An intelligent power distribution system and method

Abstract

This invention discloses an intelligent power distribution system and method, comprising: collecting and processing multi-source data to construct an input sample set, segmenting loads and power sources, and generating a mapping matrix; processing the sample set using a physical constraint graph neural ODE model, introducing the mapping matrix into the input layer, and outputting extended prediction results; inputting the extended prediction results into a modulation module, using the mapping matrix to perform cross-phase scheduling and generate a scheduling scheme; implementing a dual-time-domain event-triggered mechanism control scheme, with slow-loop rolling optimization and fast-loop rapid adjustment to generate control commands; issuing the scheduling scheme and control commands, with the power distribution controller, energy storage device, and phase modulator executing the operations; collecting execution feedback, updating the ODE model, mapping matrix, and event thresholds to form an adaptive closed loop. This invention achieves intelligent, controllable, and efficient operation of a distributed power distribution system through segmented phase energy modulation, physical constraint graph neural ODE modeling, and a dual-time-domain event-triggered mechanism.

Description

Technical Field

[0001] This invention relates to the field of intelligent power distribution technology, and more particularly to an intelligent power distribution system and method. Background Technology

[0002] With the large-scale integration of distributed energy resources and flexible loads into distribution networks, the operating characteristics of power systems have become highly dynamic and complex. Existing smart distribution methods mostly rely on traditional power flow calculations, optimized scheduling, or experience-based control logic. While these methods can achieve certain optimization effects in single scenarios, they struggle to adapt to the demands of distributed power output fluctuations, time-varying load characteristics, and coordinated control of multiple types of energy storage. In three-phase distribution networks, the imbalance in power distribution is particularly prominent. Existing methods often mitigate this imbalance through static phase reconstruction or simple load transfer, failing to achieve proactive and flexible management across multiple time scales, leading to frequent voltage fluctuations and degraded power quality.

[0003] On the other hand, existing prediction-based optimization methods often lack the embedding of physical constraints. While the predicted results may be numerically reasonable, they may be infeasible under constraints such as voltage upper and lower limits, branch current constraints, or power balance constraints. This results in the optimized solution not being directly applicable to actual operation, and may even lead to secondary adjustments and frequent equipment switching. Current event-triggered control mechanisms mostly use single thresholds or fixed logic, lacking hierarchical responses to dynamic characteristics at different time scales. This makes it difficult to capture sudden fluctuations at the second level in a timely manner, and is also prone to delays in minute-level scheduling, which adversely affects the stability of system operation and the lifespan of equipment.

[0004] In summary, existing smart power distribution methods have three main shortcomings when facing complex access environments of distributed energy resources and flexible loads: First, they lack a prediction mechanism that can balance real-time performance and feasibility, leading to a disconnect between optimization results and physical operating conditions; second, they lack segmented and flexible phase energy modulation methods, making it difficult to effectively alleviate three-phase imbalance problems; and third, they lack event-triggered control logic with dual time scales, making it impossible to ensure both safety and rapid response and equipment protection.

[0005] Therefore, how to provide an intelligent power distribution system and method 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 an intelligent power distribution system and method. This invention fully utilizes a segmented phase energy modulation mechanism, a physical constraint graph neural network (ODE) model, and dual-time-domain event-triggered control technology. It details the power prediction, optimization, and dynamic regulation process under conditions of large-scale integration of distributed energy resources and flexible loads. By segmenting the load and distributed power supply and constructing a segmented phase mapping matrix, flexible allocation of three-phase power is achieved. Physical constraints and dynamically coupled constraint terms are introduced into the graph neural network (ODE) model to ensure the physical feasibility of the prediction results. Through the dual-time-domain event-triggered mechanism, rolling optimization and rapid adjustment are achieved in minute-level and second-level dual-layer cycles, respectively, balancing the long-term stability and transient response capability of the system. This invention possesses advantages such as high power supply security, strong feasibility of optimization results, good three-phase imbalance mitigation effect, and extended equipment lifespan.

[0007] A smart power distribution method according to an embodiment of the present invention includes:

[0008] Collect multi-source data from the power distribution network, process the multi-source data, construct an input sample set, and segment the power of high-power loads and distributed power sources to generate a segmented phase mapping matrix.

[0009] The input sample set is processed based on the Physically Constrained Graph Neural (ODE) model. The ODE model introduces a piecewise phase mapping matrix in the input layer to jointly encode the electrical features of nodes and the piecewise power features. In the process of ordinary differential equation evolution, dynamic coupling constraint terms based on piecewise power features are added to form an extended prediction result set.

[0010] The extended prediction result set is input into the segmented phase energy modulation module, and cross-phase time slot scheduling is performed using the segmented phase mapping matrix. Part of the power is allocated in different phases and time slots, and a segmented phase scheduling scheme is generated by combining the energy storage output and the power distribution topology.

[0011] The segmented phase scheduling scheme is controlled based on a dual-time-domain event triggering mechanism. The slow loop performs rolling optimization within a minute-level cycle to generate a scheduling scheme, while the fast loop performs rapid adjustment within a second-level cycle based on the physical security domain triggering condition, the segmented residual triggering condition, and the dynamic evolution triggering condition of the security domain boundary to generate event-triggered control instructions.

[0012] The dispatching scheme and event-triggered control commands are sent to the power distribution controller, energy storage control device and phase energy modulator to perform power distribution and phase switching operations.

[0013] The feedback data collected after execution is compared with the extended prediction result set to update the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold, forming an adaptive closed loop of prediction-optimization-control.

[0014] Optionally, the multi-source data includes node voltage data, node current data, distributed power output data, energy storage device state of charge data, flexible load power data, environmental meteorological data, line switch status data, and transformer temperature rise data. The processing of the multi-source data includes filling in missing values, removing outliers, normalizing the data, and aligning the timestamps.

[0015] Optionally, generating the segmented phase mapping matrix refers to segmenting the power of high-power loads and distributed power sources according to a preset power threshold, and representing the mapping relationship between each segmented power and the three-phase power supply phase in matrix form. In the matrix, the element corresponding to each segmented power and a single phase has a value of 1, and the element corresponding to a non-corresponding phase has a value of 0, thus forming a segmented phase mapping matrix.

[0016] Optionally, forming the extended prediction result set includes:

[0017] Based on the distribution network topology and phase sequence information, a three-phase phase subgraph is established. The node set is copied into three groups of sub-nodes according to phases A, B, and C. The edge set is established into edge channels according to in-phase connection and cross-phase coupling. A physical constraint graph neural ODE model is constructed, which includes an input binding encoder, a continuous-time evolution unit, a physical consistency corrector, and a feasible region projector.

[0018] An input-bound encoder is constructed, which uses node voltage data and node current data as node electrical features, and uses the segmented high-power load power and distributed power source power as segmented power features. The segmented phase mapping matrix and time slice identifier are then used for joint encoding.

[0019] Establish a continuous-time evolution unit to enable the state channel and the dual embedded channel to evolve in parallel. The state channel updates the node voltage and branch current, while the dual embedded channel updates the voltage sensitivity dual quantity to active power change and the voltage sensitivity dual quantity to reactive power change. According to the segmented power characteristics, the evolution rate and aggregation weight are adjusted on the affected nodes and adjacent edges. The unbalance information is transmitted between the three-phase phase sub-diagrams according to the inter-phase coupling parameters.

[0020] After each evolution step, the physical consistency corrector and feasible region projector are executed sequentially to correct and project the intermediate state according to the power flow constraints, thermal limits, node voltage range and branch current upper limit. The safety margin is calculated by the side channel based on the distance between the current state and the constraint boundary.

[0021] After completing the continuous time evolution of the set prediction period, an extended prediction result set is output. The extended prediction result set includes the voltage prediction of each node at each prediction time, the current prediction of each branch at each prediction time, the sensitivity dual of voltage to active power change, the sensitivity dual of voltage to reactive power change, and the dynamic evolution of safety margin.

[0022] Optionally, the generation of the segmented phase scheduling scheme includes:

[0023] The segmented phase energy modulation module receives the extended prediction result set, the segmented phase mapping matrix and the segmented power components, and establishes a time slice sequence for cross-phase time slot scheduling according to a fixed time slice division.

[0024] Based on the extended prediction result set, a priority sequence of phase and time slices is generated. According to the deterministic sorting rules of safety margin priority, low sensitivity duality priority and high line thermal limit margin priority, the segmented power components are initially allocated. The mapped segmented power components are activated in the corresponding phase channel, and the unmapped segmented power components are set to zero in the phase channel to form an initial allocation table.

[0025] The initial allocation table is subjected to segmented sliding and shaping processing. The segmented power components are finely slid and rearranged in adjacent time slots. The upper limit of three-phase imbalance drift and the threshold of intra-segment jitter suppression are set. Through sliding and rearrangement, the three-phase imbalance and power fluctuation amplitude are reduced. The allocation result formed by the segmented power components and the segmented phase mapping matrix in different phases and time slots is defined as cross-phase time slot allocation.

[0026] Based on the extended prediction result set, the switching safety window is determined. Within the switching safety window, a phase switching sequence is synthesized. Constraints are set on the maximum switching frequency, minimum holding time, and minimum interval between adjacent switching. Power transition compensation is performed before and after switching in combination with the energy storage output boundary to form a segmented switching subsequence.

[0027] A feasibility check is performed on the result synthesized by the segmented sliding shaping and phase switching sequence. The feasibility check includes the upper and lower limits of node voltage, the upper limit of branch current, the power balance constraint, and the consistency constraint of segmented phase mapping. After the check is passed, the segmented phase scheduling scheme is output.

[0028] Optionally, the generation event triggers the control instruction, including:

[0029] Set minute-level slow loop period and second-level fast loop period, receive segmented phase scheduling scheme and extended prediction result set, define dynamic evolution of safety domain boundary based on safety margin changes in consecutive adjacent time moments, and set safety margin threshold, evolution threshold and segmented residual threshold.

[0030] Within each minute-level slow loop cycle, rolling optimization is performed based on the segmented phase scheduling scheme and the extended prediction result set to generate a scheduling scheme covering the prediction period, and satisfying the upper and lower limits of node voltage, the upper limit of branch current, power balance and segmented phase mapping consistency constraints.

[0031] Within each second-level fast loop cycle, the current safety margin and the dynamic evolution of the current safety domain boundary are calculated based on the real-time measurement and extended prediction result set. When the current safety margin is less than the safety margin threshold, or when the decrease in the dynamic evolution of the current safety domain boundary exceeds the evolution threshold, the physical safety domain triggering condition is determined to be met.

[0032] Within each second-level fast loop cycle, the difference between the actual cross-phase time slot allocation and the corresponding allocation in the segmented phase scheduling scheme is compared to obtain the segmented residual. When the segmented residual is greater than the segmented residual threshold, it is determined that the segmented residual triggering condition is met. The deviation between the current dynamic evolution of the security domain boundary and the reference evolution given by the extended prediction result set is compared. When the deviation exceeds the evolution threshold, it is determined that the dynamic evolution triggering condition of the security domain boundary is met.

[0033] When any triggering condition is determined to be met, an event-triggered control instruction is generated. The event-triggered control instruction includes the modification of the energy storage output plan, the incremental adjustment of the phase and time slice allocation table for cross-phase time slot allocation, and the insertion, deletion or timing rearrangement of segmented switching sub-sequences.

[0034] Optionally, the power allocation and phase switching operation includes:

[0035] Receive scheduling schemes and event-triggered control instructions, integrate event-triggered control instructions with scheduling schemes to obtain the corrected cross-phase time slot allocation table, segmented switching subsequence and energy storage output plan;

[0036] A set of control instructions for phases and time slots is generated based on the revised cross-phase time slot allocation table. The set of control instructions is received and executed by the phase energy modulator to drive each phase to complete the allocation and switching of segmented power within the corresponding time slot.

[0037] An energy storage control sequence is generated based on the revised energy storage output plan. The energy storage control sequence includes charging control commands, discharging control commands, and holding control commands, which are received and executed by the energy storage control device to balance electrical energy and suppress real-time power fluctuations.

[0038] The control command set for phase and time slices, the modified segmented switching subsequence, and the energy storage control sequence are sent to the power distribution controller, energy storage control device, and phase energy modulator to execute the corresponding power distribution, phase switching, and energy storage regulation operations.

[0039] Optionally, updating the physical constraint graph neural ODE model, the piecewise phase mapping matrix, and the event triggering threshold includes:

[0040] The feedback data includes real-time values ​​of node voltage, real-time values ​​of branch current, actual executed values ​​of energy storage output, actual executed sequence of phase switching, and actual completed results of power allocation, and is aligned hourly with the extended prediction result set.

[0041] By comparing the feedback data with the extended prediction result set, node voltage error, branch current error, energy storage output deviation and phase switching deviation are generated to form an error sequence.

[0042] The physical constraint graph neural ODE model is updated based on the error sequence. The update includes adjusting the feature weights of the input binding encoder, correcting the aggregation weights of the continuous-time evolution unit and the coupling coefficients of the dynamic coupling constraint.

[0043] The segmented phase mapping matrix is ​​corrected based on the feedback data. While maintaining the consistency of the correspondence between segmented power components and phase channels, the mapping results of some segmented power components are updated according to the actual execution situation. Unmapped segmented power components are set to zero in the corresponding phase channels.

[0044] The event trigger threshold is dynamically adjusted based on the error sequence and the operation of the safety domain, and the updated event trigger threshold is written into the dual time-domain event triggering mechanism to form an adaptive closed loop of prediction-optimization-control.

[0045] An intelligent power distribution system according to an embodiment of the present invention includes the following modules:

[0046] The data acquisition and processing module is used to acquire multi-source data from the power distribution network, construct an input sample set, and generate a segmented phase mapping matrix.

[0047] The predictive modeling module is used to generate an extended prediction result set based on the physical constraint graph neural ODE model.

[0048] The segmented phase energy modulation module is used to receive the extended prediction result set, perform cross-phase time slot scheduling using the segmented phase mapping matrix, and generate a segmented phase scheduling scheme.

[0049] The dual-time-domain event triggering control module is used to execute rolling optimization to generate scheduling schemes in the minute-level slow loop and generate event triggering control instructions in the second-level fast loop.

[0050] The execution control module is used to issue scheduling schemes and event-triggered control commands to the power distribution controller, energy storage device and phase energy modulator, and to perform power distribution and phase switching operations.

[0051] The adaptive update module is used to collect feedback data, compare it with the extended prediction result set, and update the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold.

[0052] The beneficial effects of this invention are:

[0053] This invention introduces a segmented phase energy modulation mechanism into the power distribution network, segmenting high-power loads and distributed generation power to generate a segmented phase mapping matrix. This enables flexible switching of power allocation across different phases and time slices. Compared to traditional static phase reconstruction methods, this invention actively adjusts the three-phase power distribution, dynamically reducing imbalance and voltage fluctuations, and improving power quality and supply stability.

[0054] This invention employs a physical constraint graph neural ODE model in the prediction stage, jointly encoding the electrical characteristics of nodes and the segmented power characteristics, and embedding dynamic coupling constraints during the evolution process. This ensures that the prediction results not only reflect the temporal evolution trend, but also meet physical constraints such as voltage upper and lower limits, branch current and power balance. This avoids the defect in existing methods where the prediction results cannot be directly used for scheduling, improves the physical feasibility and implementation value of the optimized solution, and provides reliable support for scheduling.

[0055] The dual-time-domain event triggering mechanism proposed in this invention achieves layered response to system operation by combining minute-level slow-loop rolling optimization with second-level fast-loop dynamic adjustment. The slow loop ensures the global optimality of overall operation, while the fast loop quickly captures sudden fluctuations and performs immediate corrections, balancing long-term system stability with rapid transient response and reducing frequent equipment switching and wear. This invention achieves integrated prediction-optimization-control, providing advantages such as high safety, superior operating efficiency, and extended equipment lifespan in complex power distribution networks. Attached Figure Description

[0056] 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:

[0057] Figure 1 Here is a flowchart of an intelligent power distribution method proposed in this invention;

[0058] Figure 2 This is a schematic diagram of the structure of an intelligent power distribution system proposed in this invention. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0060] refer to Figure 1 A smart power distribution method, comprising:

[0061] Collect multi-source data from the power distribution network, process the multi-source data, construct an input sample set, and segment the power of high-power loads and distributed power sources to generate a segmented phase mapping matrix.

[0062] The input sample set is processed based on the Physically Constrained Graph Neural (ODE) model. The ODE model introduces a piecewise phase mapping matrix in the input layer to jointly encode the electrical features of nodes and the piecewise power features. In the process of ordinary differential equation evolution, dynamic coupling constraint terms based on piecewise power features are added to form an extended prediction result set.

[0063] The extended prediction result set is input into the segmented phase energy modulation module, and cross-phase time slot scheduling is performed using the segmented phase mapping matrix. Part of the power is allocated in different phases and time slots, and a segmented phase scheduling scheme is generated by combining the energy storage output and the power distribution topology.

[0064] The segmented phase scheduling scheme is controlled based on a dual-time-domain event triggering mechanism. The slow loop performs rolling optimization within a minute-level cycle to generate a scheduling scheme, while the fast loop performs rapid adjustment within a second-level cycle based on the physical security domain triggering condition, the segmented residual triggering condition, and the dynamic evolution triggering condition of the security domain boundary to generate event-triggered control instructions.

[0065] The dispatching scheme and event-triggered control commands are sent to the power distribution controller, energy storage control device and phase energy modulator to perform power distribution and phase switching operations.

[0066] The feedback data collected after execution is compared with the extended prediction result set to update the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold, forming an adaptive closed loop of prediction-optimization-control.

[0067] In this embodiment, the multi-source data includes node voltage data, node current data, distributed power output data, energy storage device state of charge data, flexible load power data, environmental meteorological data, line switch status data, and transformer temperature rise data. The processing of the multi-source data includes filling in missing values, removing outliers, normalizing the data, and aligning the timestamps.

[0068] In this embodiment, generating a segmented phase mapping matrix refers to segmenting the power of high-power loads and distributed power sources according to a preset power threshold, and representing the mapping relationship between each segment of power and the three-phase power supply phase in matrix form. In the matrix, the element corresponding to each segment of power and a single phase has a value of 1, and the element corresponding to a non-corresponding phase has a value of 0, thus forming a segmented phase mapping matrix.

[0069] In this embodiment, forming the extended prediction result set includes:

[0070] Based on the distribution network topology and phase sequence information, a three-phase phase subgraph is established. The node set is copied into three groups of sub-nodes according to phases A, B, and C. The edge set is established into edge channels according to in-phase connection and cross-phase coupling. A physical constraint graph neural ODE model is constructed, which includes an input binding encoder, a continuous-time evolution unit, a physical consistency corrector, and a feasible region projector.

[0071] An input-bound encoder is constructed, which uses node voltage data and node current data as node electrical features, and uses the segmented high-power load power and distributed power source power as segmented power features. The segmented phase mapping matrix and time slice identifier are then used for joint encoding.

[0072] A continuous-time evolution unit is established to enable the state channel and the dual embedded channel to evolve in parallel. The state channel updates the node voltage and branch current, while the dual embedded channel updates the voltage sensitivity dual quantity to active power change and the voltage sensitivity dual quantity to reactive power change. The evolution rate and aggregation weight are adjusted on the affected nodes and adjacent edges according to the segmented power characteristics. The unbalance information is transmitted between the three-phase sub-graphs according to the inter-phase coupling parameters. The inter-phase coupling parameters refer to the cross-phase coupling coefficients extracted based on the symmetrical and asymmetrical impedance of the distribution line, the electrical mutual inductance between nodes and the segmented phase mapping relationship, which are used to describe the unbalance transmission intensity of voltage, current and power between the three-phase sub-graphs.

[0073] After each evolution step, a physical consistency corrector and a feasible region projector are executed sequentially. The intermediate state is corrected and projected according to power flow constraints, thermal limits, node voltage ranges, and branch current limits. The safety margin is calculated by the side channel based on the distance between the current state and the constraint boundary. Specifically, the calculation of the safety margin by the side channel based on the distance between the current state and the constraint boundary involves:

[0074] Node voltage margin calculation: Based on the difference between each node voltage in the intermediate state and the allowable upper and lower limits, the minimum difference is selected as the node voltage safety margin, which is used to reflect the degree to which the node voltage approaches the constraint boundary.

[0075] Branch current margin calculation: Based on the difference between the current of each branch in the intermediate state and the maximum allowable current value, the current safety margin of each branch is obtained, and the minimum value is taken as the current safety margin.

[0076] The calculation of thermal limit and power balance margin: Based on the difference between the equipment power or heat load in the intermediate state and the rated thermal limit, and combined with the distribution of segmented power components in the three phases, the thermal limit safety margin is obtained.

[0077] After completing the continuous time evolution of the set prediction period, an extended prediction result set is output. The extended prediction result set includes the voltage prediction of each node at each prediction time, the current prediction of each branch at each prediction time, the sensitivity dual of voltage to active power change, the sensitivity dual of voltage to reactive power change, and the dynamic evolution of safety margin.

[0078] In this embodiment, the generation of the segmented phase scheduling scheme includes:

[0079] The segmented phase energy modulation module receives the extended prediction result set, the segmented phase mapping matrix and the segmented power components, and establishes a time slice sequence for cross-phase time slot scheduling according to a fixed time slice division.

[0080] Based on the extended prediction result set, a priority sequence of phase and time slices is generated. According to the deterministic sorting rules of safety margin priority, low sensitivity duality priority and high line thermal limit margin priority, the segmented power components are initially allocated. The mapped segmented power components are activated in the corresponding phase channel, and the unmapped segmented power components are set to zero in the phase channel to form an initial allocation table.

[0081] The initial allocation table is subjected to segmented sliding and shaping processing. The segmented power components are finely slid and rearranged in adjacent time slots. The upper limit of three-phase imbalance drift and the threshold of intra-segment jitter suppression are set. Through sliding and rearrangement, the three-phase imbalance and power fluctuation amplitude are reduced. The allocation result formed by the segmented power components and the segmented phase mapping matrix in different phases and time slots is defined as cross-phase time slot allocation.

[0082] A switching safety window is determined based on the extended prediction result set. Within this window, a phase switching sequence is synthesized, with constraints set for the maximum switching frequency, minimum hold time, and minimum interval between adjacent switches. Power transition compensation is then performed before and after switching, incorporating the energy storage output boundary, to form a segmented switching sub-sequence. Specifically, determining the switching safety window based on the extended prediction result set involves:

[0083] Voltage stability determination: Based on the dynamic evolution trend of the voltage of each node in the extended prediction result set, the time interval in which the voltage fluctuation is within the allowable range and the rate of change is lower than the set threshold is selected and marked as the voltage stability window;

[0084] Current feasibility determination: Based on the difference between the current of each branch and the upper limit of capacity in the extended prediction result set, the time interval in which the current of each branch does not exceed the set safety margin is determined and marked as the current feasible window;

[0085] Power balance determination: Based on the phase segment power components and energy storage compensation capacity in the extended prediction result set, determine whether the total power difference before and after the switch is within the energy storage boundary compensation range, and mark the time interval that meets the condition as the power balance window;

[0086] Determining the switch safety window: The intersection of the voltage stability window, the current feasible window, and the power balance window is taken to form the final switch safety window, which is used as the time reference for the synthesis of the phase switching sequence.

[0087] A feasibility check is performed on the result synthesized from the segmented sliding shaping and phase switching sequences. The feasibility check includes constraints on node voltage upper and lower limits, branch current upper limits, power balance constraints, and segmented phase mapping consistency constraints. Upon successful check, a segmented phase scheduling scheme is output, wherein:

[0088] The node voltage upper and lower limit constraints refer to the requirement that the voltage of each node should be kept within a preset allowable range in the segmented phase scheduling scheme, that is, not lower than the lower limit of the rated voltage and not exceeding the upper limit of the rated voltage, so as to ensure the stability of the node voltage.

[0089] The upper limit constraint of branch current refers to the requirement that the current of each branch in the segmented phase dispatch scheme must not exceed the corresponding maximum allowable current value, and maintain a safety margin with the rated capacity to prevent line overload.

[0090] Power balance constraint refers to the requirement that all segmented power components, energy storage output and distributed power generation power should meet the energy conservation requirement in the segmented phase dispatch scheme, that is, the input power and output power should remain in balance during the dispatch cycle to avoid supply and demand imbalance.

[0091] The segmented phase mapping consistency constraint means that in a segmented phase scheduling scheme, the segmented power components must be consistent with the corresponding phase and time slot in the segmented phase mapping matrix, and undefined power allocation across phases or time slots is prohibited.

[0092] In this embodiment, the generation of event-triggered control instructions includes:

[0093] Set minute-level slow loop period and second-level fast loop period, receive segmented phase scheduling scheme and extended prediction result set, define dynamic evolution of safety domain boundary based on safety margin changes in consecutive adjacent time moments, and set safety margin threshold, evolution threshold and segmented residual threshold.

[0094] Within each minute-level slow loop cycle, rolling optimization is performed based on the segmented phase scheduling scheme and the extended prediction result set to generate a scheduling scheme covering the prediction period, satisfying the constraints of node voltage upper and lower limits, branch current upper limit, power balance, and segmented phase mapping consistency. Specifically, the rolling optimization involves:

[0095] At the beginning of each minute-level slow loop cycle, the latest state information in the extended prediction result set is called to update the predicted values ​​of node voltage, branch current and segmented power components in multiple future time slices;

[0096] During the rolling optimization process, constraints on the upper and lower limits of node voltage, the upper limit of branch current, the power balance constraint, and the consistency constraint of segmented phase mapping are introduced to ensure that the generated scheduling solution meets the physical feasibility requirements.

[0097] With the goals of reducing the three-phase imbalance rate, reducing the frequency of equipment switching, and improving the utilization rate of energy storage output, the distribution method of segmented power components in phase and time slices is optimized and adjusted.

[0098] By adopting a rolling time window approach, the scheduling for the current time period is fixed and executed, and the prediction results for future time periods are recalculated with the scheduling variables to form a gradually updated scheduling scheme.

[0099] After completing the rolling optimization calculation, the segmented phase scheduling scheme covering the prediction period is output and stored in the slow loop scheduling queue.

[0100] Within each second-level fast loop cycle, the current safety margin and the dynamic evolution of the current safety domain boundary are calculated based on the real-time measurement and extended prediction result set. When the current safety margin is less than the safety margin threshold, or when the decrease in the dynamic evolution of the current safety domain boundary exceeds the evolution threshold, the physical safety domain triggering condition is determined to be met.

[0101] Within each second-level fast loop cycle, the difference between the actual cross-phase time slot allocation and the corresponding allocation in the segmented phase scheduling scheme is compared to obtain the segmented residual. When the segmented residual is greater than the segmented residual threshold, it is determined that the segmented residual triggering condition is met. The deviation between the current dynamic evolution of the security domain boundary and the reference evolution given by the extended prediction result set is compared. When the deviation exceeds the evolution threshold, it is determined that the dynamic evolution triggering condition of the security domain boundary is met.

[0102] When any triggering condition is determined to be met, an event-triggered control instruction is generated. The event-triggered control instruction includes the modification of the energy storage output plan, the incremental adjustment of the phase and time slice allocation table for cross-phase time slot allocation, and the insertion, deletion or timing rearrangement of segmented switching sub-sequences.

[0103] In this embodiment, the power allocation and phase switching operation includes:

[0104] Receive scheduling schemes and event-triggered control instructions, integrate event-triggered control instructions with scheduling schemes to obtain the corrected cross-phase time slot allocation table, segmented switching subsequence and energy storage output plan;

[0105] A set of control commands for phases and time slices is generated based on the revised cross-phase time slot allocation table. This set of commands is received and executed by the phase energy modulator, driving each phase to complete the allocation and switching of segmented power within its corresponding time slice. Specifically, the generation of the control command set for phases and time slices based on the revised cross-phase time slot allocation table involves:

[0106] The corrected cross-phase time slot allocation is expanded in the order of time slots, and the segmented power components in each time slot are analyzed to generate corresponding phase switching instructions and power allocation instructions.

[0107] Based on the segmented phase mapping matrix, each segmented power component is bound to its corresponding phase and time slice. Before generating the instruction, it is verified whether the constraints of maximum switching frequency, minimum hold time and minimum interval between adjacent switching are met. If there is a conflict, it is adjusted.

[0108] During the phase switching command generation process, combined with the energy storage output boundary, power transition compensation commands before and after switching are added to ensure smooth voltage and power transition;

[0109] The switching and allocation instructions that have been mapped, verified and compensated are encapsulated to form a control instruction set for phase and time slices;

[0110] An energy storage control sequence is generated based on the revised energy storage output plan. This sequence includes charging control commands, discharging control commands, and holding control commands, which are received and executed by the energy storage control device to balance electrical energy and suppress real-time power fluctuations. Specifically, generating the energy storage control sequence based on the revised energy storage output plan involves:

[0111] When the prediction results and the segmented phase scheduling scheme indicate that there is a power surplus, the charging power and charging time of the energy storage battery are set according to the revised energy storage output plan, and a charging control command is generated.

[0112] When the prediction results and the segmented phase scheduling scheme indicate that there is a risk of insufficient power or voltage drop, the discharge power and discharge duration of the energy storage battery are set according to the revised energy storage output plan, and a discharge control command is generated.

[0113] When power supply and demand are basically balanced and there is no significant risk of fluctuation, the energy storage battery is set to maintain its current state of charge or dynamically adjust within a small range according to the revised energy storage output plan, and a maintenance control command is generated.

[0114] The control command set for phase and time slices, the modified segmented switching subsequence, and the energy storage control sequence are sent to the power distribution controller, energy storage control device, and phase energy modulator to execute the corresponding power distribution, phase switching, and energy storage regulation operations.

[0115] In this embodiment, updating the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold includes:

[0116] The feedback data includes real-time values ​​of node voltage, real-time values ​​of branch current, actual executed values ​​of energy storage output, actual executed sequence of phase switching, and actual completed results of power allocation, and is aligned hourly with the extended prediction result set.

[0117] By comparing the feedback data with the extended prediction result set, node voltage error, branch current error, energy storage output deviation and phase switching deviation are generated to form an error sequence.

[0118] The physical constraint graph neural ODE model is updated based on the error sequence. The update includes adjusting the feature weights of the input binding encoder, correcting the aggregation weights of the continuous-time evolution unit and the coupling coefficients of the dynamic coupling constraint.

[0119] The segmented phase mapping matrix is ​​corrected based on the feedback data. While maintaining the consistency of the correspondence between segmented power components and phase channels, the mapping results of some segmented power components are updated according to the actual execution situation. Unmapped segmented power components are set to zero in the corresponding phase channels.

[0120] The event trigger threshold is dynamically adjusted based on the error sequence and the operation of the safety domain, and the updated event trigger threshold is written into the dual time-domain event triggering mechanism to form an adaptive closed loop of prediction-optimization-control.

[0121] refer to Figure 2 An intelligent power distribution system includes the following modules:

[0122] The data acquisition and processing module is used to acquire multi-source data from the power distribution network, construct an input sample set, and generate a segmented phase mapping matrix.

[0123] The predictive modeling module is used to generate an extended prediction result set based on the physical constraint graph neural ODE model.

[0124] The segmented phase energy modulation module is used to receive the extended prediction result set, perform cross-phase time slot scheduling using the segmented phase mapping matrix, and generate a segmented phase scheduling scheme.

[0125] The dual-time-domain event triggering control module is used to execute rolling optimization to generate scheduling schemes in the minute-level slow loop and generate event triggering control instructions in the second-level fast loop.

[0126] The execution control module is used to issue scheduling schemes and event-triggered control commands to the power distribution controller, energy storage device and phase energy modulator, and to perform power distribution and phase switching operations.

[0127] The adaptive update module is used to collect feedback data, compare it with the extended prediction result set, and update the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold.

[0128] Example 1:

[0129] To verify the feasibility of this invention in practice, it was applied to an industrial park where the power distribution network was connected to a distributed photovoltaic power station (1.5MW installed capacity), a small wind turbine (500kW), multiple centralized electric vehicle charging stations, and cold storage loads. Due to the significant weather-dependent impact on distributed power output, coupled with the randomness of charging and cold storage loads, the power distribution network frequently experienced three-phase voltage imbalance, current exceeding limits, and frequent equipment switching. Operational statistics from June 2025 showed that during peak photovoltaic output at midday and peak evening charging periods, the three-phase voltage imbalance rate exceeded 3%, and the current in some branches approached the rated upper limit. Park users repeatedly reported issues such as flickering lighting and extended electric vehicle charging times.

[0130] In industrial parks, the intelligent power distribution method proposed in this invention is applied to acquire multi-source operating data, including three-phase voltage, current, load power and distributed power output, through a data acquisition and processing module. Electric vehicle charging load and cold storage compressor load are segmented and processed, generating segmented power components in 5kW segments. Based on this, a segmented phase mapping matrix is ​​constructed, and high-power loads are transformed into multiple controllable small units, forming a fine-grained schedulable foundation.

[0131] In the predictive modeling stage, a Physically Constrained Graph Neural Equation (ODE) model is used to jointly process the collected data and the segmented phase mapping matrix. The ODE model encodes both node electrical characteristics and segmented power characteristics at the input layer, and adds dynamic coupling constraints during the evolution process to ensure that the prediction results meet the requirements of node voltage upper and lower limits, branch current capacity limits, and energy conservation. The model prediction results indicate that, during the period from 14:00 to 16:00 on July 15th, without intervention measures, the B-phase voltage may drop to 93% of its rated value, and the branch current may exceed 5% of its rated value, posing an operational risk.

[0132] During the scheduling optimization phase, the segmented phase energy modulation module of this invention receives the prediction results and performs cross-phase time slot scheduling using the segmented phase mapping matrix. For example, it delays the start-up of some electric vehicle charging segments by 10 minutes, advances the start-up of some cold storage compressor segments by 15 minutes, and schedules energy storage batteries to compensate with a power output of 300kW during photovoltaic power fluctuations. The generated segmented phase scheduling scheme achieves a reasonable power distribution among the three phases, keeping the expected imbalance rate within 2%.

[0133] During the operation and control phase, the system employed a dual-time-domain event triggering mechanism. A minute-level slow loop performed rolling optimization to ensure overall operational stability; a second-level fast loop rapidly triggered adjustments based on real-time monitoring. When the voltage of phase A dropped to 94% of its rated value at 15:12, the fast loop triggered an event control command, immediately adjusting the energy storage output by 100kW and switching some electric vehicle charging power from phase A to phase C. The voltage level was restored within 3 seconds, ensuring power quality.

[0134] After execution, the system collects feedback data and compares it with the prediction results. The results show that the method of the present invention effectively alleviates three-phase imbalance, reduces current overruns, and lowers the switching frequency of the equipment. Statistics from the park operator during a one-month trial run show a significant improvement in power quality, increased charging efficiency on the user side, reduced temperature fluctuations in the cold storage, and extended equipment lifespan.

[0135] Table 1 Comparison of the Operation Performance of Power Distribution Networks in Industrial Parks

[0136]

[0137] As can be seen from the data comparison in Table 1, the method of the present invention effectively reduces the three-phase imbalance rate in all time periods. During the period from 10:00 AM to 12:00 PM, the three-phase imbalance rate was 2.8% under the traditional method, which decreased to 1.3% after applying the present invention. During the period from 2:00 PM to 4:00 PM, when load and photovoltaic output are higher, the imbalance rate decreased from 3.4% to 1.5%. The segmented phase energy modulation mechanism can flexibly adjust the segmented power of the load and distributed power sources, resulting in a more balanced three-phase power distribution and improved power quality.

[0138] This invention also demonstrates advantages in addressing current over-limit issues. During the high-output period from 12:00 to 14:00, the branch current over-limit rate decreased from 7.2% to 1.7%; during the most severe period from 14:00 to 16:00, the over-limit rate decreased from 8.0% to 1.8%. By introducing a physical constraint graph neural ODE model into the prediction and scheduling stages, the prediction results can automatically meet the current capacity constraints, ensuring the physical feasibility of the scheduling scheme and avoiding the risk of over-limits caused by prediction distortion during operation.

[0139] Regarding equipment operation and user experience, the method of this invention effectively reduces the switching frequency of switching equipment and user voltage complaints. Between 2 PM and 4 PM, the traditional switching frequency reaches 15 times per hour, while this method reduces it to 9 times per hour, a reduction of 40%. The number of user voltage complaints has also decreased from 5 times per week to about 2 times per week. This is attributed to the introduction of a dual-time-domain event triggering mechanism. The minute-level slow loop ensures overall operational stability, while the second-level fast loop responds promptly to voltage fluctuations and generates control commands, reducing equipment wear and extending equipment lifespan while ensuring safety.

[0140] 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 smart power distribution method, characterized in that, include: Collect multi-source data from the power distribution network, process the multi-source data, construct an input sample set, and segment the power of high-power loads and distributed power sources to generate a segmented phase mapping matrix. The input sample set is processed based on the Physically Constrained Graph Neural (ODE) model. The ODE model introduces a piecewise phase mapping matrix in the input layer to jointly encode the electrical features of nodes and the piecewise power features. In the process of ordinary differential equation evolution, dynamic coupling constraint terms based on piecewise power features are added to form an extended prediction result set. The extended prediction result set is input into the segmented phase energy modulation module, and cross-phase time slot scheduling is performed using the segmented phase mapping matrix. Part of the power is allocated in different phases and time slots, and a segmented phase scheduling scheme is generated by combining the energy storage output and the power distribution topology. The segmented phase scheduling scheme is controlled based on a dual-time-domain event triggering mechanism. The slow loop performs rolling optimization within a minute-level cycle to generate a scheduling scheme, while the fast loop performs rapid adjustment within a second-level cycle based on the physical security domain triggering condition, the segmented residual triggering condition, and the dynamic evolution triggering condition of the security domain boundary to generate event-triggered control instructions. The dispatching scheme and event-triggered control commands are sent to the power distribution controller, energy storage control device and phase energy modulator to perform power distribution and phase switching operations. The feedback data collected after execution is compared with the extended prediction result set to update the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold, forming an adaptive closed loop of prediction-optimization-control.

2. The intelligent power distribution method according to claim 1, characterized in that, The multi-source data includes node voltage data, node current data, distributed power output data, energy storage device state of charge data, flexible load power data, environmental meteorological data, line switch status data, and transformer temperature rise data. The processing of the multi-source data includes filling in missing values, removing outliers, normalizing the data, and aligning the timestamps.

3. The intelligent power distribution method according to claim 1, characterized in that, The generation of the segmented phase mapping matrix refers to segmenting the power of high-power loads and distributed power sources according to a preset power threshold, and representing the mapping relationship between each segment of power and the three-phase power supply phase in matrix form. In the matrix, the element corresponding to each segment of power and a single phase has a value of 1, and the element corresponding to a non-corresponding phase has a value of 0, thus forming a segmented phase mapping matrix.

4. The intelligent power distribution method according to claim 1, characterized in that, The formation of the extended prediction result set includes: Based on the distribution network topology and phase sequence information, a three-phase phase subgraph is established. The node set is copied into three groups of sub-nodes according to phases A, B, and C. The edge set is established into edge channels according to in-phase connection and cross-phase coupling. A physical constraint graph neural ODE model is constructed, which includes an input binding encoder, a continuous-time evolution unit, a physical consistency corrector, and a feasible region projector. An input-bound encoder is constructed, which uses node voltage data and node current data as node electrical features, and uses the segmented high-power load power and distributed power source power as segmented power features. The segmented phase mapping matrix and time slice identifier are then used for joint encoding. Establish a continuous-time evolution unit to enable the state channel and the dual embedded channel to evolve in parallel. The state channel updates the node voltage and branch current, while the dual embedded channel updates the voltage sensitivity dual quantity to active power change and the voltage sensitivity dual quantity to reactive power change. According to the segmented power characteristics, the evolution rate and aggregation weight are adjusted on the affected nodes and adjacent edges. The unbalance information is transmitted between the three-phase phase sub-diagrams according to the inter-phase coupling parameters. After each evolution step, the physical consistency corrector and feasible region projector are executed sequentially to correct and project the intermediate state according to the power flow constraints, thermal limits, node voltage range and branch current upper limit. The safety margin is calculated by the side channel based on the distance between the current state and the constraint boundary. After completing the continuous time evolution of the set prediction period, an extended prediction result set is output. The extended prediction result set includes the voltage prediction of each node at each prediction time, the current prediction of each branch at each prediction time, the sensitivity dual of voltage to active power change, the sensitivity dual of voltage to reactive power change, and the dynamic evolution of safety margin.

5. The intelligent power distribution method according to claim 1, characterized in that, The generation of the segmented phase scheduling scheme includes: The segmented phase energy modulation module receives the extended prediction result set, the segmented phase mapping matrix and the segmented power components, and establishes a time slice sequence for cross-phase time slot scheduling according to a fixed time slice division. Based on the extended prediction result set, a priority sequence of phase and time slices is generated. According to the deterministic sorting rules of safety margin priority, low sensitivity duality priority and high line thermal limit margin priority, the segmented power components are initially allocated. The mapped segmented power components are activated in the corresponding phase channel, and the unmapped segmented power components are set to zero in the phase channel to form an initial allocation table. The initial allocation table is subjected to segmented sliding and shaping processing. The segmented power components are finely slid and rearranged in adjacent time slots. The upper limit of three-phase imbalance drift and the threshold of intra-segment jitter suppression are set. Through sliding and rearrangement, the three-phase imbalance and power fluctuation amplitude are reduced. The allocation result formed by the segmented power components and the segmented phase mapping matrix in different phases and time slots is defined as cross-phase time slot allocation. Based on the extended prediction result set, the switching safety window is determined. Within the switching safety window, a phase switching sequence is synthesized. Constraints are set on the maximum switching frequency, minimum holding time, and minimum interval between adjacent switching. Power transition compensation is performed before and after switching in combination with the energy storage output boundary to form a segmented switching subsequence. A feasibility check is performed on the result synthesized by the segmented sliding shaping and phase switching sequence. The feasibility check includes the upper and lower limits of node voltage, the upper limit of branch current, the power balance constraint, and the consistency constraint of segmented phase mapping. After the check is passed, the segmented phase scheduling scheme is output.

6. The intelligent power distribution method according to claim 1, characterized in that, The generated event triggers control instructions, including: Set minute-level slow loop period and second-level fast loop period, receive segmented phase scheduling scheme and extended prediction result set, define dynamic evolution of safety domain boundary based on safety margin changes in consecutive adjacent time moments, and set safety margin threshold, evolution threshold and segmented residual threshold. Within each minute-level slow loop cycle, rolling optimization is performed based on the segmented phase scheduling scheme and the extended prediction result set to generate a scheduling scheme covering the prediction period, and satisfying the upper and lower limits of node voltage, the upper limit of branch current, power balance and segmented phase mapping consistency constraints. Within each second-level fast loop cycle, the current safety margin and the dynamic evolution of the current safety domain boundary are calculated based on the real-time measurement and extended prediction result set. When the current safety margin is less than the safety margin threshold, or when the decrease in the dynamic evolution of the current safety domain boundary exceeds the evolution threshold, the physical safety domain triggering condition is determined to be met. Within each second-level fast loop cycle, the difference between the actual cross-phase time slot allocation and the corresponding allocation in the segmented phase scheduling scheme is compared to obtain the segmented residual. When the segmented residual is greater than the segmented residual threshold, it is determined that the segmented residual triggering condition is met. The deviation between the current dynamic evolution of the security domain boundary and the reference evolution given by the extended prediction result set is compared. When the deviation exceeds the evolution threshold, it is determined that the dynamic evolution triggering condition of the security domain boundary is met. When any triggering condition is determined to be met, an event-triggered control instruction is generated. The event-triggered control instruction includes the modification of the energy storage output plan, the incremental adjustment of the phase and time slice allocation table for cross-phase time slot allocation, and the insertion, deletion or timing rearrangement of segmented switching sub-sequences.

7. The intelligent power distribution method according to claim 1, characterized in that, The power allocation and phase switching operations include: Receive scheduling schemes and event-triggered control instructions, integrate event-triggered control instructions with scheduling schemes to obtain the corrected cross-phase time slot allocation table, segmented switching subsequence and energy storage output plan; A set of control instructions for phases and time slots is generated based on the revised cross-phase time slot allocation table. The set of control instructions is received and executed by the phase energy modulator to drive each phase to complete the allocation and switching of segmented power within the corresponding time slot. An energy storage control sequence is generated based on the revised energy storage output plan. The energy storage control sequence includes charging control commands, discharging control commands, and holding control commands, which are received and executed by the energy storage control device to balance electrical energy and suppress real-time power fluctuations. The control command set for phase and time slices, the modified segmented switching subsequence, and the energy storage control sequence are sent to the power distribution controller, energy storage control device, and phase energy modulator to execute the corresponding power distribution, phase switching, and energy storage regulation operations.

8. The intelligent power distribution method according to claim 1, characterized in that, The updating of the physical constraint graph neural ODE model, the piecewise phase mapping matrix, and the event triggering threshold includes: The feedback data includes real-time values ​​of node voltage, real-time values ​​of branch current, actual executed values ​​of energy storage output, actual executed sequence of phase switching, and actual completed results of power allocation, and is aligned hourly with the extended prediction result set. By comparing the feedback data with the extended prediction result set, node voltage error, branch current error, energy storage output deviation and phase switching deviation are generated to form an error sequence. The physical constraint graph neural ODE model is updated based on the error sequence. The update includes adjusting the feature weights of the input binding encoder, correcting the aggregation weights of the continuous-time evolution unit and the coupling coefficients of the dynamic coupling constraint. The segmented phase mapping matrix is ​​corrected based on the feedback data. While maintaining the consistency of the correspondence between segmented power components and phase channels, the mapping results of some segmented power components are updated according to the actual execution situation. Unmapped segmented power components are set to zero in the corresponding phase channels. The event trigger threshold is dynamically adjusted based on the error sequence and the operation of the safety domain, and the updated event trigger threshold is written into the dual time-domain event triggering mechanism to form an adaptive closed loop of prediction-optimization-control.

9. An intelligent power distribution system, comprising executing the intelligent power distribution method according to any one of claims 1 to 8, characterized in that, Includes the following modules: The data acquisition and processing module is used to acquire multi-source data from the power distribution network, construct an input sample set, and generate a segmented phase mapping matrix. The predictive modeling module is used to generate an extended prediction result set based on the physical constraint graph neural ODE model. The segmented phase energy modulation module is used to receive the extended prediction result set, perform cross-phase time slot scheduling using the segmented phase mapping matrix, and generate a segmented phase scheduling scheme. The dual-time-domain event triggering control module is used to execute rolling optimization to generate scheduling schemes in the minute-level slow loop and generate event triggering control instructions in the second-level fast loop. The execution control module is used to issue scheduling schemes and event-triggered control commands to the power distribution controller, energy storage device and phase energy modulator, and to perform power distribution and phase switching operations. The adaptive update module is used to collect feedback data, compare it with the extended prediction result set, and update the physical constraint graph neural ODE model, the segmented phase mapping matrix, and the event triggering threshold.

Images