean standard deviation of capacity utilization is 0.10607, and the total variation of control actions is 0.097272, indicating that communication compression does not come at the cost of line overruns or control action overruns.

[0066] In this local MATLAB environment, the average computation time per cycle of the method of this invention was 0.24487 ms, and the maximum computation time was 2.4514 ms, both of which are lower than the 50 ms fast layer control cycle. These computation times are affected by computer hardware, MATLAB runtime conditions, and program implementation methods, and are only used to illustrate the feasibility of online computation at the scale of this example, and should not be considered as absolute performance limitations on different hardware platforms.

[0067] Figure 10 This is a schematic structural diagram of a device provided in an exemplary embodiment. Please refer to... Figure 10 At the hardware level, the device includes a processor 1002, an internal bus 1004, a network interface 1006, memory 1008, and non-volatile memory 1010, and may also include other hardware required for its functions. One or more embodiments of the present invention can be implemented in software, for example, the processor 1002 reads the corresponding computer program from the non-volatile memory 1010 into the memory 1008 and then runs it. Of course, in addition to software implementation, one or more embodiments of the present invention do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0068] Please refer to Figure 11 A flexible adjustment control device based on dynamic graph coordination can be applied to, for example... Figure 11 The device shown, in order to implement the technical solution of the present invention, includes: The construction unit 1101 is used to construct a dynamic electrical coupling diagram based on the power grid topology and the sensitivity, communication availability and operation mode of the actions of each sub-unit in the flexible adjustment device to the power flow of the target line. Each sub-unit only exchanges status information with the neighboring nodes in the diagram. The edge weights in the dynamic electrical coupling diagram are dynamically updated as the topology changes, sensitivity changes, communication quality changes or device exits. The first generation unit 1102 is used to solve the multi-objective constraint allocation problem on a slow time scale based on the system-level power flow regulation requirements, the available capacity and operating status of each sub-unit, and generate the local active / reactive power control reference targets and corresponding safe and feasible domain boundaries for each sub-unit, so that sub-units with high sensitivity and large thermal margin can bear greater regulation weight. The second generation unit 1103 is used to generate candidate control actions by having each subunit use the local control reference target as the tracking benchmark and the safe and feasible domain boundary as the hard constraint, based on local measurements and neighborhood information, and by using an online-updated local prediction model to perform distributed rolling optimization. The execution unit 1104 is used to correct the candidate control action by physical constraint safety projection to the safe and feasible domain jointly determined by the available capacity of the equipment, the power grid operation limit and the operation mode, so as to obtain an executable action and issue it for execution.

[0069] Optionally, the building unit 1101 is specifically used for: Each sub-unit is treated as a graph node. When the action of node j has a significant impact on the power flow of the target line where node i is located, a directed edge is established. The edge weight is obtained by weighting the electrical distance between nodes, the sensitivity of the action to the active / reactive power flow of the target line, the communication availability, and the operating mode compatibility coefficient. When the state of the line switch changes, the sub-unit is bypassed, the sensitivity change exceeds the threshold, communication timeout occurs, or the edge weight is lower than the set threshold, the corresponding edge is updated or deleted. Each node only sends compressed status messages to neighboring nodes. When no communication is triggered, neighboring nodes estimate the neighborhood status by combining the most recent valid message with local prediction.

[0070] Optionally, the safe and feasible domain boundary includes at least the upper limit of the action amplitude of a single sub-unit, the upper limit of the action change rate between adjacent cycles, and the target line load rate warning value; The multi-objective constraint allocation aims to minimize the weighted average of power flow tracking error, sub-cell capacity utilization difference, thermal aging cost, and safety margin retention term. The allocation result is dynamically updated as electrical sensitivity, remaining capacity, device temperature, and communication availability change, so that the local control reference target of high-sensitivity, low junction temperature sub-cells corresponds to a larger action amplitude allocation ratio.

[0071] Optionally, the second generating unit 1103 is specifically used for: An incremental prediction model with neighborhood coupling terms is established, and the impact of neighborhood node actions on local power flow is quantified into a neighborhood summation quantity and used as the model input. The recursive least squares method with a forgetting factor is used to update the model parameters online. When the model residuals continuously exceed the threshold, the condition number of the identification matrix is ​​too large, or the measurement quality is insufficient, the parameter updates are frozen and the model confidence is reduced, as well as the weight of the neighborhood coupling term in the rolling optimization is reduced.

[0072] Optionally, the cost function of the distributed rolling optimization includes at least: The tracking error term is based on the local active / reactive control reference target, the smoothing term suppresses sudden action changes, the consistency term suppresses action conflicts of strongly coupled neighboring nodes, and the penalty term retains the safety margin. Each node executes only the first control action of the rolling optimization result sequence, and in the next cycle, it re-identifies and re-optimizes using new measurements.

[0073] Optionally, the execution unit 1104 is specifically used for: The optimization objective is to minimize the weighted deviation between the candidate control action and the action executed in the previous cycle, and three layers of constraints are superimposed: equipment available capacity and action change rate constraints on the equipment side, line load rate and bus voltage constraints on the grid side, and feasible domain scaling constraints on the operation mode side. When no feasible solution is found for the safety projection, constraint relaxation is performed according to the priority of device current and temperature over DC side safety, DC side safety over line over-limit, and line over-limit over power flow tracking. If this is still not feasible, switch to deterministic standby control or bypass.

[0074] Optional, also includes: Monitoring unit 1105 is used to monitor communication status, model confidence, capacity margin and topology changes. When an anomaly is detected, it switches to the corresponding security degradation mode. When communication is interrupted or the device is derated, the corresponding edge weight in the dynamic graph is frozen and the adjustment task is redistributed. The return unit 1106 is used to return to the cooperative control mode without disturbance after communication is restored, capacity or topology status is restored to normal, and after a process of state consistency check, action slope limit and cooperative weight gradual increase.

[0075] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0076] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0077] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0078] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0079] For any computer-readable medium (or computer-readable storage medium) as described above or otherwise, computer instructions may be stored thereon, which, when executed by a processor, implement one or more of the embodiments described above, thereby realizing the technical solution of the present invention.

[0080] The present invention also proposes a computer program that, when executed by a processor, implements one or more of the embodiments described above, thereby realizing the technical solution of the present invention. This computer program may be specifically recorded on the above-described or other computer-readable media, and the present invention does not impose any limitations on this.

[0081] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0082] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The terminology used in one or more embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in one or more embodiments of the invention and in the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0084] It should be understood that although the terms first, second, third, etc., may be used to describe various information in one or more embodiments of the present invention, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of one or more embodiments of the present invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0085] The above description is merely a preferred embodiment of one or more embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of the present invention should be included within the protection scope of one or more embodiments of the present invention.

Claims

1. A control method for a flexible adjustment device based on dynamic graph coordination, characterized in that, include: Based on the power grid topology and the sensitivity of each sub-unit's actions to the power flow of the target line, communication availability, and operating mode in the flexible adjustment device, a dynamic electrical coupling diagram is constructed. Each sub-unit only exchanges status information with neighboring nodes in the diagram. The edge weights in the dynamic electrical coupling diagram are dynamically updated as the topology changes, sensitivity changes, communication quality changes, or the device exits. On a slow time scale, based on the system-level power flow regulation requirements, the available capacity and operating status of each sub-unit, the multi-objective constraint allocation problem is solved, and the local active / reactive power control reference targets and corresponding safe and feasible domain boundaries of each sub-unit are generated, so that sub-units with high sensitivity and large thermal margin can bear greater regulation weight. On a fast timescale, each subunit uses the local control reference target as the tracking benchmark and the boundary of the safe and feasible domain as a hard constraint. Based on local measurements and neighborhood information, it performs distributed rolling optimization using an online-updated local prediction model to generate candidate control actions. The candidate control actions are corrected by physical constraint safety projection to the safe and feasible domain determined by the available capacity of the equipment, the power grid operating limits and the operating mode, so as to obtain executable actions and issue them for execution.

2. The method according to claim 1, characterized in that, The construction of the dynamic electrical coupling diagram includes: Each sub-unit is treated as a graph node. When the action of node j has a significant impact on the power flow of the target line where node i is located, a directed edge is established. The edge weight is obtained by weighting the electrical distance between nodes, the sensitivity of the action to the active / reactive power flow of the target line, the communication availability, and the operating mode compatibility coefficient. When the state of the line switch changes, the sub-unit is bypassed, the sensitivity change exceeds the threshold, communication timeout occurs, or the edge weight is lower than the set threshold, the corresponding edge is updated or deleted. Each node only sends compressed status messages to neighboring nodes. When no communication is triggered, neighboring nodes estimate the neighborhood status by combining the most recent valid message with local prediction.

3. The method according to claim 1, characterized in that, The safe and feasible domain boundary includes at least the upper limit of the action amplitude of a single sub-unit, the upper limit of the action change rate between adjacent cycles, and the target line load rate warning value. The multi-objective constraint allocation aims to minimize the weighted average of power flow tracking error, sub-cell capacity utilization difference, thermal aging cost, and safety margin retention term. The allocation result is dynamically updated as electrical sensitivity, remaining capacity, device temperature, and communication availability change, so that the local control reference target of high-sensitivity, low junction temperature sub-cells corresponds to a larger action amplitude allocation ratio.

4. The method according to claim 1, characterized in that, The distributed rolling optimization using an online-updated local prediction model includes: An incremental prediction model with neighborhood coupling terms is established, and the impact of neighborhood node actions on local power flow is quantified into a neighborhood summation quantity and used as the model input. The recursive least squares method with a forgetting factor is used to update the model parameters online. When the model residuals continuously exceed the threshold, the condition number of the identification matrix is ​​too large, or the measurement quality is insufficient, the parameter updates are frozen and the model confidence is reduced, as well as the weight of the neighborhood coupling term in the rolling optimization is reduced.

5. The method according to claim 1, characterized in that, The cost function of the distributed rolling optimization includes at least: The tracking error term is based on the local active / reactive control reference target, the smoothing term suppresses sudden action changes, the consistency term suppresses action conflicts of strongly coupled neighboring nodes, and the penalty term retains the safety margin. Each node executes only the first control action of the rolling optimization result sequence, and in the next cycle, it re-identifies and re-optimizes using new measurements.

6. The method according to claim 1, characterized in that, The step of correcting the candidate control action to a safe and feasible region determined by the available equipment capacity, grid operating limits, and operating mode through physical constraint safety projection includes: The optimization objective is to minimize the weighted deviation between the candidate control action and the action executed in the previous cycle, and three layers of constraints are superimposed: equipment available capacity and action change rate constraints on the equipment side, line load rate and bus voltage constraints on the grid side, and feasible domain scaling constraints on the operation mode side. When no feasible solution is found for the safety projection, constraint relaxation is performed according to the priority of device current and temperature over DC side safety, DC side safety over line over-limit, and line over-limit over power flow tracking. If this is still not feasible, switch to deterministic standby control or bypass.

7. The method according to claim 1, characterized in that, Also includes: Monitor communication status, model confidence, capacity margin and topology changes. When an anomaly is detected, switch to the corresponding security degradation mode. When communication is interrupted or the device is derated, freeze the corresponding edge weights in the dynamic graph and reallocate the adjustment task. After communication is restored and capacity or topology status returns to normal, the system undergoes a process of state consistency check, action slope limitation, and gradual increase of cooperative weight, before returning to cooperative control mode without disturbance.

8. A flexible adjustment device control device based on dynamic graph coordination, characterized in that, include: Construction Unit: Based on the power grid topology and the sensitivity, communication availability and operation mode of each sub-unit in the flexible adjustment device to the power flow of the target line, a dynamic electrical coupling diagram is constructed. Each sub-unit only exchanges status information with the neighboring nodes in the diagram. The edge weights in the dynamic electrical coupling diagram are dynamically updated as the topology changes, sensitivity changes, communication quality changes or device exits. First generation unit: On a slow time scale, based on the system-level power flow regulation requirements, the available capacity and operating status of each sub-unit, solve the multi-objective constraint allocation problem, generate the local active / reactive power control reference target and the corresponding safe and feasible domain boundary of each sub-unit, so that the sub-unit with high sensitivity and large thermal margin can bear a greater regulation weight. The second generation unit: On a fast time scale, each subunit uses the local control reference target as the tracking benchmark and the boundary of the safe and feasible domain as the hard constraint. Based on local measurements and neighborhood information, it performs distributed rolling optimization using an online-updated local prediction model to generate candidate control actions. Execution unit: Corrects the candidate control action to a safe and feasible domain determined by the available capacity of the equipment, the power grid operating limit and the operating mode through physical constraint safety projection, obtains an executable action and issues it for execution.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor implements the steps of the method as described in any one of claims 1-7 by running the executable instructions.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-7.

Images