Energy storage flexible dynamic current sharing method fusing multi-physics manifold graph and intention deduction and power distribution cabinet

Abstract

The application discloses a method for energy storage flexible dynamic current sharing combining multi-physical flow graph and intention deduction and a power distribution cabinet. The method collects operation data of each energy storage branch, a PCS or a DC / DC module, a direct current busbar, a contactor, a fuse, a wind channel temperature zone and a load side power distribution circuit in the energy storage power distribution cabinet, constructs a multi-physical flow graph containing resistance coupling, heat diffusion coupling, aging coupling and topology switch coupling, and obtains a bearable current flow graph coordinate of each branch; according to an EMS scheduling instruction, photovoltaic prediction, a load curve, a standby power strategy and an on-grid and off-grid switching signal, the operation intention in a future control period is deduced; then, in combination with a heat margin, an aging margin, an impedance margin and a future task intensity, a flexible current sharing target is generated, and a DC / DC current given value, a PCS droop coefficient, a contactor opening and closing priority, a fan rotating speed, a current limiting threshold and a fault bypass strategy are output. The power distribution cabinet can correct online a current flow residual error and a temperature rise residual error to adapt to the state change in the cabinet.

Description

Technical Field

[0001] This invention relates to the field of energy storage distribution cabinet control technology, specifically to a flexible dynamic current sharing method for energy storage that integrates multi-physical manifold diagrams and intent deduction, and a distribution cabinet thereof. Background Technology

[0002] Commercial and industrial energy storage distribution cabinets typically include multiple energy storage branches, PCS or DC / DC converter modules, DC busbars, contactors, fuses, duct temperature control structures, and load-side power distribution circuits. During peak shaving and valley filling, photovoltaic power consumption, backup power switching, grid connection / off-grid conversion, and impact load startup, each energy storage branch needs to share the charging and discharging power.

[0003] Existing current sharing controls are mostly adjusted based on branch current, SOC difference, fixed droop coefficient or preset current limiting threshold, which can reduce the difference in branch current to a certain extent. However, their control basis is usually focused on the electrical parameters themselves, and it is difficult to simultaneously reflect the influence of battery internal resistance, SOH, local temperature rise of busbar, contactor voltage drop, fuse heating, differences in air duct in cabinet, humidity, insulation impedance and load-side switch status on the branch carrying capacity.

[0004] Meanwhile, existing strategies mostly rely on feedback control based on the current state, making it difficult to combine EMS dispatch instructions, photovoltaic forecasts, load curves, backup power strategies, and grid connection / off-grid signals to predict future operational intentions. When the energy storage cabinet is about to enter a high-rate discharge, reverse charging, impact load response, or fire-fighting load reduction state, if the current is still distributed according to average current sharing or a fixed coefficient, local branches are prone to bearing unsuitable current loads for a long time, which can lead to local overheating, uneven aging, PCS current limiting action, and a decrease in cabinet safety margin. Therefore, it is necessary to provide a flexible dynamic current sharing scheme for energy storage that can integrate multiple physical states within the cabinet and future operational intentions. Summary of the Invention

[0005] To address the shortcomings of existing energy storage distribution cabinets, which primarily rely on current current, SOC difference, fixed droop coefficient, or preset current limiting thresholds for current sharing during parallel operation of multiple branch energy storage units, making it difficult to comprehensively reflect changes in the electrical, thermal, aging, insulation, and topological states within the cabinet, and to proactively adjust branch load distribution in conjunction with future scheduling tasks, this invention provides a flexible dynamic current sharing method and distribution cabinet that integrates multi-physical manifold diagrams and intent extrapolation. This method treats the energy storage branches, PCS or DC / DC modules, DC busbars, contactors, fuses, duct temperature zones, and load-side distribution circuits within the energy storage distribution cabinet as a unified control object. It expresses the cross-physical field coupling relationships within the cabinet through multi-physical manifold diagrams and expresses the intensity of operational tasks within the future control window through intent extrapolation, thereby forming a flexible current sharing target that does not require absolute equality of current in each branch.

[0006] One aspect of this invention provides a flexible dynamic current sharing method for energy storage that integrates multi-physics manifold diagrams and intent extrapolation. This method collects operational data from multiple energy storage branches, PCS or DC / DC modules, DC busbars, contactors, fuses, duct temperature zones, and load-side power distribution circuits within the energy storage distribution cabinet while it is in operation. This operational data characterizes the electrical load-bearing status, thermal status, aging status, insulation status, internal heat dissipation status, and load-side topology status of each energy storage branch, ensuring that subsequent current sharing control no longer relies solely on a single current difference or SOC difference.

[0007] Specifically, operational data can include branch current, terminal voltage, SOC, SOH, estimated internal resistance, and branch temperature for each energy storage branch. It can also include busbar node temperature and infrared temperature rise at busbar connection points of the DC busbar. Furthermore, it can include contact voltage drop, fuse temperature, fan outlet wind speed, cabinet temperature zone, cabinet humidity, insulation impedance, PCS power command, DC / DC operating status, load-side circuit breaker status, and grid-connected / off-grid switch status. The above operational data is time-aligned according to the same control cycle and, after outlier removal and unit normalization, forms the branch status characteristics, cabinet thermal status characteristics, topology switch status characteristics, and load-side power distribution status characteristics corresponding to the current control cycle. When sensor data for a branch is missing or exceeds the physical allowable range, alternative status values ​​can be generated using valid data from the previous control cycle for that branch and the changing trends of adjacent branches in the same busbar section. This reduces the data reliability of that branch, ensuring that subsequent capacity assessments reflect data quality differences.

[0008] After collecting operational data, a multi-physics manifold is constructed based on this data. The multi-physics manifold uses electrical components, thermal management components, and load branches within the cabinet as nodes, and resistive coupling, thermal diffusion coupling, aging coupling, and topology switching coupling as edges. Through this graph structure, information originally scattered across battery management, PCS control, busbar temperature measurement, contactor detection, fan control, and load circuit breaker status within the energy storage distribution cabinet is organized into a unified graphical model. This allows the relationships between branch current deviation, localized temperature rise, contact resistance changes, duct heat dissipation differences, and topology switching to be collectively used for current sharing decisions.

[0009] Specifically, when constructing a multi-physics manifold, battery clusters, energy storage branch access points, DC / DC modules, PCS inputs, busbar connection points, contactors, fuses, cabinet temperature zones, fan outlets, and load branches can be set as graph nodes. The connection relationships between energy storage branches and busbars, the conductive paths between busbar connection points, and the series paths of contactors and fuses are set as resistive coupling edges. The heat transfer relationships between battery clusters, busbar connection points, contactors, fuses, cabinet temperature zones, and fan outlets are set as thermal diffusion coupling edges. The correlation between branch current ratio, temperature rise history, SOC operating range, and SOH variation trends are set as aging coupling edges. The influence of contactors, circuit breakers, PCS operating modes, and DC / DC switching states on current paths is set as topology switching coupling edges. Node characteristics of graph nodes include branch current, terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, busbar infrared temperature rise, contactor voltage drop, fan outlet wind speed, humidity, insulation impedance, and load-side circuit breaker status.

[0010] Furthermore, the current-carrying manifold coordinates of each energy storage branch are calculated based on the multi-physics manifold diagram. These current-carrying manifold coordinates represent the relative current-carrying capacity of the corresponding energy storage branch under the current cabinet condition and within a short future window. This coordinate is not a fixed rated current, nor is it a simple weighting determined by the State of Charge (SOC), but rather formed by the combined electrical manifold, thermal manifold, aging manifold, and topological manifold. This processing method enables branches with higher thermal margins, lower impedances, better State of Harmony (SOH), and better air-cooling conditions to achieve higher current-carrying capacity, while branches with higher temperature rises, abnormal contactor voltage drops, insufficient insulation margins, or severe aging achieve lower current-carrying capacity.

[0011] Specifically, electrical manifold coordinates can be calculated based on the resistive coupling edge and branch electrical state characteristics; thermal manifold coordinates can be calculated based on the thermal diffusion coupling edge and cabinet thermal state characteristics; aging manifold coordinates can be calculated based on the aging coupling edge and branch SOH variation trend; and topological manifold coordinates can be calculated based on the topology switch coupling edge and load-side power distribution state characteristics. These electrical, thermal, aging, and topological manifold coordinates are then input into a fusion function to obtain the current-carrying manifold coordinates for each energy storage branch. The fusion function can employ weighted fusion, gated fusion, or graph attention fusion methods, and is constrained by the branch rated current, busbar allowable temperature, contactor allowable voltage drop, and insulation impedance lower limit to ensure the output matches the current safety boundary within the cabinet. For scenarios without deploying deep learning models, equivalent current-carrying manifold coordinates can also be obtained using physically weighted graph embedding and constraint optimization methods.

[0012] After obtaining the manifold coordinates of the current-carrying capacity, the system acquires EMS scheduling instructions, photovoltaic forecasts, load curves, backup power strategies, and grid-connected / off-grid switching signals, and deduces the operational intent within the future control window. This operational intent characterizes the charging and discharging tasks, task durations, safety priorities, and task intensity that the energy storage distribution cabinet will execute in several future control cycles. Through this processing method, current sharing control is no longer limited to feedback adjustment at the current moment, but can adjust the current distribution and safety margin of each energy storage branch in advance before tasks such as high-rate discharge, photovoltaic charging, backup power voltage maintenance, black start preparation, impact load response, or fire-fighting load reduction occur.

[0013] Specifically, the total power change trend within the future control window is determined based on EMS dispatch instructions; the charging and discharging direction and power duration are determined based on photovoltaic forecasts and load curves; the branch redundancy and SOC range to be retained are determined based on the backup power strategy; the PCS operating mode change is determined based on grid connection / off-grid switching signals; and short-term high-rate discharge demand is determined based on the load-side circuit breaker status and impact load start signals. The future control window is divided into multiple continuous control cycles, and an operation intention label, target power, duration, safety priority, and task intensity are generated for each control cycle. The operation intention label can be one of the following: peak shaving discharge, photovoltaic charging, low power maintenance, reverse charging, backup power voltage protection, black start preparation, impact load response, or fire load reduction. When a fire alarm, insulation abnormality, or busbar over-temperature signal is triggered, fire load reduction or fault derating is determined as the operation intention with the highest safety priority.

[0014] After obtaining the manifold coordinates of the current carrying capacity and the operational intent, a flexible current sharing target for each energy storage branch is generated based on these coordinates. This flexible current sharing target does not require all energy storage branches to have completely equal current; instead, it allocates current among the total power demand of the energy storage distribution cabinet, the current carrying capacity of the branch, and the intensity of future operational tasks. Therefore, branches with high health, low temperature rise, low impedance, and good heat dissipation can safely carry more current, while branches with localized temperature rise, high contactor voltage drop, low state of equilibrium (SOH), insufficient insulation margin, or poor heat dissipation in the corresponding wind zone are automatically deloaded.

[0015] Specifically, the comprehensive carrying capacity factor C_i(t) of each energy storage branch can be calculated based on its thermal margin, aging margin, impedance margin, air cooling margin, insulation margin, and future intended matching margin. The comprehensive carrying capacity factor of a branch can be expressed as: ; Where I_total(t) represents the total charging and discharging current command of the energy storage distribution cabinet at time t; C_i(t) represents the comprehensive carrying factor of the i-th energy storage branch at time t; I_i,max represents the safe current limit value for the i-th energy storage branch to participate in current distribution in the current control cycle; C_j(t) represents the comprehensive carrying factor of the j-th energy storage branch currently in operation; I_j,max represents the safe current limit value for the j-th energy storage branch currently in operation to participate in current distribution in the current control cycle; Σ_j[C_j(t)·I_j,max] represents the summation of C_j(t)·I_j,max for all energy storage branches currently in operation that have not been bypassed, fault-locked, or disconnected by contactors. For the discharge condition, the target current is used to determine the current released by each branch to the DC busbar; for the charging condition, the target current is used to determine the current absorbed by each branch from the DC busbar, and is constrained by the SOC upper limit, charging rate and battery temperature.

[0016] Furthermore, thermal margin According to Calculation, where T_lim represents the upper limit of the allowable temperature of the branch or busbar, and T_i represents the current temperature of the branch or the corresponding busbar node temperature. This represents the predicted temperature rise within the future control window, derived from the predicted current and air-cooling conditions. T_ref represents the reference temperature. Impedance margin. According to Calculation. R_ref represents the reference impedance, R_i represents the branch internal resistance, R_bus_i represents the busbar equivalent impedance, R_con_i represents the contactor equivalent contact resistance, and ε represents the correction to prevent the denominator from being zero. When the operating intention is to respond to impact loads, increase the future intended matching margin of branches with low temperature, low resistance, and high SOH; when the operating intention is to maintain backup power and voltage, reduce the current discharge target of branches selected as subsequent emergency redundancy branches; when the operating intention is to reduce fire load, set the comprehensive carrying capacity factor of branches with abnormal temperature zones or abnormal insulation to zero or a safe derating value.

[0017] After generating the flexible current sharing target, the system outputs the DC / DC current setpoint, PCS droop coefficient, branch contactor opening and closing priority, fan speed, current limiting threshold, and fault bypass strategy according to the target. For branches equipped with DC / DC modules, the DC / DC current setpoint ensures that the corresponding branch participates in charging and discharging according to the target current. For scenarios where the PCS participates in DC-side or AC-side power distribution, the PCS droop coefficient is adjusted based on the carrying capacity and future intentions of each branch to ensure that the PCS current limiting action is consistent with the branch's flexible current sharing target. The branch contactor opening and closing priority determines the branch switching sequence, the fan speed pre-cools areas that will soon bear higher currents or experience higher local temperature rises, the current limiting threshold dynamically adjusts the safety boundaries of branches and busbar nodes, and the fault bypass strategy redistributes current or triggers overall cabinet derating when a branch malfunctions.

[0018] Furthermore, before issuing control quantities, safety interlocking verification can be performed on the flexible current sharing target and the control quantities. The safety interlocking verification includes the branch maximum current, branch terminal voltage, busbar temperature rise, contactor voltage drop, fuse temperature, insulation resistance, PCS current limiting status, DC / DC operating status, fan status, load-side circuit breaker status, and fire alarm signals. When the control quantities meet safety constraints, they are issued to the DC / DC module, PCS, contactors, fan controller, and load-side power distribution circuit. When the control quantities do not meet safety constraints, derating, bypassing, prohibiting closing, PCS power limiting, forced fan speed increase, or fault interlocking strategies are adopted, and the adjusted available carrying capacity is fed back to the EMS or cabinet control system.

[0019] After the control input is executed, the current deviation residual and temperature rise residual are collected, and the edge weights of the multi-physics manifold are updated based on these residuals. The current deviation residual reflects the difference between the target current and the measured current, while the temperature rise residual reflects the difference between the predicted temperature rise and the measured temperature rise. Through residual updates, the system can adapt to changes in contactor contact states, loose busbar connections, duct blockages, fan performance degradation, battery internal resistance changes, and cabinet environment changes, allowing the multi-physics manifold to update with the long-term operating status of the energy storage distribution cabinet.

[0020] Specifically, after execution, calculations can be performed. , ,as well as .in, I_i,meas(t) represents the current residual of the i-th energy storage branch, and I_i,meas(t) represents the measured current of the i-th energy storage branch. ΔT_i,meas(t) represents the residual temperature rise of the i-th energy storage branch or the corresponding busbar node, and ΔT_i,meas(t) represents the measured temperature rise. ΔV_con_i,meas(t) represents the measured contactor voltage drop, and ΔV_con_i,pred(t) represents the predicted contactor voltage drop based on the branch current and the contactor's equivalent contact resistance. When the current residual is consistently large, the weight of the corresponding resistance coupling side is increased or the reliability of the corresponding execution channel is decreased; when the temperature rise residual is consistently large, the weight of the thermal diffusion coupling side is increased or the thermal margin of the corresponding branch is decreased; when the contactor voltage drop residual is consistently large, the estimated value of the contactor's equivalent contact resistance is increased, and the opening and closing priority and carrying capacity of that branch are decreased.

[0021] Optionally, the graph edge weights or margin parameters can be updated using a restricted recursive method, expressed as follows: Where θ represents the graph edge weights or margin parameters, η represents the learning rate, L_res represents the residual loss consisting of the current residual, temperature rise residual, and contactor voltage drop residual, and θ_min and θ_max represent the parameter safety boundaries. The above online updates do not exceed the hard constraints set by the safety interlocking module. When the residual exceeds the safety threshold, derating or bypass control is prioritized instead of further increasing the model update magnitude.

[0022] Another aspect of the present invention provides a flexible dynamic current sharing distribution cabinet for energy storage that integrates multi-physics manifold diagrams and intent extrapolation. The distribution cabinet includes a cabinet body, energy storage branch access units, a DC busbar, a PCS access unit, a DC / DC or current sharing execution module, contactors, fuses, a load-side distribution circuit, branch current sensors, branch voltage sampling units, busbar temperature sensors, contactor voltage drop sampling units, insulation detection units, duct temperature control modules, communication modules, and an edge computing controller. The energy storage branch access units are connected to the DC busbar via contactors and fuses. The DC / DC or current sharing execution module is located between the energy storage branch access units and the DC busbar. The PCS access unit is connected to the DC busbar and used for power exchange with the AC-side distribution system. The load-side distribution circuit is connected to the output side of the energy storage distribution cabinet via a circuit breaker. The edge computing controller is used to execute the above-mentioned flexible dynamic current sharing method for energy storage, and sends control commands to the DC / DC or current sharing execution module, PCS access unit, contactor, duct temperature control module and load-side power distribution circuit according to the calculation results.

[0023] Specifically, the edge computing controller includes a multi-source data acquisition module, a multi-physics manifold construction module, a branch carrying capacity assessment module, an intent deduction module, a flexible current sharing optimization module, a safety interlocking module, and an online self-calibration module. The multi-source data acquisition module acquires branch current, terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, busbar infrared temperature rise, contactor voltage drop, fan outlet wind speed, cabinet humidity, insulation impedance, PCS power command, and load-side circuit breaker status. The multi-physics manifold construction module establishes a multi-physics manifold diagram including resistive coupling edges, thermal diffusion coupling edges, aging coupling edges, and topology switch coupling edges. The branch carrying capacity assessment module calculates the current-carrying manifold coordinates of each energy storage branch based on the multi-physics manifold diagram. The intent deduction module generates operating intents based on EMS dispatch commands, photovoltaic forecasts, load curves, backup power strategies, and grid-connected / off-grid switching signals. The flexible current sharing optimization module generates target currents and control quantities for each energy storage branch. The safety interlocking module is used to perform derating, bypass, or prohibition of closing control when there is branch overcurrent, busbar overtemperature, abnormal contactor voltage drop, or abnormal insulation impedance. The online self-calibration module is used to update the graph edge weights based on the bias current residual, temperature rise residual, and contactor voltage drop residual.

[0024] Furthermore, the duct temperature control module includes a fan, duct, fan outlet wind speed sensor, and cabinet interior temperature zone sensor. The duct is arranged along multiple energy storage branches, the DC busbar, and the area where the PCS access unit is located. The fan outlet wind speed sensor provides cooling margin calculation data to the edge computing controller, and the cabinet interior temperature zone sensor provides thermal manifold coordinate calculation data to the edge computing controller. A contactor voltage drop sampling unit is connected to both ends of the branch contactor contacts and collects the contact voltage drop when the contactor is closed. The edge computing controller calculates the equivalent contact resistance of the contactor based on the contact voltage drop and branch current. An insulation detection unit detects the DC-side insulation impedance and outputs the insulation status to the safety interlocking module. A communication module communicates with the EMS, PCS, DC / DC module, fan controller, and load-side circuit breaker, enabling the energy storage distribution cabinet to synchronously adjust branch current, PCS droop coefficient, contactor opening and closing priority, fan speed, current limiting threshold, and fault bypass strategy according to the flexible current sharing target.

[0025] In summary, this application first collects the status data of multiple energy storage branches, PCS or DC / DC modules, DC busbars, contactors, fuses, duct temperature zones, and load-side power distribution circuits during the operation of the energy storage distribution cabinet, and constructs a multi-physical manifold diagram that can express resistive coupling, thermal diffusion coupling, aging coupling, and topology switching coupling. Then, based on this multi-physical manifold diagram, the manifold coordinates of the current carrying capacity of each branch are obtained. At the same time, the operating intention within the future control window is deduced by combining EMS dispatch instructions, photovoltaic forecasts, load curves, backup power strategies, and grid connection / disconnection switching signals. Subsequently, a flexible current sharing target is formed based on the branch carrying capacity and operating intention, and this target is converted into DC / DC current setpoint, PCS droop coefficient, contactor opening and closing priority, fan speed, current limiting threshold, and fault bypass strategy. Finally, the graph edge weights are updated using the bias current residual, temperature rise residual, and voltage drop residual after execution, forming a closed-loop control process that can be corrected according to the changes in the cabinet's status.

[0026] Compared to existing current sharing schemes based on current current, SOC difference, or fixed droop coefficient, this application extends the current sharing control object from a single battery branch to multiple branch energy storage units, PCS or DC / DC modules, DC busbars, contactors, fuses, duct temperature zones, and load-side distribution circuits within the energy storage distribution cabinet. This allows the current sharing target to be simultaneously constrained by electrical condition, thermal condition, aging condition, insulation condition, air-cooling condition, and topology condition. Since the branch comprehensive carrying capacity factor is jointly determined by thermal margin, aging margin, impedance margin, air-cooling margin, insulation margin, and future intended matching margin, this application avoids neglecting the problems of local temperature rise, contactor voltage drop, and differences in duct airflow within the cabinet when power is distributed solely based on current averaging or SOC averaging.

[0027] Secondly, this application uses intent extrapolation to transform EMS dispatch instructions, photovoltaic forecasts, load curves, backup power strategies, grid connection / off-grid switching signals, and impact load initiation signals into operational intents within the future control window. This enables the energy storage distribution cabinet to adjust branch current distribution before tasks such as high-rate discharge, reverse charging, backup power voltage maintenance, black start preparation, impact load response, and fire load reduction occur. When future operational tasks require higher instantaneous discharge capacity, branches with lower temperatures, lower resistance, and higher state of equilibrium (SOH) can handle a higher proportion of current; when future tasks require maintaining backup power capacity, some healthy branches can be retained as emergency redundancy; when safety anomalies are triggered, abnormal branches can be derated or bypassed. Thus, a correspondence is established between branch current distribution and the intensity of future tasks.

[0028] This application also converts the flexible current sharing target into multiple types of executable control quantities, so that current sharing control is no longer limited to the DC / DC current setpoint, but can synchronously adjust the PCS droop coefficient, branch contactor switching priority, fan speed, current limiting threshold, and fault bypass strategy. This approach enables electrical current sharing, thermal management, and safety interlocking to operate in coordination under the same control logic. When a branch experiences abnormal contactor voltage drop, local busbar overheating, or abnormal insulation impedance, the system can reduce the target current and switching priority of that branch, while simultaneously increasing the fan speed in the corresponding wind zone or executing bypass control, thereby reducing the possibility of the local anomaly continuing to expand.

[0029] Finally, this application performs online self-correction of the multi-physical manifold diagram using offset current residuals, temperature rise residuals, and contactor voltage drop residuals, enabling the graph edge weights and margin parameters to be updated as the actual operating conditions change. For long-term operating energy storage distribution cabinets, contactor contacts, cable connections, busbar connections, air duct unobstructedness, fan performance, and battery branch internal resistance may all change. Through residual-driven constrained updates, this application can adjust the resistive coupling edge, thermal diffusion coupling edge, and load-bearing capacity parameters without breaking safety interlocking constraints, allowing current sharing control to continuously adapt to changes in cabinet conditions. Attached Figure Description

[0030] To more clearly illustrate the technical solution of this application, the accompanying drawings are briefly described below. The drawings are used to illustrate the processing flow, device structure, data relationships, and application scenarios of this application, and do not constitute a limitation on the physical location of each module, the number of interfaces, the wiring method, or the scale of the illustrations.

[0031] Figure 1 This is a schematic diagram of the overall process of the flexible dynamic current sharing method for energy storage provided in an embodiment of the present invention.

[0032] Figure 2 This is a schematic diagram of the system architecture of the flexible dynamic current sharing distribution cabinet for energy storage provided in an embodiment of the present invention.

[0033] Figure 3 This is a schematic diagram illustrating the construction relationship of a multiphysics manifold graph provided in an embodiment of the present invention.

[0034] Figure 4 This is a schematic diagram illustrating the deduction of operational intent and the generation of future control windows, provided for an embodiment of the present invention.

[0035] Figure 5 This is a schematic diagram of the process for generating flexible flow sharing targets and control quantities, provided in an embodiment of the present invention.

[0036] Figure 6 This is a schematic diagram of the online self-correction closed loop of the bias flow residual, temperature rise residual, and graph edge weight provided in the embodiments of the present invention.

[0037] Figure 7 This is a schematic diagram of the application structure of the industrial and commercial energy storage distribution cabinet in a photovoltaic-storage load scenario, as provided in an embodiment of the present invention.

[0038] Figure 8 This is a schematic diagram comparing the effects of flexible dynamic flow equalization control provided in an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] The embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the described embodiments are used to explain the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Where there is no conflict, the technical features in the following embodiments can be combined with each other. The parameters such as control cycle, threshold, weight, temperature, current, and number of branches involved in the following embodiments can be adjusted according to the rated power of the energy storage distribution cabinet, DC voltage level, number of energy storage branches, duct structure, PCS type, DC / DC module capacity, and on-site operation strategy. Example 1

[0041] This embodiment provides a flexible dynamic current sharing method for energy storage that integrates multi-physics manifold graphs and intent extrapolation. This method can be executed by an edge computing controller within the energy storage distribution cabinet, or it can be executed collaboratively by the local controller of the energy storage distribution cabinet and the EMS. For example... Figure 1 As shown, this method starts with the acquisition of multi-source data in the cabinet, and then proceeds to the construction of a multi-physical manifold graph, the calculation of the manifold coordinates of the branch current carrying capacity, the deduction of future operation intentions, the generation of flexible current sharing targets, the verification of safety interlocking and the issuance of control quantities. Finally, the graph edge weights are updated by executing residual feedback to form a continuous control closed loop.

[0042] When the energy storage distribution cabinet is in operation, the edge computing controller collects operating data within the cabinet according to a preset control cycle. This operating data includes branch current, terminal voltage, SOC, SOH, estimated internal resistance, and branch temperature for each energy storage branch; busbar node temperature and infrared temperature rise at busbar connection points of the DC busbar; contact voltage drop in the contactor's closed state; fuse temperature; fan outlet wind speed; cabinet internal temperature zone; cabinet internal humidity; DC side insulation impedance; PCS power command; DC / DC operating status; load-side circuit breaker status; and grid-connected / off-grid switch status. For energy storage branches with independent DC / DC modules, the edge computing controller can also collect the corresponding DC / DC module's current setpoint, actual output current, current limiting status, and fault status. For energy storage branches without independent DC / DC modules, the edge computing controller can collect branch contactor status, branch current limiting threshold, and PCS-side power distribution status.

[0043] After acquiring operational data, the edge computing controller performs time alignment on data from different sources. Since the refresh cycles of branch current sampling, temperature sampling, insulation detection, PCS communication, and EMS scheduling commands may differ, the edge computing controller uses the current sharing control cycle as a benchmark, writing valid data within the same control cycle into the state buffer and marking data exceeding the allowable delay as low-confidence data. For sampled values ​​of physical quantities significantly exceeding the allowable range, the edge computing controller does not directly use them as node features in calculations. Instead, it combines the valid value from the previous control cycle, the changing trends of adjacent branches, and sensor status to generate alternative state values. These alternative state values ​​are only used to maintain control continuity and reduce the confidence level of the corresponding branch's carrying capacity in the current cycle.

[0044] After data preprocessing, the edge computing controller constructs a multi-physics manifold graph based on the cabinet's internal topology and current operating data. For example... Figure 3 As shown, the nodes in the multiphysics manifold include battery cluster nodes, energy storage branch access nodes, DC / DC nodes, PCS input nodes, busbar connection point nodes, contactor nodes, fuse nodes, cabinet temperature zone nodes, fan outlet nodes, and load branch nodes. Each node is loaded with node characteristics based on its corresponding component. For example, battery cluster nodes are loaded with branch current, terminal voltage, SOC, SOH, estimated internal resistance, and branch temperature; busbar connection point nodes are loaded with busbar temperature, infrared temperature rise, and estimated equivalent impedance; contactor nodes are loaded with contact voltage drop, open / closed state, and equivalent contact resistance; fan outlet nodes are loaded with wind speed, fan speed, and corresponding wind zone number; and load branch nodes are loaded with circuit breaker open / closed state, load type, and impact load identification.

[0045] Edges in the multiphysics manifold are used to represent the relationships between different physical fields and topological states within the cabinet. The connections between energy storage branches and busbars, the conductive paths between busbar connection points, and the series paths of contactors and fuses are set as resistive coupling edges. The heat transfer relationships between battery clusters, busbar connection points, contactors, fuses, internal temperature zones, and fan outlets are set as thermal diffusion coupling edges. The relationships between branch current ratio, temperature rise history, SOC operating range, and SOH trend are set as aging coupling edges. The impact of contactors, load-side circuit breakers, PCS operating modes, and DC / DC switching states on the current path is set as topology switching coupling edges. When a contactor disconnects or a load-side circuit breaker changes state, the topology switching coupling edges associated with that switching state are updated synchronously in the next control cycle, enabling the multiphysics manifold to reflect real-time current paths.

[0046] After establishing the multi-physics manifold graph, the edge computing controller calculates the current-carrying manifold coordinates of each energy storage branch. Specifically, the edge computing controller calculates the electrical manifold coordinates based on the resistive coupling edge and the branch's electrical state characteristics, the thermal manifold coordinates based on the thermal diffusion coupling edge and the cabinet's thermal state characteristics, the aging manifold coordinates based on the aging coupling edge and the branch's SOH (State of Harm) variation trend, and the topological manifold coordinates based on the topology switch coupling edge and the load-side power distribution state characteristics. These four types of manifold coordinates can be obtained through graph embedding with physical weights, or through graph neural networks, graph attention networks, or gated fusion networks. Regardless of the specific calculation method used, the output is constrained by the branch's rated current, the busbar's allowable temperature, the contactor's allowable voltage drop, the lower limit of insulation impedance, and the PCS current-limiting state.

[0047] In one specific implementation, the edge computing controller represents the node characteristics of the i-th energy storage branch as x_i(t), where x_i(t) = [I_i, U_i, SOC_i, SOH_i, R_i, T_i, ΔT_bus_i, ΔV_con_i, v_air_i, H_i, Z_ins_i, q_i]. Here, I_i represents the branch current, U_i represents the terminal voltage, SOC_i represents the state of charge, SOH_i represents the health state, R_i represents the estimated internal resistance, T_i represents the branch temperature, ΔT_bus_i represents the infrared temperature rise of the corresponding busbar node, ΔV_con_i represents the contactor voltage drop, v_air_i represents the corresponding duct velocity, H_i represents the humidity inside the cabinet, Z_ins_i represents the insulation resistance, and q_i represents the branch or load-side topology state code. The edge computing controller calculates the electrical manifold coordinates z_i^e, thermal manifold coordinates z_i^h, aging manifold coordinates z_i^a, and topological manifold coordinates z_i^s respectively, and obtains the current-carrying manifold coordinates z_i through a fusion function. Where c_i represents the branch's rated parameters and historical operating statistics, This represents a constrained fusion function.

[0048] While calculating the manifold coordinates that can carry current, the edge computing controller acquires external scheduling and load-side information. For example... Figure 4 As shown, the external scheduling and load-side information includes EMS scheduling commands, photovoltaic forecasts, load curves, time-of-use pricing, backup power strategies, grid connection / off-grid switching signals, load-side circuit breaker status, impact load initiation signals, and fire safety signals. The edge computing controller determines the total power change trend within the future control window based on EMS scheduling commands, determines the charging and discharging direction and duration based on photovoltaic forecasts and load curves, determines the branch redundancy and SOC range to be retained based on the backup power strategy, determines the PCS operating mode change based on the grid connection / off-grid switching signals, and determines short-term high-rate discharge demand based on the load-side circuit breaker status and impact load initiation signals.

[0049] The edge computing controller divides the future control window into multiple consecutive control cycles and generates an operation intent label, target power, duration, safety priority, and task intensity for each control cycle. The operation intent label can be identified as peak shaving discharge, photovoltaic charging, low-power maintenance, reverse charging, backup power voltage protection, black start preparation, impact load response, or fire-fighting load reduction. When the time-of-use price is in the peak range and the EMS issues a discharge power command, the operation intent can be identified as peak shaving discharge. When the predicted photovoltaic power is higher than the current load demand and there is a reverse power constraint at the grid connection point, the operation intent can be identified as photovoltaic charging or photovoltaic consumption. When the load-side circuit breaker status indicates that a high-power device is about to be connected, and the historical load curve shows that there is a short-term high power demand during the device's startup phase, the operation intent can be identified as impact load response. When the insulation impedance is lower than the safety threshold, the busbar temperature rise exceeds the temperature lockout threshold, or a fire signal is triggered, the operation intent is identified as fire-fighting load reduction or fault derating, and assigned the highest safety priority.

[0050] After obtaining the manifold coordinates of each branch's current carrying capacity and its future operational intent, the edge computing controller generates a flexible current sharing target. For example... Figure 5 As shown, the flexible current sharing target does not aim to make the current in each energy storage branch absolutely equal, but rather determines the target current distribution relationship between the total current demand, the branch's safe carrying capacity, and the future workload. The edge computing controller calculates the branch's comprehensive carrying capacity factor C_i(t) based on the thermal margin, aging margin, impedance margin, air cooling margin, insulation margin, and future intended matching margin of each energy storage branch. The formula for calculating the branch's comprehensive carrying capacity factor is: .in, Indicates heat margin, Indicates aging margin, Indicates impedance margin, Indicates the air-cooling margin, Indicates the insulation margin, Indicates the future intention matching margin. w_T, w_A, w_R, w_F, w_Z, and w_P represent the corresponding weights, and clip means to limit the calculation result between C_min and C_max.

[0051] The thermal margin M_i^T can be calculated according to where T_ref < T_lim. T_lim represents the upper limit of the allowable temperature of the branch or busbar, T_i represents the current branch temperature or the corresponding busbar node temperature, ΔT_i^pred represents the predicted temperature rise obtained from the predicted current and air-cooling state within the future control window, and T_ref represents the reference temperature. The impedance margin M_i^R can be calculated according to where R_ref represents the reference impedance, R_i represents the internal resistance of the branch, R_bus_i represents the equivalent impedance of the busbar, R_con_i represents the equivalent contact resistance of the contactor, and ε represents the correction amount to prevent the denominator from being zero. The aging margin M_i^A can be determined according to SOH, the number of cycles, the historical high-rate operation time, and the temperature rise history. The air-cooling margin M_i^F can be determined according to the outlet wind speed of the fan, the estimated value of the air duct resistance, the temperature in the cabinet area, and the fan state. The insulation margin M_i^Z can be determined according to the distance between the current insulation impedance and the lower limit of the insulation impedance. The future intention matching margin M_i^P is adjusted according to the operation intention and is used to make the branch allocation target match the future task intensity.

[0052] Among them, the predicted temperature rise can be calculated using the discrete thermal resistance-capacitance model: where a_i represents the thermal inertia coefficient, b_i represents the electro-thermal conversion coefficient, c_i represents the air-cooling heat transfer coefficient, and T_amb represents the reference temperature of the incoming air in the cabinet. The above coefficients can be obtained from the factory thermal test, historical temperature rise data, or on-site identification and are limited within the preset physical range.

[0053] When the operation intention is to respond to impact loads, the edge computing controller increases the future intention matching margin of the branches with low temperature, low resistance, high SOH, and good air-cooling margin, so that they carry a higher proportion of the current during short-term high-rate discharges. When the operation intention is to maintain the voltage during standby power, the edge computing controller reduces the current discharge target of the branches selected as subsequent emergency redundant branches, so that they retain a certain SOC and thermal margin. When the operation intention is to charge with photovoltaic power, the edge computing controller increases the charging carrying factor of the branches with low SOC, low temperature, and small charging internal resistance. When the operation intention is to reduce the load for fire protection or derate due to faults, the edge computing controller sets the comprehensive carrying factor of the branches with abnormal temperature zones, abnormal insulation, or abnormal contactor voltage drop to zero or the safety derating value.

[0054] After determining the comprehensive carrying capacity factor of the branch, the edge computing controller calculates the target current I_i^tar(t) of the i-th energy storage branch based on the total charging and discharging current command, and its expression is: Where I_total(t) represents the total charging and discharging current command of the energy storage distribution cabinet, and ΣC_j(t) represents the sum of the comprehensive carrying factors of the energy storage branches participating in operation. The edge computing controller can also add current change smoothing constraints after the target current is calculated, so that the target current change between adjacent control cycles does not exceed the preset change rate, thereby reducing the abrupt changes in the power regulation of the DC / DC module and PCS.

[0055] Before issuing control inputs, the edge computing controller performs safety interlocking verification. This verification includes branch maximum current verification, branch terminal voltage verification, busbar temperature rise verification, contactor voltage drop verification, fuse temperature verification, insulation impedance verification, PCS current limiting status verification, DC / DC operating status verification, fan status verification, load-side circuit breaker status verification, and fire signal verification. When the flexible current sharing target meets the above safety constraints, the edge computing controller outputs the corresponding branch current setpoint to the DC / DC module, the droop coefficient or power distribution correction to the PCS, the opening and closing priority to the contactor control unit, the fan speed command to the fan controller, and the current limiting threshold or bypass strategy to the load-side distribution circuit. When the flexible current sharing target does not meet the safety constraints, the edge computing controller recalculates the comprehensive load factor or directly executes derating, bypass, prohibition of closing, PCS power limiting, or fault interlocking control.

[0056] After the control input is issued and executed, the edge computing controller collects the execution results and calculates the residuals. For example... Figure 6 As shown, the bias flow residual The residual temperature rise e_i^T(t) can be expressed as The contactor voltage drop residual e_i^V(t) can be expressed as Where I_i,meas(t) represents the measured current of the i-th energy storage branch, ΔT_i,meas(t) represents the measured temperature rise, ΔV_con_i,meas(t) represents the measured contactor voltage drop, and ΔV_con_i,pred(t) is the contactor voltage drop predicted based on the branch current and the equivalent contact resistance of the contactor. If the bias current residual is consistently large, it indicates that the actual impedance, execution capability, or current path of the corresponding branch deviates from the model. If the temperature rise residual is consistently large, it indicates that the corresponding heat diffusion edge, duct cooling capacity, or busbar connection status needs to be corrected. If the contactor voltage drop residual is consistently large, it indicates that the contactor contact status has changed or the estimated contact resistance needs to be increased.

[0057] The edge computing controller performs constrained updates to the graph edge weights or margin parameters based on the residual loss L_res(t). The constrained update can be represented as... Where θ represents the graph edge weights or margin parameters, η represents the learning rate, and θ_min and θ_max represent the safety boundaries. This update process does not violate the hard constraints of the safety interlocking module. When the residuals are within the allowable range, the edge computing controller progressively corrects the weights of the resistively coupled edges, the weights of the thermally diffused coupled edges, the estimated equivalent contact resistance of the contactor, and the branch carrying capacity parameters. When the residuals exceed the safety threshold, the edge computing controller prioritizes derating, bypassing, or fault interlocking control and suspends the adaptive gain amplification of the corresponding branch.

[0058] In a specific implementation, the residual loss can be expressed as Where μ_I, μ_T, and μ_V are residual weights, ΔT_lim is the upper limit of allowable temperature rise, and ΔV_con,lim is the allowable voltage drop of the contactor. I_i,limit represents the upper limit of allowable current for the i-th energy storage branch in the current control cycle. The weights are non-negative and can be normalized according to the control objective. During online updates, only the resistive coupling edge, thermal diffusion coupling edge, or contactor equivalent resistance parameters related to the residual source are corrected.

[0059] Using the method described in this embodiment, the energy storage distribution cabinet can generate branch-level flexible current sharing targets based on the multi-physical states within the cabinet and future operational intentions under various operating conditions, including peak shaving discharge, photovoltaic charging, backup power voltage maintenance, grid connection / off-grid switching, and impact load response. This method enables branch current distribution to correspond to branch carrying capacity and future task intensity, and can continuously correct the multi-physical manifold diagram by executing residuals. Example 2

[0060] This embodiment describes the specific implementation of multi-physics manifold graph construction and current-carrying manifold coordinate calculation. This embodiment can be implemented based on Embodiment 1, or it can be used as an independent data modeling process in the flexible dynamic current sharing method for energy storage. Through this embodiment, the edge computing controller can unify the electrical connection relationships, heat transfer relationships, aging effect relationships, and switch topology relationships within the energy storage distribution cabinet into the same graph structure, and output the carrying capacity characterization quantity for branch current distribution based on the graph structure.

[0061] like Figure 3As shown, the multiphysics manifold graph within the energy storage distribution cabinet can be represented as G(t) = {V(t), E_e(t), E_h(t), E_a(t), E_s(t), X(t)}. Here, V(t) represents the set of nodes at time t, E_e(t) represents the set of resistive coupling edges, E_h(t) represents the set of thermal diffusion coupling edges, E_a(t) represents the set of aging coupling edges, E_s(t) represents the set of topology switch coupling edges, and X(t) represents the node characteristic matrix. This graph structure is updated with the control cycle. When the branch contactor state, load-side circuit breaker state, PCS operating mode, DC / DC switching state, or sensor data state changes, the node characteristics and some graph edge weights are updated accordingly.

[0062] The node set V(t) is established according to the physical components and controlled objects within the energy storage distribution cabinet. For each energy storage branch, battery cluster nodes, energy storage branch access nodes, and DC / DC nodes are set. Battery cluster nodes are used to carry branch current, terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, and historical operating characteristics. Energy storage branch access nodes are used to carry branch access status, branch current limiting value, and branch switch status. DC / DC nodes are used to carry current setpoint, actual output current, module temperature, module current limiting status, and module fault status. For the cabinet's common connection structure, busbar connection point nodes and PCS input nodes are set. Busbar connection point nodes are used to carry busbar node temperature, infrared temperature rise, busbar section equivalent impedance, and spatial location coding. PCS input nodes are used to carry PCS power commands, current power, PCS current limiting status, and operating mode. For the safety and thermal management structure, contactor nodes, fuse nodes, cabinet internal temperature zone nodes, and fan outlet nodes are set. Contactor nodes are used to display contactor open / closed status, contact voltage drop, and equivalent contact resistance. Fuse nodes are used to display fuse temperature, rated current, and thermal accumulation status. Cabinet temperature zone nodes are used to display temperature, humidity, and corresponding space area information for the temperature zone. Fan outlet nodes are used to display fan speed, air velocity, and fan operating status. For load-side structures, load branch nodes are provided, which are used to display the load-side circuit breaker open / closed status, load type, impulsive load identification, and estimated load power.

[0063] The set of resistive coupling edges, E_e(t), represents the actual conductive path and equivalent impedance relationship within the energy storage distribution cabinet. The edge between the battery cluster node and the DC / DC node represents the internal impedance of the energy storage branch and the module input impedance. The edge between the DC / DC node and the contactor node represents the branch output path. The edge between the contactor node and the fuse node represents the series path of the safety device. The edge between the fuse node and the busbar connection point node represents the branch access to the busbar path. The edge between multiple busbar connection point nodes represents the conductive path of the busbar section. The edge between the busbar connection point node and the PCS input node represents the DC-side power collection path. The edge between the PCS input node and the load branch node represents the path of power transfer to the load side via the PCS or distribution circuit. The initial weight of the resistive coupling edges can be determined based on the busbar section length, conductor cross-sectional area, contactor rated parameters, fuse parameters, and branch cable specifications, and is corrected during operation based on current residuals and contactor voltage drop residuals.

[0064] The set of thermal diffusion coupling edges, E_h(t), represents the relationships of heat conduction, heat convection, and air cooling effects between different components within the energy storage distribution cabinet. Thermal diffusion coupling edges are set between battery cluster nodes and internal temperature zone nodes to describe the impact of battery cluster heating on the internal temperature zone. Thermal diffusion coupling edges are set between busbar connection point nodes and internal temperature zone nodes to describe the diffusion of localized busbar temperature rise into the surrounding space. Thermal diffusion coupling edges are set between contactor nodes, fuse nodes, and internal temperature zone nodes to describe the impact of contactor contact heating and fuse heat accumulation on the surrounding temperature zone. Air cooling coupling edges are set between fan outlet nodes and internal temperature zone nodes to describe the suppression effect of fan outlet air velocity on the corresponding temperature zone's temperature rise. For energy storage distribution cabinets with a clearly defined airflow direction, directional thermal diffusion coupling edges can be set between the fan outlet node and multiple temperature zone nodes downstream of the airflow. The weights of the thermal diffusion coupling edges can be determined based on component spatial distance, airflow direction, air velocity, ventilation cross-sectional area, and temperature rise history, and are updated based on the temperature rise residual.

[0065] The aging coupling edge set E_a(t) is used to represent the influence relationship between current rate, temperature rise history, SOC operating range, cycle number, and SOH changes. Aging coupling edges can be set between battery cluster nodes to express the comparison of aging rates of different branches within the same energy storage cabinet under similar tasks. Aging coupling edges can be set between battery cluster nodes and busbar connection points to express the impact of long-term high-current operation of branches on connection point temperature rise and contact state changes. Aging coupling edges can be set between battery cluster nodes and contactor nodes to express the impact of frequent branch switching and high-rate current on contact state. Aging coupling edges do not directly determine the current path, but they affect the aging margin in branch capacity assessment. For branches with rapid SOH decline, long-term high-temperature operation, or a large number of high-rate discharge cycles, their aging margin is reduced, and the overall capacity factor is reduced through manifold fusion.

[0066] The set of topology switch coupling edges, E_s(t), represents the impact of contactor, circuit breaker, PCS operating mode, and DC / DC switching state on the current path. When a contactor in an energy storage branch is in the open state, the corresponding topology switch coupling edge is set to the open state, and this branch does not participate in the current distribution of the current control cycle. When a load-side circuit breaker changes from the open state to the closed state, the topology switch coupling edge between the load branch node and the PCS input node is activated. The intention inference module can use this topology change to determine whether there is an impact load response requirement. When the PCS switches from grid-connected mode to off-grid mode, the topology coupling state between the PCS input node and the load branch node changes, and the edge computing controller adjusts the branch redundancy and safety priority accordingly. The topology switch coupling edges are usually discrete, but continuous reliability weights can also be set based on the health status of the switching devices, contact voltage drop, and number of switching actions.

[0067] The node feature matrix X(t) is composed of feature vectors from different nodes. To facilitate graph computation, the edge computing controller normalizes features of different dimensions to preset ranges. For example, branch current can be normalized to the branch's rated current, temperature to the upper limit of allowable temperature, contactor voltage drop to the upper limit of allowable voltage drop, insulation impedance to be reverse-normalized to the lower safety limit, and wind speed to the fan's rated wind speed. Boolean switch states can be represented by 0 or 1; discrete states such as PCS operating modes and operation intention labels can be represented by state coding. The normalized node features retain the relative magnitudes of each physical quantity and facilitate the fusion of different physical fields.

[0068] In one specific implementation, the edge computing controller can construct adjacency matrices A_e(t), A_h(t), A_a(t), and A_s(t) according to the electrical domain, thermal domain, aging domain, and topological domain, respectively. Here, the element a_e,ij in A_e(t) represents the resistive coupling strength between node i and node j; the element a_h,ij in A_h(t) represents the thermal diffusion coupling strength between node i and node j; the element a_a,ij in A_a(t) represents the aging effect strength between node i and node j; and the element a_s,ij in A_s(t) represents the topological switching coupling state or reliability between node i and node j. These adjacency matrices can be updated according to the control cycle, and the node numbers in each adjacency matrix remain consistent, enabling the fusion of graph computation results from different physical domains within the same branch dimension.

[0069] The calculation of the current-carrying manifold coordinates can be performed using a physical weighted graph embedding method. The edge computing controller first calculates the electrical manifold coordinates z_i^e based on A_e(t) and the electrical characteristic X_e(t), giving branches with low resistance, low voltage drop, and stable current handling capability a higher electrical carrying capacity character. Then, it calculates the thermal manifold coordinates z_i^h based on A_h(t) and the thermal characteristic X_h(t), giving branches with low temperature rise, good air cooling, and stable busbar connection point temperature a higher thermal carrying capacity character. Subsequently, it calculates the aging manifold coordinates z_i^a based on A_a(t) and the aging characteristic X_a(t), giving branches with high SOH, short historical high-temperature operating time, and low cycle loss a higher aging carrying capacity character. Finally, it calculates the topological manifold coordinates z_i^s based on A_s(t) and the topological characteristic X_s(t), giving branches in an effective connection state, stable switching state, and available load paths a higher topological carrying capacity character.

[0070] In another specific implementation, the edge computing controller can employ a graph neural network or a graph attention network to calculate manifold coordinates. The graph neural network takes as input a node feature matrix X(t) and multi-domain adjacency matrices A_e(t), A_h(t), A_a(t), and A_s(t). The first layer of graph computation is used to transmit electrical, thermal, and topological states between adjacent nodes. The second layer of graph computation is used to extract the coupling characteristics between branches and busbars, contactors, temperature zones, and load branches. The output layer generates the electrical manifold coordinates, thermal manifold coordinates, aging manifold coordinates, and topological manifold coordinates for each energy storage branch. The graph attention network can also assign attention weights to different adjacent nodes. For example, when the current deviation in a branch is mainly caused by abnormal contactor voltage drop, the attention weight of the contactor node for the branch's load-bearing capacity is increased; when the load reduction in a branch is mainly caused by an increase in the duct temperature zone, the attention weights of the cabinet temperature zone node and the fan outlet node for the branch's load-bearing capacity are increased.

[0071] In a specific implementation, the graph neural network can include two graph convolutional layers, one gated fusion layer, and one fully connected output layer. The hidden dimension can be set to 16 to 64. The input is the node feature matrix and four types of adjacency matrices, and the output is each branch. The training data is divided into training, validation, and test sets in chronological order. The loss function consists of the prediction error of the current carrying capacity, the prediction error of the temperature rise, the prediction error of the contactor voltage drop, and the penalty for breach of the safety boundary. During deployment, the model output must undergo amplitude-limited projection and safety interlocking verification and is not directly issued as the final control quantity.

[0072] For the implementation involving model training, training data can come from historical operation records, factory test data, and simulation data of the energy storage distribution cabinet. Each training sample includes the node feature matrix, multi-domain adjacency matrix, actual branch current, busbar temperature rise, contactor voltage drop, fan status, PCS power command, and safety results within a certain control cycle. Training objectives can include prediction errors for branch current carrying capacity, temperature rise, bias current residual prediction errors, and safety boundary violation penalties. During training, branch currents that do not experience over-temperature, over-current, or insulation abnormalities and operate stably can be used as positive samples, while control cycles experiencing abnormal temperature rises, voltage drops, PCS current limiting, or branch bypasses can be used as risk samples. After model training is completed, when deployed to the edge computing controller, the output results still need to be constrained by the safety interlocking module and cannot directly exceed the branch rated current, busbar allowable temperature, contactor allowable voltage drop, and insulation impedance lower limit.

[0073] During the fusion phase, the edge computing controller inputs the electrical manifold coordinates z_i^e, thermal manifold coordinates z_i^h, aging manifold coordinates z_i^a, and topological manifold coordinates z_i^s into the fusion function to obtain the current-carrying manifold coordinates z_i of the i-th energy storage branch. The fusion function can use... This indicates that c_i represents the branch's rated parameters and historical operating statistics. For a rule-based implementation, the fusion function can use weighted summation followed by amplitude limiting. For a gated fusion implementation, the edge computing controller adjusts the gating coefficients of different manifold coordinates based on the operating intent and safety status. For example, in high-temperature environments or when the fan malfunctions, the constraint weight of the thermal manifold coordinates is increased; after a branch has undergone long-term high-rate operation, the constraint weight of the aging manifold coordinates is increased; during grid-connected / off-grid switching or changes in the load circuit breaker status, the constraint weight of the topology manifold coordinates is increased.

[0074] In one specific implementation, during the rule-based implementation, the edge computing controller normalizes various physical quantities to the 0-1 range and calculates the coordinates of the four types of manifolds using the following formula: Wherein, each β is a non-negative weight and the sum of the weights within the same domain is 1; q_i represents the valid identifier of the branch access; R_i, ΔV_con_i, T_i, ΔT_i^pred, v_air_i, N_i^cycle, and t_i^hot are the normalized impedance, contactor voltage drop, temperature, predicted temperature rise, wind speed, cycle count, and high temperature accumulation time, respectively. The superscript n indicates the normalized quantity; s_i,con represents the contactor availability status or confidence level; s_i,pcs represents the PCS operating mode availability status; s_i,load represents the load path availability status; and q_i represents the valid identifier of the branch access.

[0075] The current-carrying manifold coordinate z_i is not the final target current, but rather a characterization of the branch's ability to participate in flexible current sharing optimization. The edge computing controller further calculates thermal margin, aging margin, impedance margin, air cooling margin, insulation margin, and future intended matching margin based on z_i, forming a comprehensive carrying factor C_i(t). When z_i indicates that a branch is in good condition in the electrical and aging domains, but the corresponding busbar node is in a high-temperature zone, the comprehensive carrying factor will be limited by the thermal margin. Conversely, when a branch has a low SOC but abnormal contactor voltage drop and insufficient insulation margin, this branch will not be allocated excessive charging current solely due to its low SOC. Therefore, the multi-physics manifold diagram enables branch current allocation to simultaneously consider multiple safety constraints.

[0076] To improve the adaptability of the graph structure during long-term operation, the edge computing controller updates the graph edge weights based on the execution residuals after each control cycle. If the measured current of the i-th branch is consistently lower than the target current, and the DC / DC module is fault-free, the edge computing controller can determine that the actual resistance path of the branch is higher than the model estimate and increase the weight of the corresponding resistance coupling edge. If the temperature rise of a busbar connection point is higher than the predicted value under the same current conditions, the thermal diffusion coupling weight between the busbar node and the adjacent temperature zone node is increased, or the air cooling margin of that section is reduced. If the contactor voltage drop is consistently higher than the predicted value, the equivalent contact resistance of the contactor node is increased, and the opening and closing priority of that branch in subsequent switching is reduced. If a short-term bias current occurs after a change in the load-side topology, the edge computing controller can adjust the confidence level of the corresponding topology switch coupling edge to make subsequent intention inference and current sharing distribution closer to the actual current path.

[0077] The multi-physics manifold graph described in this embodiment can also support deployment methods with different levels of precision. In power distribution cabinets with limited hardware resources, a fusion of physical weight graphs and rules can be used to calculate manifold coordinates. This method has a lower computational load and is suitable for scenarios with short control cycles. In power distribution cabinets with edge computing controllers that have high computing power, graph neural networks or graph attention networks can be used for graph representation learning. This method can learn the nonlinear relationships between branches, electrical connections, thermal diffusion, and topology states from historical operating data. In scenarios with high safety requirements, a physical weight graph can be used as the main control link, and the output of the graph neural network can be used as a load-bearing capacity correction factor, with all outputs projected to the allowable range through a safety interlocking module.

[0078] Using the method described in this embodiment, the energy storage distribution cabinet can transform the coupling relationships between multiple branch energy storage units, PCS or DC / DC modules, busbars, contactors, fuses, duct temperature zones, and load branches into a computable graph structure. This graph structure can reflect the current state of the branches and can also adapt to cabinet aging and environmental changes through residual updates, providing a basis for branch-level load-bearing capacity for subsequent flexible current sharing targets. Example 3

[0079] This embodiment describes the specific implementation of operational intent deduction and flexible current sharing target generation. This embodiment can be implemented in conjunction with Embodiments 1 and 2 to further convert the branch carrying capacity output from the multi-physical manifold diagram into branch target currents and various control quantities. The intent deduction module in this embodiment refers to a functional unit located within the edge computing controller, used to parse EMS scheduling information, prediction information, load-side status, and safety signals to generate the operational intent of the future control window. This functional unit can be implemented by software programs, state machines, rule bases, time series prediction models, or combinations thereof within the edge computing controller. The flexible current sharing optimization module refers to a functional unit located within the edge computing controller, used to generate branch target currents and control quantities based on the manifold coordinates of the carrying current and the operational intent. The above functional units can be integrated with the edge computing controller in Embodiment 1 within the same processor, industrial controller, or embedded control board, or the edge computing controller can call partial prediction results provided by an external EMS and complete the calculation locally.

[0080] like Figure 4As shown, the intent deduction module receives EMS dispatch instructions, photovoltaic forecasts, load curves, time-of-use pricing, backup power strategies, grid connection / off-grid switching signals, load-side circuit breaker status, impact load initiation signals, PCS operating modes, and fire safety signals. These inputs are not used in isolation but are uniformly parsed according to a future control window. The future control window can consist of multiple continuous control cycles, and its length can be set according to the application scenario of the energy storage distribution cabinet. For impact load response, the future control window can be set to several seconds to tens of seconds; for peak shaving and valley filling and photovoltaic consumption, the future control window can be set to several minutes to tens of minutes; for backup power voltage maintenance, the future control window can be set to a longer strategy cycle. Different window lengths can coexist in the same system. The edge computing controller can use short windows for fast power response and long windows for SOC and branch lifetime allocation.

[0081] The intent deduction module first analyzes the total power target of the energy storage distribution cabinet within the future window based on the EMS dispatch instructions. If the EMS dispatch instructions indicate discharging to the load side during high-load periods, and the load curve shows that the load demand within the future window is consistently higher than the preset load baseline, the intent deduction module marks the corresponding window as peak-shaving discharge. If the photovoltaic forecast shows that the photovoltaic power generation within the future window is higher than the local load absorption capacity, and the grid connection point power constraint requires a reduction in reverse power, the intent deduction module marks the corresponding window as photovoltaic charging or photovoltaic absorption. If the EMS instructions require the energy storage cabinet to maintain a certain SOC range, and the critical load backup power strategy is enabled, the intent deduction module marks the corresponding window as backup power voltage protection. If the grid connection / off-grid switching signal indicates that the energy storage cabinet will switch from grid connection to off-grid status, or the PCS operating mode is about to switch, the intent deduction module marks the corresponding window as black start preparation or grid connection / off-grid switching preparation. If the load-side circuit breaker status shows that a high-power load is about to close, or the load curve shows a rapid upward trend in a short period of time, the intent deduction module marks the corresponding window as impact load response. If a fire alarm, busbar overheating, insulation abnormality, or smoke detector signal is triggered, the intention simulation module will mark the corresponding window as fire load reduction or fault derating.

[0082] After determining the operational intent label, the intent deduction module generates target power, duration, safety priority, and task intensity for each control cycle within the future control window. Target power can be determined by EMS power commands, the difference between photovoltaic predicted power and load predicted power, backup power strategy requirements, or estimated impact load power. Duration can be determined by the scheduling plan, the duration of the prediction curve, the load startup duration, or the duration of a safety event. Safety priority is used to determine the control sequence when multiple operational intents exist simultaneously. Fire-fighting load reduction and insulation anomalies have higher safety priority than peak shaving and photovoltaic absorption; backup power voltage maintenance has higher safety priority than general power dispatch; and impact load response has higher safety priority than low-power maintenance. Task intensity represents the requirement for branch current carrying capacity within the future window, and can be determined by the target power change rate, peak power, duration, and safety priority.

[0083] In one specific implementation, the intent information for the k-th control cycle within the future control window can be represented as p_k = (y_k, P_k, τ_k, ρ_k, α_k). Here, y_k represents the operational intent label, P_k represents the target power, τ_k represents the expected duration, ρ_k represents the safety priority, and α_k represents the task intensity. The task intensity α_k can be calculated by normalizing the ratio of target power to rated power, the power change rate, the duration, and the safety priority. For impact load response, the power change rate and peak power have higher weights; for backup power voltage maintenance, the SOC maintenance requirement and branch redundancy requirement have higher weights; for fire load reduction, the safety priority directly derates or bypasses abnormal branches, rather than prioritizing meeting the original power command.

[0084] The operational intent output by the intent deduction module is passed to the flexible flow sharing optimization module. For example... Figure 5 As shown, the flexible current sharing optimization module receives the current-carrying manifold coordinates obtained in Example 2 and the operating intent obtained in this example, and calculates the comprehensive carrying factor of each energy storage branch. The comprehensive carrying factor is jointly determined by thermal margin, aging margin, impedance margin, air cooling margin, insulation margin, and future intent matching margin. All of the above margins are limited within a preset range to prevent a single index anomaly from causing a sudden change in the target current. Thermal margin reflects the margin of the branch or busbar from the upper limit of temperature; aging margin reflects the branch's SOH and historical operating loss status; impedance margin reflects the branch's internal resistance, busbar impedance, and contactor contact resistance status; air cooling margin reflects wind speed, fan status, and heat dissipation capacity of the temperature zone; insulation margin reflects the DC side insulation safety status; and future intent matching margin reflects whether the branch is suitable to undertake tasks within the future window.

[0085] The flexible current sharing optimization module calculates the branch comprehensive load-bearing factor C_i(t), which is expressed as C_i(t) = clip(w_TM_i^T + w_A M_i^A + w_R M_i^R + w_F M_i^F + w_Z M_i^Z + w_P M_i^P, C_min, C_max). Here, M_i^T represents thermal margin, M_i^A represents aging margin, M_i^R represents impedance margin, M_i^F represents air cooling margin, M_i^Z represents insulation margin, M_i^P represents future intended matching margin, w_T, w_A, w_R, w_F, w_Z, and w_P represent corresponding weights, and clip restricts the calculation result to between C_min and C_max. The weights w_T, w_A, w_R, w_F, w_Z, and w_P can be determined by factory calibration, historical operating data, and scenario configuration. In conventional peak-shaving discharge, thermal margin, impedance margin, and aging margin can have similar weights; in high-temperature environments, the weights of thermal margin and air-cooling margin are increased; in long-term backup power scenarios, the weights of aging margin and SOC maintenance strategy are increased; in grid-connected / off-grid switching or impact load response, the weights of future intention matching margin and topology availability are increased; in insulation anomaly or fire-fighting load reduction scenarios, safety interlocking logic directly limits or covers the comprehensive load factor.

[0086] The thermal margin M_i^T is calculated based on the current temperature and the predicted temperature rise. The edge computing controller can predict the temperature rise ΔT_i^pred within a future window based on the current target current, branch internal resistance, busbar equivalent impedance, contactor contact resistance, fan outlet velocity, and cabinet temperature zone. When the current temperature of a branch is close to the upper temperature limit, or the infrared temperature rise at the busbar connection point is high, M_i^T decreases. For branches with lower temperature zones and sufficient fan outlet velocity, M_i^T increases. The aging margin M_i^A can be calculated based on SOH, cycle count, cumulative high-rate discharge duration, cumulative high-temperature operation duration, and historical bypass count. The impedance margin M_i^R is calculated based on the branch internal resistance, busbar equivalent impedance, and contactor equivalent contact resistance. The air-cooling margin M_i^F is calculated based on fan status, velocity, estimated duct resistance, and temperature zone gradient. The insulation margin M_i^Z is calculated based on the distance between the current insulation impedance and the lower limit of the insulation impedance safety margin. When the insulation resistance approaches the safety lower limit, the margin decreases and triggers the safety interlocking module to strengthen the constraint.

[0087] The future intention matching margin M_i^P is jointly determined by the task type and branch status output by the intention deduction module. If the operational intention is peak-shaving discharge, M_i^P allows branches with low temperature, low resistance, high SOH, and sufficient SOC to bear more discharge current. If the operational intention is photovoltaic charging, M_i^P allows branches with low SOC, low temperature, and low charging internal resistance to bear more charging current. If the operational intention is backup power and voltage maintenance, M_i^P reduces the discharge ratio of some healthy branches in the current period to retain subsequent emergency redundancy. If the operational intention is impact load response, M_i^P prioritizes branches with strong instantaneous load capacity, large DC / DC current limiting margin, and low corresponding wind zone temperature. If the operational intention is fire-fighting load reduction, M_i^P sets zero or a safe derating value for branches in abnormal temperature zones, insulation abnormalities, contactor voltage drop abnormalities, or smoke detector-related areas.

[0088] After the branch comprehensive bearing factor is determined, the flexible current sharing optimization module calculates the target current of each energy storage branch according to the total charging and discharging current command of the energy storage distribution cabinet. The target current I_i^tar(t) of the i-th energy storage branch is calculated according to I_i^tar(t)=I_total(t)·[C_i(t)·I_i,max] / Σ_j[C_j(t)·I_j,max]. Where I_total(t) represents the total charging and discharging current command of the energy storage distribution cabinet at time t; C_i(t) represents the comprehensive carrying capacity factor of the i-th energy storage branch at time t; I_i,max represents the safe current limit value for the i-th energy storage branch to participate in current distribution during the current control cycle; C_j(t) represents the comprehensive carrying capacity factor of the j-th energy storage branch currently in operation; I_j,max represents the safe current limit value for the j-th energy storage branch currently in operation to participate in current distribution during the current control cycle; Σ_j[C_j(t)·I_j,max] represents the summation of C_j(t)·I_j,max for all energy storage branches currently in operation that have not been bypassed, fault-locked, or disconnected by contactors. This target current expression enables the total current command to be distributed according to the comprehensive carrying capacity of the branches. When Σ_j[C_j(t)·I_j,max] is zero, or the total carrying capacity of the energy storage branches involved in operation is insufficient to meet I_total(t), the flexible current sharing optimization module does not forcibly allocate the target current of the branch according to the original total current command, but sets the target current of the abnormal branch to zero or a safe derating value, and outputs the cabinet derating command to the EMS or PCS.

[0089] To ensure the target current remains continuous between adjacent control cycles, the flexible current sharing optimization module can solve a constrained optimization problem based on the above allocation results. This optimization problem can be expressed as min Σ a_i·(I_i−I_i^tar)^2+λ_1Σ(I_i−I_i,last)^2+λ_2Ψ_T+λ_3Ψ_age+λ_4Ψ_switch. Here, I_i represents the actual branch current setpoint to be issued, I_i,last represents the branch current setpoint of the previous control cycle, a_i represents the branch reliability weight, λ_1 represents the current smoothing weight, λ_2 represents the temperature rise risk weight, λ_3 represents the aging risk weight, λ_4 represents the switch action penalty weight, Ψ_T represents the temperature rise risk term, Ψ_age represents the aging risk term, and Ψ_switch represents the contactor or circuit breaker action penalty term. The optimization problem is subject to the following constraints: ΣI_i=I_total, 0≤|I_i|≤I_i,limit, T_i^pred≤T_lim, ΔV_con_i≤ΔV_con,lim, Z_ins≥Z_ins,min, and the PCS and DC / DC capability boundaries.

[0090] Where Ψ_T can be taken as Σ_i max(0,T_i^pred−T_lim)^2, Ψ_age can be taken as Ψ_age=Σ_i(1−SOH_i^n)|I_i| / I_i,limit, where SOH_i^n represents SOH normalized to the interval between 0 and 1, and Ψ_switch can be taken as the number of contactor actions or the weighted value of the number of actions within the control cycle.

[0091] During the specific solution process, if all branches meet the constraints of temperature, voltage drop, insulation, and current, the flexible current sharing optimization module outputs a branch current setpoint that is close to the target current and changes smoothly. If some branches are close to the upper limit of temperature or voltage drop, the flexible current sharing optimization module reduces the current setpoint of that branch and allocates the difference to branches that still have a carrying margin. If the sum of the available carrying factors of all branches is insufficient, the system reduces the total current target of the entire cabinet and feeds back the derating capability to the EMS. If the safety interlocking module determines that a branch has over-temperature, insulation abnormality, or contactor voltage drop abnormality, the target current of that branch is limited to zero or a safe value, and determines whether to disconnect the contactor or execute a bypass strategy based on the branch status.

[0092] The flexible current sharing target is further converted into specific control variables. For branches with DC / DC modules, the branch current setpoint is directly sent to the corresponding DC / DC module, causing it to charge or discharge according to the target current. For scenarios where the PCS participates in power distribution, the flexible current sharing optimization module generates a PCS droop coefficient based on the total power demand, the branch target current, and the PCS current limiting status. The PCS droop coefficient is used to adjust the power response of the PCS when the DC bus voltage changes or the AC side power changes, so that the PCS current limiting action is consistent with the branch flexible current sharing target. For scenarios requiring branch switching, the system determines the contactor opening and closing priority based on the comprehensive load factor, contactor voltage drop, number of switching actions, and operating intention. Branches with high thermal margin, low contactor voltage drop, and high SOH can be prioritized for switching; branches with abnormal voltage drop, abnormal temperature rise, or insufficient insulation margin are given lower switching priority or prohibited from closing.

[0093] The fan speed is jointly determined by the results of flexible current sharing optimization and temperature zone prediction. When a branch corresponding to a certain temperature zone is about to bear a high current in the future window, the duct temperature control module increases the corresponding fan speed in advance, so that the temperature zone enters a pre-cooling state. When the operation intention is to maintain low power and the temperature zone temperature is stable, the fan speed can be reduced to the level required to maintain heat dissipation. When the infrared temperature rise at the busbar connection point is rapid, even if the branch current has not exceeded the current limiting threshold, the edge computing controller can also increase the fan speed of the corresponding air zone and reduce the load factor of the corresponding busbar section branch. The current limiting threshold is dynamically determined by the comprehensive load-bearing capacity of the branch, and the fault bypass strategy is determined by the safety interlocking module according to the type of abnormality. For branches with abnormal contactor voltage drop, the system can first reduce the target current of the branch and observe the voltage drop change; for branches with abnormal insulation impedance or triggered by fire signal, the system directly executes bypass or disconnection control.

[0094] In peak-shaving discharge scenarios, the energy storage distribution cabinet receives an EMS discharge command with a total current command of I_total(t). If the SOC of branches 1 and 2 is high, but the busbar connection point of branch 2 experiences a high temperature rise, the SOC of branch 3 is moderate, but the SOH is good and the air-cooling conditions are good, and the contactor voltage drop of branch 4 is slightly high, then the flexible current sharing optimization module will not simply distribute the discharge current according to the SOC. Instead, it will increase the carrying factor of branch 3 and decrease the carrying factors of branches 2 and 4 after considering thermal margin, impedance margin, and aging margin, allowing branch 3 to bear more discharge current. This treatment can prevent the high-temperature busbar connection point from continuing to heat up during peak-shaving discharge.

[0095] In photovoltaic (PV) charging scenarios, PV forecasts indicate surplus PV power within a future window, requiring the energy storage distribution cabinet to absorb charging current. If the first branch has a low State of Charge (SOC) and low temperature, the second branch has a low SOC but abnormal contactor voltage drop, the third branch has a high SOC close to its upper limit, and the fourth branch has a moderate SOC and low charging resistance, then the flexible current sharing optimization module increases the charging targets for the first and fourth branches, while decreasing the charging targets for the second and third branches. For the second branch with abnormal contactor voltage drop, even with a low SOC, it will not be allocated excessive charging current. Therefore, the charging current distribution during PV absorption can simultaneously meet both SOC absorption capacity and cabinet safety constraints.

[0096] In backup power scenarios, critical load backup power strategies require energy storage cabinets to retain a certain emergency capacity. If some branches have high State of Harm (SOH), low temperature rise, and State of Charge (SOC) within a suitable range, the edge computing controller can mark some of these branches as emergency redundant branches and reduce their current discharge targets, allowing them to retain higher available capacity during subsequent grid-connected / off-grid switching or black-start preparation. The remaining branches undertake current low-power discharge or maintenance tasks within a safe range. This approach differs from average current sharing because its goal is not to make the current of each branch completely equal, but rather to allocate branch margins between the current task and subsequent backup power tasks.

[0097] In the impact load response scenario, the load-side circuit breaker status indicates that a high-power device is about to start, and the power change rate is high within a short window. The intent inference module marks this window as an impact load response and increases the task intensity α_k. The flexible current sharing optimization module prioritizes branches with low resistance, low temperature, high SOH, large DC / DC current limiting margin, and high wind speed in the corresponding wind zone to bear the short-term high-rate current. The duct temperature control module increases the fan speed in the corresponding temperature zone in advance, and the PCS droop coefficient is also adjusted accordingly to reduce the inconsistency between the PCS current limiting action and the branch current distribution. If the bias current residual exceeds the threshold after the impact load starts, the online self-correction module corrects the weight of the corresponding resistive coupling side and reduces the target current of the bias current branch in the next control cycle.

[0098] In fire protection load reduction scenarios, if an over-temperature alarm occurs in a certain temperature zone within a cabinet, or if the insulation impedance falls below the safety lower limit, the intention-inference module sets fire protection load reduction or fault derating to the highest safety priority. The flexible current sharing optimization module sets the comprehensive load-bearing factor of the abnormal temperature zone associated branch or the branch with abnormal insulation to zero or a safe derating value. The safety interlocking module prohibits that branch from continuing to undertake ordinary current sharing tasks and performs bypass, contactor disconnection, or overall cabinet derating according to the abnormality level. The remaining branches redistribute the total current according to the remaining load-bearing capacity. If the remaining branches cannot safely carry the original total current target, the system reports a reduction in available power to the EMS and executes a load-side graded load reduction strategy.

[0099] Through the above processing, this embodiment enables the operational intent projection results to directly influence the flexible current sharing target. The generation of the branch target current depends not only on the current difference or SOC difference, but also on thermal margin, aging margin, impedance margin, air cooling margin, insulation margin, and the task intensity within the future window. This method can generate different branch allocation strategies under different operating scenarios, and further convert these strategies into DC / DC current setpoints, PCS droop coefficients, contactor opening and closing priorities, fan speed, current limiting thresholds, and fault bypass strategies, thereby forming a coordinated control process for electrical current sharing, thermal management, and safety protection. Example 4

[0100] This embodiment provides a flexible dynamic current sharing distribution cabinet for energy storage that integrates multiphysics manifold diagrams and intent inference. This distribution cabinet executes the flexible dynamic current sharing method for energy storage described in the preceding embodiments, and implements the processes of multiphysics manifold diagram construction, intent inference, flexible current sharing optimization, safety interlocking, and online self-correction into the specific cabinet structure, sampling unit, execution unit, and edge computing controller. For example... Figure 2 As shown, the distribution cabinet includes a cabinet, an energy storage branch access unit, a DC busbar, a PCS access unit, a DC / DC or current sharing execution module, contactors, fuses, a load-side distribution circuit, a branch current sensor, a branch voltage sampling unit, a busbar temperature sensor, a contactor voltage drop sampling unit, an insulation detection unit, a duct temperature control module, a communication module, and an edge computing controller.

[0101] The energy storage branch access unit is used to connect battery clusters, battery packs connected in series and parallel, or energy storage units with independent DC / DC converters to the energy storage distribution cabinet. Each energy storage branch access unit corresponds to one energy storage branch and works with a branch current sensor, a branch voltage sampling unit, a contactor, a fuse, and a DC / DC converter or current sharing execution module. This unit provides the edge computing controller with branch current, branch terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, and branch switch status, enabling the edge computing controller to determine the current electrical load status of the branch and its availability for current sharing. Specific implementations of the energy storage branch access unit may include branch copper busbars, branch cables, branch input terminals, branch sampling terminals, branch contactor mounting positions, branch fuse mounting positions, and branch communication interfaces. For branches with DC / DC converters, the branch access unit also includes DC / DC input / output terminals and a current closed-loop control interface. For branches without independent DC / DC converters, the branch access unit can participate in current sharing control through branch contactor switching, branch current limiting threshold, and PCS droop coefficient.

[0102] The DC busbar is used to collect DC power from multiple energy storage branches and connect the PCS access unit and the load-side power distribution circuit. The DC busbar may include a positive busbar, a negative busbar, branch access terminals, PCS access terminals, and load-side output terminals. Busbar temperature sensors are installed at busbar connection points, branch access points, bends, end connection points, or high current density areas to collect busbar node temperatures. Busbar temperature sensors can be surface-mount temperature sensors, infrared temperature sensors, fiber optic temperature sensors, or combinations thereof. The DC busbar corresponds to a busbar connection point node in the multiphysics manifold diagram. The collected temperature and infrared temperature rise are used for thermal manifold coordinate calculations and also for the safety interlock module to determine if the busbar has an over-temperature risk.

[0103] The PCS access unit connects the DC busbar to the AC distribution system and performs power conversion in grid-connected, off-grid, charging, discharging, and standby modes. The PCS access unit provides the edge computing controller with the current PCS power, PCS power command, DC-side voltage, AC-side status, current-limiting status, and operating mode. The PCS access unit also receives the droop factor, power distribution correction, or power-limiting command output from the edge computing controller. When the flexible current sharing optimization module determines that the carrying capacity of certain branches has decreased, the PCS access unit can adjust the power response according to the droop factor or power-limiting command to prevent the PCS from continuing to drive high-risk branches to bear excessive current according to the original total power command.

[0104] The DC / DC or current sharing execution module is used to control the charging and discharging current of each energy storage branch according to the current setpoint issued by the edge computing controller. For each energy storage branch with a DC / DC module, the DC / DC or current sharing execution module receives the target current of the corresponding branch and feeds back the actual output current, module temperature, module current limiting status, and fault status to the edge computing controller. This module can be a bidirectional DC / DC converter, a branch current control module, or a current sharing execution unit with current limiting function. For distribution cabinets without independent DC / DC modules, the DC / DC or current sharing execution module can be replaced by a branch current limiter, a controllable switching device, or a PCS-side power distribution control unit, and the edge computing controller can achieve approximately flexible current sharing through branch switching priority and PCS droop coefficient.

[0105] Contactors are used to control the electrical connection between energy storage branches and DC busbars. Each energy storage branch can be equipped with at least one main contactor, and pre-charge contactors or auxiliary contactors can be set according to the DC voltage level. A contactor voltage drop sampling unit is connected to both ends of the contactor contacts to collect the contact voltage drop when the contactor is closed. The edge computing controller calculates the equivalent contact resistance of the contactor based on the contact voltage drop and the branch current, and uses it as a node characteristic of the contactor node in the multiphysics manifold. When the contactor voltage drop is higher than the allowable voltage drop or the voltage drop rises abnormally, the safety interlocking module reduces the carrying capacity of the corresponding branch, the flexible current sharing optimization module reduces the target current of the branch, and the contactor opening and closing priority is reduced accordingly.

[0106] Fuses are used to provide overcurrent protection for energy storage branches. Fuses can be configured with temperature sampling points to obtain their thermal state. In the multiphysics manifold, the fuse node is adjacent to the energy storage branch, contactor, and busbar nodes; its temperature and rated current are used to assess the branch's safe carrying capacity. When the fuse temperature approaches the allowable value, or when the branch containing the fuse operates at high rates for an extended period, the edge computing controller reduces the branch's thermal margin and overall carrying capacity factor to prevent the fuse from being in a state of high heat accumulation for a prolonged period.

[0107] The load-side power distribution circuit is used to connect critical loads, general loads, grid-connected / off-grid switching circuits, fire-fighting load reduction circuits, or graded disconnection circuits. The load-side power distribution circuit includes load-side circuit breakers, contactors, load detection units, and communication interfaces. The status of the load-side circuit breaker is sent to the edge computing controller to construct the topology switch coupling edge and participate in intent inference. When the load-side circuit breaker indicates that an impending load is about to be connected, the intent inference module generates the impending load response intent in the future control window; when the fire-fighting load reduction circuit is triggered, the safety interlocking module executes fault derating, branch bypass, or load-side graded disconnection strategies.

[0108] Branch current sensors are used to collect the charging and discharging current of each energy storage branch. Branch current sensors can be Hall current sensors, shunts, or other DC current sampling devices. Branch voltage sampling units are used to collect the terminal voltage of each energy storage branch and the input and output voltages of the DC / DC module. Branch current and terminal voltage are used to calculate branch power, current deviation, branch execution residual, and branch electrical status. Busbar temperature sensors are used to collect the temperature of the DC busbar and its connection points. Contactor voltage drop sampling units are used to collect the contactor contact voltage drop, and insulation detection units are used to detect the DC side insulation impedance to ground. All of the above sampling units are connected to the edge computing controller, and their sampling results are entered into the multi-source data acquisition module after time alignment and outlier processing.

[0109] The duct temperature control module controls heat dissipation within the energy storage distribution cabinet and provides cooling margin calculation data to the edge computing controller. The module includes a fan, duct, fan outlet wind speed sensor, cabinet interior temperature zone sensor, and fan controller. The duct is arranged along multiple energy storage branches, DC busbars, PCS access units, and the areas where DC / DC modules are located. The fan outlet wind speed sensor measures the actual wind speed, and the cabinet interior temperature zone sensor measures the temperature in different areas. The edge computing controller predicts the temperature rise within the future control window based on wind speed, fan speed, temperature zone, and branch current, and calculates the cooling margin and thermal margin accordingly. When a temperature zone is about to handle a large current, the duct temperature control module increases the corresponding fan speed in advance according to the edge computing controller's instructions.

[0110] The communication module is used to enable data exchange between the energy storage distribution cabinet and the EMS, PCS, DC / DC modules, wind turbine controllers, load-side circuit breakers, fire protection systems, and upper-level monitoring systems. The communication module can use CAN, RS485, Ethernet, fiber optic communication, or industrial fieldbus. It provides the edge computing controller with EMS scheduling commands, photovoltaic forecasts, load curves, time-of-use pricing, backup power strategies, grid connection / off-grid switching signals, PCS status, and load-side circuit breaker status. It also transmits control commands generated by the edge computing controller to the execution components.

[0111] The edge computing controller is used to implement a flexible dynamic current sharing method for energy storage. This controller can be an industrial controller, an embedded processor, an edge computing gateway, or a local control board for the energy storage cabinet. The edge computing controller includes a multi-source data acquisition module, a multi-physical manifold construction module, a branch capacity assessment module, an intent deduction module, a flexible current sharing optimization module, a safety interlocking module, and an online self-calibration module. Each module can be a software function module, or it can be implemented collaboratively by software and sampling interfaces, communication interfaces, storage units, and safety logic circuits.

[0112] The multi-source data acquisition module is used to acquire branch current, terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, busbar infrared temperature rise, contactor voltage drop, fan outlet wind speed, cabinet humidity, insulation impedance, PCS power command, and load-side circuit breaker status. This module also performs time alignment on data from different sampling periods, removes or replaces abnormal sampled values, and generates identifiers for data reliability. The specific implementation of the multi-source data acquisition module may include a sampling interface driver, a communication parsing unit, a timestamp buffering unit, an anomaly detection unit, and a feature normalization unit. The sampling interface driver reads local sensor data, the communication parsing unit parses EMS, PCS, and DC / DC module data, the timestamp buffering unit organizes data according to the control cycle, the anomaly detection unit identifies out-of-limit and missing values, and the feature normalization unit forms a node feature matrix.

[0113] The multi-physics manifold construction module is used to build a multi-physics manifold graph containing resistive coupling edges, thermal diffusion coupling edges, aging coupling edges, and topology switching coupling edges. This module determines the node set based on the cabinet configuration file, determines resistive coupling edges based on branch and busbar connections, determines thermal diffusion coupling edges based on component spatial locations and airflow direction, determines aging coupling edges based on SOH changes, temperature rise history, and current ratios, and determines topology switching coupling edges based on contactor, circuit breaker, PCS operating modes, and DC / DC switching states. The specific implementation of this module may include a topology configuration reading unit, a node generation unit, an edge weight initialization unit, a dynamic topology update unit, and a graph data caching unit.

[0114] The branch carrying capacity assessment module is used to calculate the carrying current manifold coordinates of each energy storage branch based on a multi-physics manifold graph. This module calculates the electrical manifold coordinates, thermal manifold coordinates, aging manifold coordinates, and topological manifold coordinates separately, and obtains the carrying current manifold coordinates of each branch through a fusion function. This module can be implemented as a physical weight graph embedding unit, a graph neural network inference unit, or a graph attention fusion unit. If a graph neural network inference unit is used, the input to this unit is a node feature matrix and a multi-domain adjacency matrix, and the output is a branch carrying capacity representation, which needs to be processed by safety boundary projection.

[0115] The intent derivation module generates operational intents based on EMS dispatch instructions, photovoltaic forecasts, load curves, backup power strategies, and grid connection / off-grid switching signals. This module can also receive load-side circuit breaker status, impact load initiation signals, and fire safety signals. The specific implementation of the intent derivation module may include a dispatch instruction parsing unit, a prediction curve parsing unit, an operational state machine, a task intensity calculation unit, and a safety priority judgment unit. The dispatch instruction parsing unit reads the EMS total power instruction and dispatch plan; the prediction curve parsing unit calculates the future trends of photovoltaic power and load power changes within a window; the operational state machine determines operational intents such as peak shaving discharge, photovoltaic charging, backup power voltage maintenance, impact load response, or fire load reduction; the task intensity calculation unit generates task intensity; and the safety priority judgment unit overrides ordinary operational intents when a safety event occurs.

[0116] The flexible current sharing optimization module generates the target current and control quantities for each energy storage branch. This module receives the manifold coordinates of the current carrying capacity from the branch carrying capacity assessment module and the operational intent from the intent deduction module. It calculates the thermal margin, aging margin, impedance margin, air cooling margin, insulation margin, and future intent matching margin for each branch, and generates the branch's comprehensive carrying capacity factor. Subsequently, the module calculates the target current based on the total charging and discharging current command of the energy storage distribution cabinet and the comprehensive carrying capacity factor of each branch, and generates the control quantities to be issued through constraint optimization or amplitude limiting projection. The specific implementation of the flexible current sharing optimization module may include a margin calculation unit, a comprehensive carrying capacity factor calculation unit, a target current allocation unit, a smoothing constraint unit, and a control quantity mapping unit. The control quantity mapping unit converts the target current into a DC / DC current setpoint, PCS droop coefficient, contactor opening and closing priority, fan speed, current limiting threshold, and fault bypass strategy.

[0117] The safety interlock module is used to perform protective control when there is branch overcurrent, busbar overtemperature, abnormal contactor voltage drop, abnormal fuse temperature, abnormal insulation impedance, abnormal PCS current limiting, fan failure, or fire signal triggering. The safety interlock module receives the control quantity generated by the flexible current sharing optimization module and performs safety verification on it. If the control quantity meets the safety boundary, the safety interlock module allows the control quantity to be issued. If the control quantity does not meet the safety boundary, the safety interlock module can limit the target current of the high-risk branch to a safe value, prohibit the contactor of the high-risk branch from closing, increase the fan speed, execute branch bypass, reduce PCS power, or output the overall cabinet derating capability to the EMS. The specific implementation of the safety interlock module may include a hard threshold verification unit, an anomaly type identification unit, a protection action generation unit, and an alarm output unit.

[0118] The online self-calibration module updates the graph edge weights based on the bias current residual, temperature rise residual, and contactor voltage drop residual. After the control input is executed, this module receives the measured branch current, measured temperature rise, and measured contactor voltage drop, and compares them with the target current, predicted temperature rise, and predicted voltage drop to generate a residual loss. The online self-calibration module updates the resistive coupling edge, thermal diffusion coupling edge, estimated equivalent contact resistance of the contactor, and branch carrying capacity parameters based on the residual loss. The specific implementation of this module may include a residual calculation unit, a restricted learning unit, an edge weight update unit, and a parameter boundary verification unit. The restricted learning unit updates parameters according to safety boundaries, preventing the self-calibration process from exceeding the hard constraints set by the safety interlocking module.

[0119] There are clear data and control relationships between the above modules. The multi-source data acquisition module outputs node characteristics and status data to the multi-physics manifold construction module. The multi-physics manifold construction module outputs the graph structure to the branch carrying capacity assessment module. The branch carrying capacity assessment module outputs the manifold coordinates of the current carrying capacity to the flexible current sharing optimization module. The intent deduction module outputs the operating intent to the flexible current sharing optimization module. The flexible current sharing optimization module outputs the control quantity to the safety interlocking module. The safety interlocking module sends the verified control quantity to the DC / DC or current sharing execution module, PCS access unit, contactor, duct temperature control module, and load-side power distribution circuit. The online self-calibration module feeds back the graph edge weights and margin parameters to the multi-physics manifold construction module and the branch carrying capacity assessment module based on the execution results.

[0120] In one specific distribution cabinet structure, eight energy storage branches are installed within the cabinet. Each energy storage branch is connected to a DC busbar via a branch contactor, fuse, and bidirectional DC / DC module. The DC busbar connects to a PCS, which in turn connects to the AC power distribution system. Each energy storage branch is equipped with a branch current sensor and a branch voltage sampling terminal. A busbar temperature sensor is installed between every two branch connection points on the busbar, and voltage drop sampling terminals are installed at both ends of the branch contactor contacts. A vertical air duct is installed at the back of the cabinet, and a wind speed sensor is installed at the fan outlet. Temperature zone sensors are installed in the energy storage branch area, busbar area, and PCS area, respectively. The edge computing controller communicates with the DC / DC module and PCS via a CAN bus, reads temperature, voltage drop, insulation, and wind speed data via RS485 or a local sampling interface, and communicates with the EMS via Ethernet. This structure can support multi-physics manifold construction, intent deduction, and flexible current sharing control as described in the previous embodiments.

[0121] The distribution cabinet described in this embodiment is not limited to eight energy storage branches, nor is it limited to each branch being equipped with an independent DC / DC module. For distribution cabinets with a small number of branches or those sharing PCS control, flexible current sharing can be achieved through PCS droop coefficient, branch contactor switching priority, branch current limiting threshold, and fan control. For modular energy storage distribution cabinets with a large number of branches, multiple branches can be divided into busbar sections or duct temperature zones, and local multi-physics manifold maps can be established in the edge computing controller, and then the local maps can be merged into a whole cabinet manifold map. For outdoor cabinet-type energy storage systems, humidity, condensation, insulation, and ambient temperature sampling can be added, and these can be used as node features in insulation margin and thermal margin calculations.

[0122] Through the distribution cabinet structure of this embodiment, the multi-source sampling, graphical model construction, intent deduction, flexible current sharing optimization, safety interlocking, and online self-correction in the aforementioned methods all have corresponding hardware and functional modules. This distribution cabinet can generate branch-level current distribution targets based on the real-time status inside the cabinet and future operating intentions, and execute control through DC / DC modules, PCS, contactors, fans, and load-side power distribution circuits, thereby forming a dynamic current sharing control device at the energy storage distribution cabinet level. Example 5

[0123] This embodiment describes the implementation process of the technical solution of this application, taking the operation scenario of a photovoltaic-storage distribution cabinet in an industrial and commercial park as an example. The energy storage distribution cabinet described in this embodiment is connected to the low-voltage distribution busbar of the factory area and operates in conjunction with rooftop photovoltaic systems, the public power grid, critical loads, and ordinary loads. Its main tasks include peak shaving, photovoltaic power consumption, short-term backup power, and response to impact loads. Figure 7 As shown, the energy storage distribution cabinet is equipped with multiple energy storage branches, DC / DC module groups, DC busbars, PCS, load-side power distribution circuits, heat dissipation ducts, sensor sampling units, and edge computing controllers; externally, it establishes electrical or communication connections with the EMS dispatch system, photovoltaic power generation units, public power grid, critical loads, and ordinary loads. Figure 8 This is used to illustrate the processing effect trends of this embodiment in terms of branch flow deviation, temperature rise distribution, branch load matching, and online self-correction.

[0124] This embodiment applies to a commercial and industrial energy storage distribution cabinet with a rated power of 100kW and a nominal DC bus voltage of 750V. The cabinet includes eight energy storage branches, each connected to a battery cluster and connected to a DC busbar via a bidirectional DC / DC module. Each energy storage branch has a continuous current of 85A, a short-time current of 110A, a branch contactor voltage drop threshold of 80mV, a busbar connection point temperature rise warning threshold of 18℃, a busbar connection point temperature rise lockout threshold of 25℃, and a DC side insulation impedance safety lower limit of 500kΩ. The PCS has a rated AC output power of 100kW and supports grid-connected operation, off-grid backup power, and power derating control. Plant loads include continuously operating lighting and control loads, periodically starting air compressor loads, and production line motor loads, with short-term power surges during air compressor startup.

[0125] The technical requirements of this embodiment are as follows: During peak-shaving discharge, the energy storage distribution cabinet should limit the bias current of each branch while meeting the EMS power command, avoiding a single branch from bearing a large current for a long time; during photovoltaic power consumption, the energy storage distribution cabinet should allocate charging current according to the SOC, temperature, internal resistance, and contactor voltage drop of each branch, avoiding charging allocation only based on low or high SOC; before the air compressor or production line motor starts, the energy storage distribution cabinet should identify the impulse load response intention in advance and retain the instantaneous discharge margin of branches with low temperature, low resistance, and high SOH; when the busbar temperature rises, contactor voltage drop, or insulation impedance is abnormal, the energy storage distribution cabinet should promptly reduce the current target of the abnormal branch, and if necessary, perform bypass or derating of the entire cabinet. The above technical requirements mean that the control target in this embodiment is not an absolute average of the current of each branch, but rather the allocation of current within the safety boundary according to the branch carrying capacity and future operating intention.

[0126] The hardware selection for this embodiment is as follows: Each energy storage branch is equipped with a Hall current sensor with a range of ±150A and a sampling accuracy of 0.5 class; each energy storage branch is equipped with a terminal voltage sampling channel, and provides SOC, SOH, and estimated internal resistance to the edge computing controller through the battery cluster management unit. The DC busbar is equipped with patch temperature sensors at the eight branch access points and the PCS access point, and infrared temperature measurement points are set at the middle and ends of the busbar to collect the temperature rise at the busbar connection points. Voltage drop sampling terminals are set at both ends of each branch contactor contact, with a sampling range of 0 to 200mV. A vertical air duct is set at the back of the cabinet, with a filter status detection device at the duct inlet and a wind speed sensor at the fan outlet. Temperature zone sensors are set in the energy storage branch area, busbar area, and PCS area respectively. An insulation detection unit is configured on the DC side for periodically detecting the insulation impedance of the positive and negative busbars to ground. The edge computing controller adopts an industrial embedded controller and has a local sampling interface, a CAN communication interface, an RS485 communication interface and an Ethernet interface, which are respectively connected to the DC / DC module, PCS, fan controller, EMS and load-side circuit breaker status acquisition unit.

[0127] In this embodiment, the sampling period for branch current and terminal voltage is set to 100ms, the sampling period for busbar temperature and cabinet temperature zone is set to 1s, the sampling period for contactor voltage drop is set to 500ms, the sampling period for wind speed is set to 1s, and the insulation impedance detection period is set to 5s. The current sharing control period is set to 1s, the short-time intentional projection window is set to 30s for impact load response and grid-connected / off-grid switching preparation, the medium-time intentional projection window is set to 10min for peak shaving discharge and photovoltaic absorption, and the backup power voltage protection strategy window is set to 30min. The lower limit C_min of the branch comprehensive carrying capacity factor C_i(t) is set to 0.15 under normal operating conditions and can be set to 0 under fault blocking conditions; the upper limit C_max is set to 1.00. The thermal margin weight w_T is set to 0.25, the aging margin weight w_A to 0.18, the impedance margin weight w_R to 0.20, the air-cooling margin weight w_F to 0.12, the insulation margin weight w_Z to 0.10, and the future intended matching margin weight w_P to 0.15. In the impact load response scenario, w_P is increased to 0.25, and the non-critical weights other than the aging margin and air-cooling margin are correspondingly reduced. In the busbar temperature rise early warning scenario, w_T is increased to 0.35, making the temperature rise have a stronger constraint on the branch target current. The above parameters are used to illustrate the configuration method of this embodiment and do not limit other energy storage cabinets to using the same values.

[0128] In this scenario, the execution flow of this embodiment is as follows. After the energy storage distribution cabinet is started, the edge computing controller first reads the cabinet configuration file to confirm the numbers and topological relationships of the 8 energy storage branches, 8 DC / DC modules, 9 busbar temperature measurement points, 8 contactor voltage drop sampling channels, 3 cabinet internal temperature zones, 2 fan outlet wind speed points, and load-side circuit breaker status points. Subsequently, the edge computing controller loads the branch rated current, busbar allowable temperature rise, contactor allowable voltage drop, insulation impedance lower limit, fan rated wind speed, and PCS power boundary. The above configuration is used to generate the initial node set and initial edge weights of the multiphysics manifold graph.

[0129] During operation, the edge computing controller collects multi-source data according to the control cycle and performs time alignment on the data from different cycles. At a certain moment, the SOC of the eight energy storage branches were 62%, 66%, 58%, 64%, 71%, 55%, 68%, and 60%, respectively, and the SOH were 96%, 94%, 91%, 97%, 95%, 90%, 96%, and 93%, respectively. The estimated internal resistance of branch 3 was about 18% higher than the average, the SOH of branch 6 was lower, and the SOC and temperature of branch 5 were higher. Busbar temperature measurement showed that the temperature rise at the busbar connection point between branch 2 and branch 3 was 16.5℃, close to the warning threshold of 18℃; the contactor voltage drop of branch 7 was 72mV, close to the allowable voltage drop threshold of 80mV; the contactor voltage drops of the remaining branches were between 35mV and 55mV. Wind speed sampling showed that the wind speed in the wind zones corresponding to branches 1 to 4 was 2.1 m / s, and the wind speed in the wind zones corresponding to branches 5 to 8 was 2.8 m / s. Insulation testing showed that the DC side insulation resistance was 1.8 MΩ, which is higher than the safety lower limit.

[0130] The edge computing controller constructs a multi-physics manifold graph based on the above data. Battery clusters, DC / DC modules, busbar connection points, contactors, fuses, cabinet temperature zones, fan outlets, and load branches are set as nodes. The connection path from the branch to the busbar is set as a resistive coupling edge, the connection between the busbar connection point and the cabinet temperature zone is set as a thermal diffusion coupling edge, the branch's state of equilibrium (SOH) and historical temperature rise are set as aging coupling relationships, and the contactor and load-side circuit breaker status are set as topology switching coupling edges. The edge computing controller calculates the electrical manifold coordinates, thermal manifold coordinates, aging manifold coordinates, and topology manifold coordinates of each branch and forms the current-carrying manifold coordinates. Due to the high internal resistance of branch 3 and the high temperature rise of the adjacent busbar connection point, its overall carrying capacity is reduced; due to the high voltage drop of the contactor in branch 7, its impedance margin and opening / closing priority are reduced; due to the low temperature, good air-cooling conditions, and high SOH of branch 5, its carrying factor is increased.

[0131] Meanwhile, the edge computing controller receives EMS scheduling information. The EMS issues a peak-shaving discharge command for the next 10 minutes, with a target discharge power of 80kW; photovoltaic forecasts show a decrease in photovoltaic output over the next 10 minutes, and the plant load curve shows the production line load remains high; the load-side circuit breaker status indicates the air compressor will start in approximately 20 seconds, with the startup phase expected to last 8 to 12 seconds. The intent extrapolation module marks the medium-time window as peak-shaving discharge and the short-time window as impact load response, setting the short-time window task intensity to a level higher than normal peak-shaving discharge. Because no fire alarm signal, insulation abnormality, or busbar over-temperature lockout was detected, the safety priority did not cover normal power tasks, but the busbar temperature rise warning between branch 2 and branch 3 increased the thermal margin weight.

[0132] When generating the flexible current sharing target, the edge computing controller calculates the comprehensive load factor for each of the eight branches. After calculation, the comprehensive load factors for branches 1 to 8 are 0.78, 0.63, 0.49, 0.82, 0.92, 0.58, 0.66, and 0.76, respectively. Since the target discharge power is 80kW and the DC bus voltage is approximately 750V, the corresponding total discharge current is approximately 106.7A. If an average current sharing method is used, the target current for each branch is approximately 13.3A. After adopting the flexible current sharing method of this embodiment, the discharge target currents for branches 1 to 8 are approximately 14.7A, 11.9A, 9.2A, 15.5A, 17.4A, 11.0A, 12.5A, and 14.4A, respectively. This allocation result reduces the current targets for branches 3 and 7, increases the participation ratio of branches 5 and 4, and ensures that the current of each branch still meets the total power requirement.

[0133] When the intent simulation module detects that the air compressor is about to start, the edge computing controller increases the fan speed of the corresponding air zones in branches 4 and 5 5 seconds before startup, raising the corresponding air velocity from 2.8 m / s to 3.5 m / s, and increases the short-time discharge carrying factor of branches 4 and 5 to within the safe upper limit. Branch 3 continues to reduce load due to internal resistance and temperature rise factors, while branch 7 does not participate in short-time high-rate load due to the contactor voltage drop approaching the threshold. When the air compressor starts, the PCS droop coefficient is adjusted to make the PCS power response consistent with the branch flexibility target. After the impact load lasts for about 10 seconds, the system gradually restores the target current of each branch to the distribution state under the peak-shaving discharge window to avoid sudden changes in current setting.

[0134] After control execution, the edge computing controller collects measured current, measured temperature rise, and contactor voltage drop. In the first round of execution, the measured current of branch 5 was 17.0A, the target current was 17.4A, and the bias current residual was small; the measured current of branch 3 was 10.5A, higher than the target current of 9.2A, indicating that its actual current path or DC / DC response differed from the model estimate; the temperature rise at the busbar connection point between branch 2 and branch 3 increased from 16.5℃ to 17.2℃, which did not reach the early warning blocking threshold, but was higher than the model prediction of 16.8℃; the contactor voltage drop of branch 7 increased from 72mV to 76mV, close to the allowable voltage drop of 80mV. Based on this, the online self-calibration module increased the weight of the relevant resistive coupling edge of branch 3, increased the thermal diffusion coupling weight of the busbar connection point between branch 2 and branch 3, and reduced the opening and closing priority and comprehensive load factor of branch 7 in subsequent control cycles.

[0135] After approximately 15 minutes of operation, the system enters the photovoltaic charging window. At this time, the photovoltaic output recovers, the EMS provides a charging power of 60kW, the DC bus voltage is approximately 760V, and the corresponding total charging current is approximately 78.9A. The SOC of branch 5 has risen to 74%, while the SOC of branch 6 remains at 56% but its SOH is low. Branch 3 has a high internal resistance, and the corresponding busbar temperature rise is still at a high level. If charging is only performed based on the low SOC, branch 6 may receive a higher charging current. In this embodiment, the flexible current sharing optimization module considers SOH, temperature, impedance, and air cooling margin simultaneously, limiting the charging target of branch 6 within a safe range and increasing the charging ratio of branches 1, 4, and 8. Therefore, the photovoltaic charging process will not subject the low-SOC branch to excessive charging current due to its low SOH or high impedance.

[0136] To illustrate the effectiveness of this embodiment compared to existing methods, a comparison was conducted using the same energy storage distribution cabinet under similar load curves and photovoltaic conditions. Comparison Method 1 employed traditional average current sharing and a fixed PCS droop coefficient, with branch current targets primarily distributed evenly according to the number of participating branches, and protection action only initiated after overcurrent or overtemperature thresholds were reached. Comparison Method 2 employed the multi-physics manifold diagram, intent deduction, and flexible dynamic current sharing method described in this embodiment. Both methods were operated in three phases: peak-shaving discharge, air compressor startup, and photovoltaic charging, with 30 minutes of operational data collected. The comparison records are shown in the table below, where the values ​​represent results under the test or simulation conditions of this embodiment, used to illustrate the treatment effect and not constituting a numerical limitation on the protection range.

[0137] As can be seen from the above comparison, under the traditional average current sharing method, although branch 3 has a higher internal resistance and a higher temperature rise of the adjacent busbar, it will still bear a current close to the average value during the peak-shaving discharge phase, causing the temperature rise of the busbar connection point to continue to increase. During the air compressor startup phase, the PCS responds according to a fixed droop coefficient. The low-resistance branch and the branch with a faster response bear a larger current for a short time, and the voltage drop of the contactor in branch 7 exceeds the allowable threshold and triggers protection. After adopting this embodiment, the carrying factors of branches 3 and 7 have been reduced before the task arrives. Branches 4 and 5 bear more short-term current due to their low temperature, low resistance, and higher air-cooling margin. The pre-cooling of the fan makes the temperature rise of the corresponding temperature zone rise more slowly. As a result, the maximum temperature rise of the busbar, the maximum voltage drop of the contactor, and the number of PCS current-limiting actions are all reduced.

[0138] Figure 8 The first sub-graph in the diagram corresponds to the aforementioned branch current deviation changes. Under the traditional average current sharing method, the current deviation decreases slowly during load changes and the start-up of impact loads, and new fluctuations occur when voltage drop anomalies occur in some branches. In this embodiment, the flexible current sharing target is updated with the branch load-bearing factor, and the current deviation residual gradually decreases over multiple control cycles. Figure 8The second sub-diagram corresponds to the temperature rise distribution in the busbar and contactor areas. In the conventional approach, the busbar connection point between branch 2 and branch 3 and the contactor in branch 7 are in a high temperature rise region; in this embodiment, by reducing the target current of the corresponding branch and increasing the local fan speed, the temperature rise in the above-mentioned areas is limited. Figure 8 The third sub-graph in the diagram corresponds to the relationship between the overall load factor and the actual distributed current. Branches with higher load factors receive higher current distribution, while branches with lower load factors are deloaded. Figure 8 The fourth subgraph in the diagram corresponds to the convergence trend of the bias flow residual and the temperature rise residual during the online self-calibration process.

[0139] This embodiment also includes settings for quality recording and subsequent self-calibration. At the end of each control cycle, the edge computing controller records branch node characteristics, busbar temperature rise, contactor voltage drop, insulation impedance, operating intention label, comprehensive load-bearing factor, target current, actual current, PCS droop coefficient, fan speed, and residual value. For branch 7 where the contactor voltage drop is close to the threshold, the system marks it as a branch requiring maintenance and review, and reduces its opening / closing priority in subsequent control cycles. For branch 2 to branch 3 where the busbar connection point temperature rise exceeds the predicted value, the system increases the heat diffusion edge weight for that busbar section and reduces the thermal margin of the corresponding branch before the next peak-shaving discharge. Through this recording process, the energy storage distribution cabinet can gradually develop load-bearing capacity parameters adapted to its own cabinet connection state, air duct state, and battery branch state during long-term operation.

[0140] This embodiment illustrates that the technical solution of this application does not only balance the SOC at the battery management layer, nor does it only adjust the fixed droop coefficient at the PCS side. Instead, it incorporates multiple branch energy storage units, DC / DC modules, PCS, busbars, contactors, fuses, air duct temperature zones, and load-side power distribution circuits within the energy storage distribution cabinet into the control process. Through the synergy of multi-physical manifold diagrams, intent extrapolation, flexible current sharing targets, safety interlocking, and online self-correction, the energy storage distribution cabinet can reduce the risks of local current bias, local temperature rise, and continuous load on abnormal branches while meeting the total power demand.

[0141] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A flexible dynamic current sharing method for energy storage that integrates multi-physics manifold diagrams and intent extrapolation, characterized in that, include: When the energy storage distribution cabinet is in operation, operational data of multiple energy storage branches, PCS or DC / DC modules, DC busbars, contactors, fuses, duct temperature zones, and load-side power distribution circuits within the cabinet are collected. A multi-physical manifold diagram (MPM) is constructed based on this data, with electrical components, thermal management components, and load branches within the cabinet as nodes, and resistive coupling, thermal diffusion coupling, aging coupling, and topology switching coupling as edges. The manifold coordinates of the current carrying capacity of each energy storage branch are calculated based on the MPM. EMS dispatch instructions, photovoltaic forecasts, load curves, backup power strategies, and grid connection / off-grid switching signals are obtained to predict the operational intent within the future control window. Based on the current-carrying manifold coordinates and operating intentions, generate flexible current sharing targets for each energy storage branch; output DC / DC current setpoints, PCS droop coefficients, branch contactor opening and closing priorities, fan speeds, current limiting thresholds, and fault bypass strategies according to the flexible current sharing targets. Collect the flow deviation residual and temperature rise residual after execution, and update the graph edge weights of the multiphysics manifold graph based on the flow deviation residual and temperature rise residual.

2. The flexible dynamic current sharing method for energy storage according to claim 1, characterized in that, The system collects operational data from multiple energy storage branches, PCS or DC / DC modules, DC busbars, contactors, fuses, duct temperature zones, and load-side power distribution circuits within the cabinet. This includes: collecting branch current, terminal voltage, SOC, SOH, estimated internal resistance, and branch temperature for each energy storage branch according to the same control cycle; collecting busbar node temperature and infrared temperature rise at busbar connection points of the DC busbar; collecting contact voltage drop in the closed state of the contactors and calculating the equivalent contact resistance; and collecting fuse temperature, fan outlet velocity, cabinet temperature zones, cabinet humidity, and insulation resistance. The system monitors the resistance, PCS power command, DC / DC operating status, load-side circuit breaker status, and grid-connected / off-grid switch status. It performs time alignment, outlier removal, and unit normalization on the collected data to obtain branch status characteristics, cabinet thermal status characteristics, topology switch status characteristics, and load-side power distribution status characteristics corresponding to the current control cycle. When sensor data for a branch is missing or exceeds the physical allowable range, it uses the valid data from the previous control cycle of that branch and the changing trends of adjacent branches in the same busbar section to generate a substitute status value, thus reducing the reliability of the data for that branch.

3. The flexible dynamic current sharing method for energy storage according to claim 1, characterized in that, When constructing a multi-physics manifold, the battery cluster, energy storage branch access point, DC / DC module, PCS input point, busbar connection point, contactor, fuse, cabinet temperature zone, fan outlet, and load branch are respectively set as graph nodes; the connection relationship between the energy storage branch and the busbar, the conductive path between the busbar connection points, and the series path of the contactor and fuse are set as resistive coupling edges; the heat transfer relationship between the battery cluster, busbar connection point, contactor, fuse, cabinet temperature zone, and fan outlet is set as thermal diffusion coupling edges; the correlation between branch current ratio, temperature rise history, SOC operating range, and SOH change trend is set as aging coupling edges; the influence of contactor, circuit breaker, PCS operating mode, and DC / DC switching state on the current path is set as topology switching coupling edges; the node characteristics of the graph nodes include branch current, terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, busbar infrared temperature rise, contactor voltage drop, fan outlet wind speed, humidity, insulation impedance, and load-side circuit breaker status.

4. The flexible dynamic current sharing method for energy storage according to claim 1, characterized in that, The current-carrying manifold coordinates of each energy storage branch are calculated based on the multi-physics manifold diagram, including: calculating electrical manifold coordinates based on resistive coupling edges and branch electrical state characteristics; calculating thermal manifold coordinates based on thermal diffusion coupling edges and cabinet thermal state characteristics; calculating aging manifold coordinates based on aging coupling edges and branch SOH variation trends; and calculating topological manifold coordinates based on topology switch coupling edges and load-side power distribution state characteristics. The electrical manifold coordinates, thermal manifold coordinates, aging manifold coordinates, and topological manifold coordinates are input into a fusion function to obtain the current-carrying manifold coordinates of each energy storage branch. The fusion function adopts weighted fusion, gated fusion, or graph attention fusion methods and is constrained by the branch rated current, busbar allowable temperature, contactor allowable voltage drop, and insulation impedance lower limit to match the output current-carrying manifold coordinates with the current safety boundary within the cabinet.

5. The flexible dynamic current sharing method for energy storage according to claim 1, characterized in that, The simulation of operational intent within the future control window includes: determining the total power change trend within the future control window based on EMS scheduling instructions; determining the charging and discharging direction and power duration based on photovoltaic forecasts and load curves; determining the branch redundancy and SOC range to be retained based on the backup power strategy; determining the PCS operating mode change based on grid connection / off-grid switching signals; and determining the short-term high-rate discharge demand based on the load-side circuit breaker status and impact load start signals. The future control window is divided into multiple continuous control cycles, and an operational intent label, target power, duration, safety priority, and task intensity are generated for each control cycle. The operational intent label includes one of the following: peak shaving discharge, photovoltaic charging, low power maintenance, reverse charging, backup power voltage maintenance, black start preparation, impact load response, and fire load reduction. When a fire alarm, insulation abnormality, or busbar over-temperature signal is triggered, fire load reduction or fault derating is determined as the operational intent with the highest safety priority.

6. The flexible dynamic current sharing method for energy storage according to claim 1, characterized in that, Generate the flexible current sharing targets for each energy storage branch according to the current-carrying manifold coordinates and the operating intention, including: calculating the branch comprehensive carrying factor \(C_i(t)\) based on the thermal margin, aging margin, impedance margin, air-cooling margin, insulation margin and future intention matching margin of each energy storage branch, where \(C_i(t)=\text{clip}(w_T M_i^T + w_A M_i^A+w_R M_i^R + w_F M_i^F+w_Z M_i^Z + w_P M_i^P,C_{min},C_{max})\), \(M_i^T\) represents the thermal margin, \(M_i^A\) represents the aging margin, \(M_i^R\) represents the impedance margin, \(M_i^F\) represents the air-cooling margin, \(M_i^Z\) represents the insulation margin, \(M_i^P\) represents the future intention matching margin, \(w_T\), \(w_A\), \(w_R\), \(w_F\), \(w_Z\) and \(w_P\) represent the corresponding weights, and \(\text{clip}\) means to limit the calculation result between \(C_{min}\) and \(C_{max}\); calculate the target current \(I_i^{tar}(t)\) of the \(i\)-th energy storage branch according to \(I_i^{tar}(t)=I_{total}(t)\cdot[C_i(t)\cdot I_{i,max}] / \sum_j[C_j(t)\cdot I_{j,max}]\), where \(I_{i,max}\) represents the safe current limit value of the \(i\)-th energy storage branch in the current control period, \(C_j(t)\) and \(I_{j,max}\) respectively represent the comprehensive carrying factor and the safe current limit value of the \(j\)-th energy storage branch participating in the operation currently, and \(\sum_j[C_j(t)\cdot I_{j,max}]\) represents the summation of all energy storage branches participating in the operation currently and not bypassed, fault-blocked or disconnected by the contactor.

7. The flexible dynamic current sharing method for energy storage according to claim 6, characterized in that, The thermal margin \(M_i^T\) is calculated according to \(M_i^T=\text{clip}((T_{lim}-T_i-\Delta T_i^{pred}) / (T_{lim}-T_{ref}),0,1)\), where \(T_{ref}<T_{lim}\), \(T_{lim}\) represents the upper limit of the allowable temperature of the branch or busbar, \(T_i\) represents the current branch temperature or the corresponding busbar node temperature, \(\Delta T_i^{pred}\) represents the predicted temperature rise obtained from the predicted current and the air-cooling state within the future control window, and \(T_{ref}\) represents the reference temperature; the impedance margin \(M_i^R\) is calculated according to \(M_i^R=\text{clip}(R_{ref} / (R_i + R_{bus_i}+R_{con_i}+\varepsilon),0,1)\), \(R_{ref}\) represents the reference impedance, \(R_i\) represents the internal resistance of the branch, \(R_{bus_i}\) represents the equivalent impedance of the busbar, \(R_{con_i}\) represents the equivalent contact resistance of the contactor, and \(\varepsilon\) represents the correction amount to prevent the denominator from being zero; when the operating intention is impact load response, increase the future intention matching margin of the branches with lower temperature, lower resistance and higher SOH; when the operating intention is standby power voltage maintenance, reduce the current discharge target of the branches selected as the subsequent emergency redundant branches; when the operating intention is fire load reduction, set the comprehensive carrying factor of the abnormal temperature area or the branch with abnormal insulation to zero or the safe derating value.

8. A flexible dynamic current sharing distribution cabinet for energy storage that integrates multi-physics manifold diagrams and intent extrapolation, characterized in that, The system includes a cabinet, an energy storage branch access unit, a DC busbar, a PCS access unit, a DC / DC or current sharing execution module, contactors, fuses, a load-side power distribution circuit, branch current sensors, branch voltage sampling units, busbar temperature sensors, contactor voltage drop sampling units, insulation detection units, a duct temperature control module, a communication module, and an edge computing controller. The energy storage branch access unit is connected to the DC busbar via contactors and fuses. The DC / DC or current sharing execution module is located between the energy storage branch access unit and the DC busbar. The PCS access unit is connected to the DC busbar and used to exchange power with the AC-side power distribution system. The load-side power distribution circuit is connected to the output side of the energy storage distribution cabinet via a circuit breaker. The edge computing controller executes the flexible dynamic current sharing method for energy storage as described in any one of claims 1 to 7 and sends control commands to the DC / DC or current sharing execution module, the PCS access unit, contactors, the duct temperature control module, and the load-side power distribution circuit based on the calculation results.

9. The energy storage flexible dynamic current sharing distribution cabinet according to claim 8, characterized in that, The edge computing controller includes a multi-source data acquisition module, a multi-physics manifold construction module, a branch carrying capacity assessment module, an intent deduction module, a flexible current sharing optimization module, a safety interlocking module, and an online self-calibration module. The multi-source data acquisition module is used to acquire branch current, terminal voltage, SOC, SOH, estimated internal resistance, branch temperature, busbar infrared temperature rise, contactor voltage drop, fan outlet wind speed, cabinet humidity, insulation impedance, PCS power command, and load-side circuit breaker status. The multi-physics manifold construction module is used to establish a multi-physics manifold including resistive coupling edges, thermal diffusion coupling edges, aging coupling edges, and topology switch coupling edges. The system comprises the following modules: a branch carrying capacity assessment module for calculating the current carrying capacity manifold coordinates of each energy storage branch based on the multi-physical manifold diagram; an intent deduction module for generating operational intent based on EMS dispatch instructions, photovoltaic forecasts, load curves, backup power strategies, and grid connection / disconnection switching signals; a flexible current sharing optimization module for generating target currents and control quantities for each energy storage branch; a safety interlocking module for performing derating, bypass, or prohibition of closing control when there is branch overcurrent, busbar overtemperature, abnormal contactor voltage drop, or abnormal insulation impedance; and an online self-correction module for updating graph edge weights based on bias current residuals, temperature rise residuals, and contactor voltage drop residuals.

10. The energy storage flexible dynamic current sharing distribution cabinet according to claim 8, characterized in that, The air duct temperature control module includes a fan, an air duct, a fan outlet wind speed sensor, and an in-cabinet temperature zone sensor. The air duct is arranged along multiple energy storage branches, DC busbars, and the area where the PCS access unit is located. The fan outlet wind speed sensor is used to provide air cooling margin calculation data to the edge computing controller, and the in-cabinet temperature zone sensor is used to provide thermal manifold coordinate calculation data to the edge computing controller. The contactor voltage drop sampling unit is connected to both ends of the contactor contacts in the branch circuit and collects the contact voltage drop when the contactor is closed. The edge computing controller calculates the equivalent contact resistance of the contactor based on the contact voltage drop and the branch current. The insulation detection unit is used to detect the DC side insulation impedance and output the insulation status to the safety interlocking module. The communication module is used to communicate with the EMS, PCS, DC / DC module, wind turbine controller and load-side circuit breaker, so that the energy storage distribution cabinet can synchronously adjust the branch current, PCS droop coefficient, contactor opening and closing priority, wind turbine speed, current limiting threshold and fault bypass strategy according to the flexible current sharing target.

Images