A module-level distributed active equalization method and system

Abstract

The application relates to the technical field of energy storage systems, in particular to a module-level distributed active balancing method and system. The module-level distributed active balancing method and system comprises an SOC estimation and uncertainty quantification module, a balancing potential field construction module, a balancing channel health assessment module, a working condition coupling pre-judgment module, a virtual balancing token management module, a three-level balancing scheduling module, a potential field gradient balancing execution module, an abnormality detection and potential field isolation module, a strategy evolution and versioning module and a balancing result recording and reporting module. Compared with the prior art, the discrete 'collection-computation-distribution' process is replaced by a continuous potential field gradient driven self-organizing energy flow through the balancing potential field modeling, the balancing response time is shortened, and the balancing system is naturally adapted to the topological expansion of any scale system; through the balancing channel health (BHI) online assessment, the balancing path selection perceives the channel degradation degree, avoids selecting the degraded channel, and prolongs the overall life of the balancing system.

Description

Technical Field

[0001] This invention relates to the field of energy storage system technology, specifically a module-level distributed active balancing method and system. Background Technology

[0002] With the large-scale application of electrochemical energy storage technology in power systems, the installed capacity of energy storage power stations is rapidly increasing from the megawatt-hour (MWh) level to the hundred MWh level. Large-scale energy storage systems typically adopt a multi-stage series-parallel structure, composed of thousands to tens of thousands of lithium-ion battery cells, organized in a hierarchical manner of "cell → module → cluster → stack". During long-term operation, due to factors such as differences in cell manufacturing, temperature gradients caused by installation location, inconsistent self-discharge rates, and differences in capacity decay rates, the state of charge (SOC) between cells and modules gradually deviates, leading to a decrease in the overall usable capacity of the system, a shortened cycle life, and in severe cases, even overcharge and over-discharge safety risks.

[0003] Battery balancing control is the core method for solving battery consistency issues. Existing active balancing technologies have the following drawbacks and limitations in large-scale energy storage system applications:

[0004] First, existing active balancing schemes all adopt a "command-and-control" paradigm: after the central controller or host computer collects the SOC of each node, it calculates the balancing command through deterministic rules or optimization algorithms and issues it one by one. This command-and-control paradigm is essentially a centralized decision-making architecture, which faces the combinatorial explosion problem when the number of nodes increases, and lacks the ability to continuously model the global state of the system.

[0005] Second, existing equalization schemes only focus on the state of health (SOH) of the battery itself, ignoring the degradation of the equalization channel itself. During long-term operation, equalization converters experience problems such as aging of switching devices, increased contact resistance, and drift of the saturation point of magnetic components, causing the actual transfer efficiency of the equalization channel to gradually deviate from the design value. Existing schemes use fixed efficiency parameters for path selection, which will lead to the selection of suboptimal or even ineffective paths after the equalization channel degrades.

[0006] Third, the existing balancing triggering strategy is purely reactive: balancing is only initiated after the SOC deviation has exceeded the threshold. This "deteriorate first, then fix" approach cannot take advantage of foreseeable low-load windows during system operation (such as off-peak charging intervals of energy storage power stations and off-peak periods for industrial and commercial energy storage), leading to mutual interference between balancing operations and main power charging and discharging, thus reducing the overall system efficiency.

[0007] Fourth, existing distributed load balancing control schemes employ traditional "request-response" or "master-slave" communication architectures, requiring a central coordinator to arbitrate concurrent load balancing requests from multiple modules. The presence of a central coordinator introduces the risk of single point of failure and creates a communication bottleneck in high-density load balancing scenarios. While existing consensus algorithms can achieve decentralized control, their convergence speed is limited by network topology, making them unsuitable for load balancing scenarios requiring rapid response.

[0008] Fifth, existing adaptive equilibrium parameter schemes only support parameter adjustments in a monotonic direction (such as "lowering the trigger threshold when SOH decreases"), lacking the ability to systematically evaluate and backtrack on the effects of historical strategies. When the parameter adjustment direction is incorrect or excessive, it is impossible to revert to previously validated parameter combinations, leading to oscillations or continuous deterioration of the equilibrium effect.

[0009] Sixth, existing equilibrium techniques are based on deterministic models and rely heavily on accurate SOC estimates for decision-making. When there are biases in the SOC estimates, the equilibrium action may be counterproductive. Existing schemes lack the quantification and utilization of SOC estimation uncertainty, and therefore cannot make robust equilibrium decisions under uncertain conditions. Summary of the Invention

[0010] To overcome the shortcomings of existing technologies, this invention provides a modular-level distributed active balancing method and system. By modeling the balancing potential field, the discrete "acquisition-computation-allocation" process is replaced with a self-organized energy flow driven by a continuous potential field gradient, shortening the balancing response time and naturally adapting to the topology expansion of systems of any scale. Online assessment of balancing channel health (BHI) allows the balancing path selection to perceive the degree of channel degradation, avoiding the selection of degraded channels and extending the overall lifespan of the balancing system. A condition coupling prediction mechanism identifies low-load balancing windows in advance, initiating balancing at the optimal time, eliminating conflicts between balancing and main power operations, and improving balancing efficiency. A virtual balancing token mechanism achieves fully decentralized balancing permission management, eliminating the risk of single-point failure of the central coordinator, with token transmission latency controllable to the millisecond level. A strategy evolution and versioning mechanism enables bidirectional adjustment of balancing parameters and second-level rollback of historical optimal strategies, avoiding effect oscillations caused by parameter adjustments. By introducing SOC estimation uncertainty as a modulation factor for the potential field strength, the balancing intensity is automatically reduced when SOC estimation is uncertain, achieving robust balancing decisions under uncertain conditions.

[0011] To achieve the above objectives, a module-level distributed active load balancing method and system are designed, characterized in that the system includes:

[0012] SOC Estimation and Uncertainty Quantification Module: This module is deployed in each module controller and is responsible for online SOC estimation of each cell within the module, and synchronously quantifying the uncertainty of the SOC estimation. The input of this module is the sampled values ​​of each cell voltage, module current and temperature, and the output is the estimated SOC value of each cell, the SOC uncertainty, the average SOC of the module and the SOC deviation vector between modules. The uncertainty information is reported along with the SOC deviation and used for the modulation of the potential field strength.

[0013] Equilibrium Potential Field Construction Module: This module maps the SOC deviation of all modules in the system to a continuous equilibrium potential field Φ(x), where x is the position coordinate of the module in the equilibrium topology; this module operates in a distributed manner, that is, each module controller calculates the local potential field gradient based on the local SOC deviation and the potential energy information of neighboring modules, and independently determines the direction and intensity of energy inflow or outflow.

[0014] Balanced Channel Health Assessment Module: This module is responsible for online assessment of the health status of each balanced channel. This module is updated regularly, and during the update, the newly measured channel efficiency is fitted with the historical efficiency sequence to determine the trend.

[0015] Operating Condition Coupling Prediction Module: This module is responsible for pattern recognition based on historical power curves to predict the charging and discharging operating conditions of the system within a future time window. The operating condition prediction adopts a power pattern matching method based on sliding window statistics. The output of this module includes the expected start time, expected duration and confidence level of the next equalization window.

[0016] Virtual Balanced Token Management Module: This module implements decentralized balanced permission management based on virtual tokens;

[0017] The three-level equilibrium scheduling module is responsible for determining the equilibrium level that needs to be activated based on the potential field gradient strength and the equilibrium window status.

[0018] Potential field gradient equalization execution module: This module is responsible for controlling the bidirectional equalization converter to perform energy transfer based on the local potential field gradient and token holding state;

[0019] Anomaly detection and potential field isolation module: This module monitors the operating status of each module and equalization channel in real time, and isolates the abnormal module from the potential field when an anomaly is detected.

[0020] Strategy Evolution and Versioning Module: After each round of equilibrium is completed, this module calculates the equilibrium efficiency index ηbal and stores all current equilibrium strategy parameters and equilibrium efficiency index ηbal as a strategy version.

[0021] Balance Result Recording and Reporting Module: This module records complete information for each round of balance, including potential field distribution snapshot, token passing trajectory, BHI changes, working condition prediction accuracy, strategy version number, and balance efficiency indicators. The recorded information is reported to the operation and maintenance platform to generate balance trend analysis and system health assessment reports.

[0022] The distributed active load balancing method is as follows:

[0023] S1, Multi-channel data acquisition and preprocessing: Each module controller synchronously acquires the terminal voltage of all cells in the module, the total current of the module and the temperature of each temperature measuring point through a high-precision ADC at a fixed period. The acquired raw data is filtered by median to remove pulse noise and then filtered by moving average to smooth high-frequency fluctuations.

[0024] S2, SOC estimation and uncertainty are quantified simultaneously: Each module controller estimates the operating SOC of each cell in the module based on the filtered voltage and current data using an extended Kalman filter;

[0025] S3, Module Average SOC and Uncertainty Convergence: Each module controller calculates the module average SOC and average uncertainty;

[0026] S4, Equilibrium Potential Field Construction: The equilibrium potential field construction module maps the SOC deviation of each module into a potential field;

[0027] S5, Balanced Channel Health Assessment: For each balanced channel (i,j), the balanced channel health assessment calculates its health index;

[0028] S6, Operating Condition Coupling Prediction: The operating condition coupling prediction module identifies the equalization window based on historical power curves;

[0029] S7, Virtual Balance Token Transmission and Global State Broadcast: Virtual balance tokens are transmitted between modules in logical ring order;

[0030] S8, Potential field gradient-driven equilibrium execution: Modules holding tokens perform equilibrium based on the local potential field gradient;

[0031] S9, Real-time evaluation of balance effect: After each round of full-loop transmission of the token, the system calculates the global balance effect;

[0032] S10, Equilibrium Termination Decision;

[0033] S11, Anomaly Detection and Potential Field Isolation: The anomaly detection module calculates the comprehensive anomaly score of each module in real time;

[0034] S12, Strategy Evolution and Version Management: After each round of balancing is completed, the strategy evolution and versioning module performs the following steps: recording the current strategy version and its corresponding balancing efficiency, storing the current strategy version and its corresponding balancing efficiency in a circular buffer, selecting the highest version as the baseline, generating a new parameter combination in the parameter space neighborhood of the highest version, and rollback determination.

[0035] S13, BHI is updated regularly: After each round of balancing is completed, the health assessment of the balancing channel updates the BHI for each used balancing channel;

[0036] S14, Operating condition prediction model update: Every day at dawn, the operating condition prediction module updates its historical power database;

[0037] S15, Equilibrium Result Recording and Reporting: After each round of equilibrium, the system records the potential field distribution snapshot, token transmission trajectory and holding time of each module, BHI change trend, operating condition prediction accuracy, strategy version number and rollback event, SOC standard deviation before and after equilibrium, and equilibrium efficiency indicators. The total balanced energy consumption information is then reported to the operation and maintenance platform.

[0038] In the equilibrium potential field construction module, the construction rule for the potential field is: modules with a SOC higher than the global average are given positive potential energy, and modules with a SOC lower than the global average are given negative potential energy.

[0039] In step S2, the state equation of the extended Kalman filter is based on a first-order RC equivalent circuit model: ;in, For state vectors, This is the estimated SOC value for the k-th sampling period. is the RC polarization voltage; A is the state transition matrix, where B is the input matrix. , ; This refers to the charging and discharging current. This is the process noise vector; The sampling period; Polarization resistor; Polarizing capacitor; Coulomb efficiency; This refers to the nominal capacity.

[0040] In step S3, the formulas for calculating the module's average SOC and average uncertainty are as follows: ; ;in, Let be the average SOC of the i-th module; The number of battery cells in the module, a positive integer, typically 16; This represents the module's average SOC uncertainty.

[0041] In step S4, the potential energy is defined as: ;in, Let m be the potential energy value of the m-th module; This is the potential field gain coefficient, with a default value of 10.0, which is adjusted by the strategy evolution module. Average SOC of the module; The global average SOC; The uncertainty modulation function is defined as: ;in, The module's average SOC uncertainty; For reference uncertainty, the default value is 0.02; when hour When it is close to 1, hour Close to 0.

[0042] In step S5, the health index of the balanced channel is calculated: ;in, The balance health of channel (i,j) is represented by a value between 0 and 1. The efficiency decay weight is set to 0.5 by default. = / , which is the ratio of the current efficiency to the initial efficiency; This is the contact resistance weight, defaulting to 0.3; , is the normalized complement of the contact resistance degradation; The current contact resistance is... The initial contact resistance, Maximum permissible contact resistance; The cumulative energy weight is 0.2 by default. , is the complement of the ratio of cumulative equilibrium energy to channel design lifetime energy; Accumulated energy transfer for the channel, Total energy transferred over the channel's design lifetime.

[0043] In step S6, the operating condition coupling prediction module identifies the equalization window based on the following historical power curves:

[0044] (1) Extracting the past The system total power curve at the same time on the same day ;

[0045] (2) Calculate the mean and standard deviation of power for the predicted period: ; in, To predict the historical average power for a given period; The historical power standard deviation; For predicted future time points; The default number of days for historical data lookup is 7 days.

[0046] (3) Criteria for determining the equilibrium window: ;in, This is a Boolean flag for the equalization window. This is the system's rated power; This is the window power ratio threshold, with a default value of 0.2. This is the power stability ratio threshold, with a default value of 0.1.

[0047] (4) Calculate the window confidence level: ;in, The confidence level for the window, ranging from 0 to 1.

[0048] In step S7, the token data structure includes the following fields:

[0049] (1) token_gen: Token algebra number, a positive integer;

[0050] (2) holder_id: The current holder module number, a positive integer;

[0051] (3) soc_global_snapshot: Global SOC mean snapshot, dimensionless;

[0052] (4) field_strength_max: The maximum value of the global field gradient, dimensionless;

[0053] (5) window_active: Whether the current window is in a balanced window period, Boolean;

[0054] (6) strategy_version: Current strategy version number, a positive integer;

[0055] The token holding module executes the following logic during the holding period:

[0056] (1) Update the global SOC snapshot in the token;

[0057] (2) Determine whether this module needs to perform load balancing;

[0058] (3) If balancing is required and the current balancing window is in progress, perform the balancing operation in step eight;

[0059] (4) If balancing is not required or balancing has been completed, pass the token to the next module in the logical ring;

[0060] Token loss recovery mechanism: Each module maintains a token timeout timer. If no token message is detected on the CAN bus within the timeout period, the active module with the smallest number regenerates a new token and restarts the loop from its own position.

[0061] In step S8, the magnitude of the equilibrium current is proportional to the magnitude of the potential field gradient: ;in, The equalization current setting, in amperes (A); The upper limit of the rated current of the equalization converter is set to 5A; This is the potential-to-current conversion coefficient, with a default value of 2.0A, which is adjusted by the strategy evolution module. This represents the magnitude of the local potential field gradient. The BHI conduction coefficient of the channel;

[0062] The equilibrium direction is determined by the sign of the potential field gradient: When the voltage is greater than 0, energy flows from module i to module j. The converter duty cycle is calculated using the following formula: ;in, This represents the PWM duty cycle, ranging from 0.1 to 0.9. The target output voltage; Input voltage; For converter conversion efficiency;

[0063] The maximum equilibrium duration during a single token holding period is Single equilibrium energy The calculation is as follows = ;in, This represents the average voltage during the equalization period.

[0064] In step S9, the metrics for the global equilibrium effect are calculated: ;in, The standard deviation of the system's SOC is given; the equilibrium efficiency index for this round is also calculated. ;in, To balance efficiency indicators; This represents the change in the standard deviation of SOC before and after this round of equilibrium. This represents the total balanced energy consumption for this round.

[0065] In step S10, the equilibrium termination determination logic is to terminate the current equilibrium cycle when any of the following conditions are met:

[0066] (1) Potential field convergence: The maximum value of the gradient of the potential field is lower than the convergence threshold;

[0067] (2) End of Balanced Window: The balanced window period marked by the working condition prediction module ends;

[0068] (3) Equilibrium timeout: The duration of a single equilibrium cycle exceeds;

[0069] (4) Abnormal interruption: The abnormal detection module triggers potential field isolation.

[0070] In step S11, the anomaly detection module calculates the comprehensive anomaly score for each module in real time: ;in, For the anomaly score of the i-th module; Weighted by the rate of change of internal resistance, default value is 0.4; The time rate of change of the module's internal resistance; Used as a reference rate of change in internal resistance; The weight for temperature deviation is 0.35 by default. For module temperature; This is the temperature warning value; This is the upper limit of the temperature range; This is the communication quality weight, with a default value of 0.25. For communication packet loss rate;

[0071] when Exceeding the isolation threshold At that time, perform potential field isolation: isolate the potential energy of the module. Forced zeroing is equivalent to "removing" it from the potential field. The potential field is automatically redistributed among the remaining modules without explicit path replanning. If the module is holding a token, the token is forcibly passed to the next active module.

[0072] In step S12, the strategy evolution and versioning module performs the following operations:

[0073] (1) Record the current strategy version and its corresponding equilibrium efficiency ;

[0074] (2) Stored in a circular buffer, buffer capacity ;

[0075] (3) From the most recent Choose from the following versions (5 by default): The highest version As a benchmark;

[0076] (4) In Generate new parameter combinations within the parameter space neighborhood. : ;in, The value of parameter p in the new version; The value of parameter p in the baseline version; The exploration range for parameter p, with a default value of 0.1; for[ Uniformly distributed random numbers over an interval;

[0077] (5) Rollback determination: If continuous After the next update All below of Then it will automatically roll back to The parameter snapshot is completed within one control cycle.

[0078] In step S13, the BHI periodic update process is as follows:

[0079] (1) Measure the actual transfer efficiency of this round. = / ;

[0080] (2) The efficiency history sequence is appended to channel (i,j);

[0081] (3) Fit a linear trend to the historical efficiency sequence and update ;

[0082] (4) Measure the channel contact resistance and update. ;

[0083] (5) Cumulative energy += ,renew ;

[0084] (6) Recalculate and .

[0085] Compared with existing technologies, this invention provides a modular distributed active balancing method and system. By modeling the balancing potential field, it replaces the discrete "acquisition-computation-allocation" process with a self-organized energy flow driven by a continuous potential field gradient, shortening the balancing response time and naturally adapting to the topology expansion of systems of any scale. Through online assessment of balancing channel health (BHI), it enables balancing path selection to perceive the degree of channel degradation, avoiding the selection of degraded channels and extending the overall lifespan of the balancing system. Through a condition coupling prediction mechanism, it identifies low-load balancing windows in advance and initiates balancing at the optimal time, eliminating conflicts between balancing and main power operations and improving balancing efficiency. Through a virtual balancing token mechanism, it achieves fully decentralized balancing permission management, eliminating the risk of single-point failure of the central coordinator, with token transmission latency controllable to the millisecond level. Through a strategy evolution and versioning mechanism, it enables bidirectional adjustment of balancing parameters and second-level rollback of historical optimal strategies, avoiding effect oscillations caused by parameter adjustments. By introducing SOC estimation uncertainty as a modulation factor for the potential field strength, it automatically reduces the balancing intensity when SOC estimation is uncertain, achieving robust balancing decisions under uncertain conditions. Attached Figure Description

[0086] Figure 1 This is a diagram of the overall system architecture of the present invention.

[0087] Figure 2 This is the control flowchart of the present invention.

[0088] Figure 3 This is a schematic diagram of the gradient of the equilibrium potential field.

[0089] Figure 4 A sequence diagram for virtual token transfer and loss recovery.

[0090] Figure 5 This is a sequence of contour lines showing the convergence of the potential field in an embodiment of the present invention. Detailed Implementation

[0091] The present invention will now be further described with reference to the accompanying drawings.

[0092] like Figure 1 As shown, this invention presents the overall system architecture.

[0093] Figure 1 The left half (potential field mechanism link): draws the data flow link from top to bottom. The top rectangle "SOC estimation and uncertainty quantification module" outputs two arrows: one labeled "SOC deviation" pointing to the "equilibrium potential field construction module", and the other labeled "uncertainty σ" also pointing to the "equilibrium potential field construction module" (labeled "modulation potential field strength"). The "equilibrium potential field construction module" is filled with a gradient color (red = high potential energy → blue = low potential energy), and its left side receives the "conduction coefficient β" from the "BHI evaluation module". The "equilibrium potential field construction module" outputs "potential field gradient ∇Φ" pointing to the "potential field gradient equalization execution module". The "potential field gradient equalization execution module" is connected to the "bidirectional equalization converter" and "equalization bus" of each module.

[0094] Figure 1 The right half (token coordination link): Draw a logical ring consisting of 8 module nodes, with the nodes connected end-to-end by unidirectional arrows to form a loop, labeled "token transfer" next to the arrows. A token icon (circular coin shape) is placed in the middle of the loop, indicating the current holder. The token data structure fields are labeled around the outside of the loop.

[0095] Bottom horizontal modules: horizontal rectangle "Operating Condition Coupling Prediction Module" (input "Historical Power Curve" on the left, output "Window Flag" to the token ring on the right); "Anomaly Detection and Potential Field Isolation Module" (red dashed box, connected to both the potential field construction module and the token ring); "Policy Evolution and Versioning Module" (bottom, receives ηbal input, outputs "Parameter Update / Rollback" to the potential field construction and token management module).

[0096] At the very bottom: The "Balanced Result Recording and Reporting Module" aggregates all data streams.

[0097] Color scheme: Potential field related modules use a blue-red gradient; token related modules use a gold color scheme; BHI module uses green; anomaly detection uses a red border; strategy evolution uses purple.

[0098] I. SOC Estimation and Uncertainty Measurement Module.

[0099] This module, deployed in each module controller, is responsible for online SOC estimation of each cell within the module and simultaneously quantifying the uncertainty of the SOC estimation. The SOC estimation employs an extended Kalman filter (EKF), using the battery equivalent circuit model as the state equation and the terminal voltage as the observed quantity. While outputting the SOC estimate, the diagonal elements of the EKF's error covariance matrix represent the variance of the SOC estimate; its square root is taken as the uncertainty σ of the SOC estimation. SOC,i The module takes as input the sampled values ​​of cell voltage, module current, and temperature, and outputs the estimated state of charge (SOC) of each cell, SOC uncertainty, average SOC of the module, and SOC deviation vector between modules. The uncertainty information is reported along with the SOC deviation and is used for modulation of the potential field strength.

[0100] II. Equilibrium Potential Field Construction Module.

[0101] This module is one of the core innovative modules of this invention. It maps the SOC deviation of all modules in the system to a continuous equilibrium potential field Φ(x), where x is the position coordinate of the module in the equilibrium topology. The construction rule of the potential field is: modules with SOC higher than the global average are assigned positive potential energy ("potential energy highlands"), and modules with SOC lower than the global average are assigned negative potential energy ("potential energy lowlands"). The potential field gradient ∇Φ represents the direction and intensity of the energy flow. The potential field intensity is modulated by the SOC estimation uncertainty: when the SOC estimation uncertainty of a module is large, the potential field intensity corresponding to that module is attenuated, avoiding radical equilibrium decisions based on uncertain information. The equilibrium energy flow flows naturally along the potential field gradient, without the need for a central controller to calculate paths and issue commands one by one. This module operates in a distributed manner: each module controller calculates its local potential field gradient based on its local SOC deviation and the potential energy information of its neighboring modules, independently determining the direction and intensity of energy inflow or outflow.

[0102] III. Balanced Channel Health (BHI) Assessment Module.

[0103] This module is responsible for online evaluation of the health status of each equalization channel. A BHI index is defined for each equalization channel, comprehensively considering the channel's historical transfer efficiency decay trend, contact resistance variation trend, and cumulative equalization energy. The BHI value ranges from 0 to 1, where 1 indicates a fully healthy channel and 0 indicates a failed channel. The BHI evaluation result is used to determine the conduction coefficient of the corresponding edge in the modulation potential field: channels with lower BHI have reduced conduction coefficients at their corresponding potential field edges, causing energy flow to automatically bypass the degraded channel. The BHI is updated periodically (after each equalization cycle), and during the update, the newly measured channel efficiency is trend-fitted with the historical efficiency sequence.

[0104] IV. Operating Condition Coupling Prediction Module.

[0105] This module is responsible for pattern recognition based on historical power curves, predicting the system's charging and discharging conditions within a future time window, and marking the "balancing window period" (the interval where the expected low load duration exceeds a set threshold). The condition prediction uses a power pattern matching method based on sliding window statistics: extracting power curves from the same time period over the past N days (default 7 days), calculating their mean and standard deviation, and marking the predicted future time period as a balancing window period when the predicted power mean is lower than a certain percentage (default 20%) of the rated power and the standard deviation is lower than a set value. The module outputs include: the expected start time, expected duration, and confidence level of the next balancing window. The balancing trigger logic prioritizes initiating balancing within the balancing window period; outside the window period, only emergency balancing is handled.

[0106] V. Virtual Balance Token Management Module.

[0107] This module is another core innovation of this invention. It implements decentralized load balancing permission management based on virtual tokens. A virtual load balancing token is set up in the system, and the token is passed between all module controllers according to a predefined logical ring order. The token data structure carries the following information: a snapshot of the global SOC mean, a global field strength summary, the current holder's ID, and the token generation number. Only the module holding the token has the right to initiate load balancing operations (actively pushing energy as a supplier or actively pulling energy as a demander). The maximum duration of token holding is Ttoken (default 500ms). After the timeout, the token must be passed to the next module in the logical ring. When the module holding the token does not need load balancing, it immediately passes the token to the next module (fast pass-through mode). Token transmission uses CAN bus broadcasting, and the transmission delay can be controlled within 5ms. Token loss detection mechanism: If any module does not receive a token within a predetermined timeout period (default 2 seconds), the active module with the smallest ID regenerates the token and restarts the ring.

[0108] VI. Three-level balanced scheduling module.

[0109] This module is responsible for determining the current balancing level to be activated based on the potential field gradient strength and the balancing window status. Unlike traditional three-level scheduling, this module's scheduling decision is directly driven by the magnitude of the potential field gradient: when the local potential field gradient exceeds the cell-level threshold but not the module-level threshold, cell-level balancing is activated; when the potential field gradient exceeds the module-level threshold, module-level balancing is activated; when the inter-cluster potential field gradient exceeds the cluster-level threshold, inter-cluster balancing is achieved through PCS power adjustment. The balancing scheduling is also constrained by the operating condition prediction module: non-emergency balancing is only performed within the balancing window period.

[0110] VII. Potential Field Gradient Equalization Execution Module.

[0111] This module is responsible for controlling the bidirectional equalizer to perform energy transfer based on the local potential field gradient and token holding status. When the module holds a token and the local potential field gradient exceeds the execution threshold, the converter executes energy flow according to the direction of the potential field gradient: when the gradient points outward (the module's potential energy is higher than its neighbors), the converter operates in boost mode to output energy; when the gradient points inward (the module's potential energy is lower than its neighbors), the converter operates in buck mode to absorb energy. The equalization current is proportional to the magnitude of the potential field gradient and is limited by the BHI conduction coefficient.

[0112] 8. Anomaly Detection and Potential Field Isolation Module.

[0113] This module monitors the operating status of each module and the equalization channel in real time. When an anomaly is detected, it isolates the abnormal module from the potential field. Unlike traditional bypass, potential field isolation directly sets the potential field contribution of the abnormal module to zero, allowing the potential field to be automatically redistributed among the remaining modules without the need for explicit path replanning. Anomaly detection integrates three dimensions: internal resistance change rate, temperature deviation, and communication quality.

[0114] IX. Strategy Evolution and Versioning Module.

[0115] This module is the third core innovation of this invention. After each round of balancing, this module calculates the balancing efficiency index ηbal and associates all current balancing strategy parameters (potential gain, BHI weight, token holding duration, condition prediction window length, etc.) with the balancing efficiency index ηbal, storing them as a "strategy version". The strategy version is managed using a circular buffer, storing the K most recent versions (default K=20). When updating parameters, the module uses the version with the highest ηbal among the M most recent versions (default M=5) as a benchmark and searches for new parameter combinations in its parameter space neighborhood. If, after several (default) updates, the result is still lower than the historical best version, it automatically rolls back to the historical best strategy version. The strategy version contains a complete parameter snapshot, and the rollback operation is completed within one control cycle.

[0116] 10. Balanced Result Recording and Reporting Module.

[0117] This module records complete information for each round of balancing, including a snapshot of the potential field distribution, token passing trajectory, BHI changes, accuracy of operational condition prediction, strategy version number, and balancing efficiency metrics. The recorded information is reported to the operations and maintenance platform to generate balancing trend analysis and system health assessment reports.

[0118] The connection relationship of the above ten modules is as follows: the SOC estimation and uncertainty quantification module outputs the SOC deviation and uncertainty to the equilibrium potential field construction module; the equilibrium potential field construction module constructs the potential field based on the SOC deviation and receives the channel health information from the BHI evaluation module to modulate the conduction coefficient of the potential field edge; the operating condition coupling prediction module outputs the equilibrium window information to the three-level equilibrium scheduling module; the virtual equilibrium token management module transmits tokens and global state summaries between modules; the three-level equilibrium scheduling module determines the equilibrium level by combining the potential field gradient and window state; the potential field gradient equilibrium execution module executes energy transfer based on the token holding state and the local gradient driving the converter; the anomaly detection and potential field isolation module monitors in real time and isolates the module from the potential field when anomalies occur; the strategy evolution and versioning module evaluates the effect after each round of equilibrium and updates or rolls back the strategy parameters; the equilibrium result recording and reporting module gathers all information and reports it.

[0119] like Figure 2 As shown, the control flow of the present invention is as follows:

[0120] Step 1: Multi-channel data acquisition and preprocessing: Each module controller synchronously acquires the terminal voltage of all cells in the module, the total current of the module, and the temperature of each measuring point at a fixed period (default 1 second) through a high-precision ADC. The acquired raw data is filtered by median to remove pulse noise, and then filtered by moving average to smooth high-frequency fluctuations.

[0121] Step 2, SOC estimation and uncertainty quantification simultaneously: Each module controller performs SOC estimation for each cell in the module based on the filtered voltage and current data using an extended Kalman filter (EKF).

[0122] The state equations of the Extended Kalman Filter (EKF) are based on a first-order RC equivalent circuit model: ;in, For state vectors, Let SOC be the estimated value for the k-th sampling period (dimensionless, from 0 to 1). RC polarization voltage (in V); A is the state transition matrix, where B is the input matrix. , ; This is the charging / discharging current (unit: A, positive for charging). This is the process noise vector; The sampling period (in seconds); Polarization resistance (unit: Ω); Polarization capacitance (unit: F); Coulomb efficiency (dimensionless); Nominal capacity (unit: As).

[0123] The observation equation is: ;in, Terminal voltage observation (unit: V); This is a lookup table function for SOC to open-circuit voltage; Internal resistance in ohms (unit: Ω); To observe noise.

[0124] The (1,1) element of the error covariance matrix of the Extended Kalman Filter (EKF) is the variance of the SOC estimate. Taking its square root yields the SOC estimate uncertainty. ;in, Let be the SOC estimation uncertainty of the i-th cell. It is dimensionless, and the larger the value, the more uncertain the estimation.

[0125] Step 3, Module Average SOC and Uncertainty Convergence: Each module controller calculates the module average SOC and average uncertainty. ; ;in, Let be the average SOC of the i-th module, which is dimensionless; The number of battery cells in the module, a positive integer, typically 16; The module average SOC uncertainty is dimensionless.

[0126] Each module will and It is transmitted along with the token information within the logical ring. The global average SOC is calculated from the global snapshot carried by the token: ;in, The total number of modules within the cluster is a positive integer, typically 16.

[0127] Step four, as Figure 3 As shown, the equilibrium potential field is constructed by mapping the SOC deviation of each module to a potential field. The potential energy of the m-th module is defined as follows: ;in, Let be the potential energy value of the m-th module, which is dimensionless; This is the potential field gain coefficient, dimensionless, with a default value of 10.0, which is adjusted by the strategy evolution module; Average SOC of the module; The global average SOC; The uncertainty modulation function is defined as: ;in, The module's average SOC uncertainty; For reference uncertainty, dimensionless, default value is 0.02; when hour Approaching 1 (potential field strength does not decay), when hour Approaching 0 (the potential field strength decreases significantly, avoiding equilibrium under high uncertainty).

[0128] The potential gradient between adjacent modules i and j is defined as: ;in, Let be the potential gradient from module i to module j, with a positive value indicating that energy should flow from i to j; is the BHI conduction coefficient of channel (i,j), which is dimensionless and takes a value from 0 to 1, provided by the BHI evaluation module.

[0129] Step 5, Balanced Channel Health (BHI) Assessment: For each balanced channel (i,j), the BHI assessment module calculates its health index: ;in, The equilibrium health of channel (i,j) is dimensionless and ranges from 0 to 1; The efficiency decay weight is dimensionless and defaults to 0.5. = / , which is the ratio of the current efficiency to the initial efficiency, and is dimensionless; This is the contact resistance weight, dimensionless, default value 0.3; , is the normalized complement of the contact resistance degradation, and is dimensionless; This represents the current contact resistance (in mΩ). This is the initial contact resistance (in mΩ). Maximum permissible contact resistance (in mΩ); This is the cumulative energy weight, dimensionless, with a default value of 0.2. , is the complement of the ratio of cumulative equilibrium energy to channel design lifetime energy, and is dimensionless; Accumulated energy transferred through the channel (unit: Wh). Total energy transferred over the channel's design lifetime (in Wh).

[0130] BHI is also used to calculate the conduction coefficient of the potential field edge: ;in, This is the BHI sensitivity index, dimensionless, with a default value of 2.0; The larger the value, the stronger the penalty for the conduction coefficient when the BHI decreases.

[0131] Step 6, Operating Condition Coupling Prediction: The operating condition coupling prediction module identifies the equalization window based on historical power curves.

[0132] (1) Extracting the past Total system power curve for the same period of the day (default 7 days) .

[0133] (2) Calculate the mean and standard deviation of power for the predicted period: ; in, The historical average power (in kW) for the predicted period; Historical power standard deviation (in kW); For predicted future time points; The default number of days for historical data back is 7 days.

[0134] (3) Criteria for determining the equilibrium window: ;in, This is a Boolean flag for the equalization window. The system's rated power (in kW); This is the window power ratio threshold, dimensionless, with a default value of 0.2. This is the power stability ratio threshold, dimensionless, with a default value of 0.1.

[0135] (4) Calculate the window confidence level: ;in, represents the window confidence level, which is dimensionless and ranges from 0 to 1.

[0136] Step seven, as Figure 4 As shown, virtual balancer token passing and global state broadcasting: Virtual balancer tokens are passed between modules in logical ring order. The token data structure contains the following fields:

[0137] (7) token_gen: Token algebra number, a positive integer;

[0138] (8) holder_id: The current holder module number, a positive integer;

[0139] (9) soc_global_snapshot: Global SOC mean snapshot, dimensionless;

[0140] (10) field_strength_max: The maximum value of the global field gradient, dimensionless;

[0141] (11) window_active: Whether the current window is in a balanced window period, Boolean;

[0142] (12) strategy_version: Current strategy version number, a positive integer.

[0143] The token holding module executes the following logic during the holding period:

[0144] (5) Update the global SOC snapshot in the token (replace the old value in the corresponding position in the token with the latest SOC of this module).

[0145] (6) Determine whether this module needs to perform balancing (whether the local potential gradient exceeds the execution threshold).

[0146] (7) If balancing is required and the current period is within the balancing window (or for emergency balancing), perform the balancing operation in step eight.

[0147] (8) If balancing is not required or balancing has been completed, pass the token to the next module in the logical ring.

[0148] Token loss recovery mechanism: Each module maintains a token timeout timer. (Default 2 seconds). If no token message is detected on the CAN bus within the timeout period, the active module with the smallest number regenerates a new token (token_gen increments by 1) and restarts the loop from its own position.

[0149] Step 8, Potential Field Gradient-Driven Equalization Execution: The module holding the token performs equalization based on the local potential field gradient. The magnitude of the equalization current is proportional to the magnitude of the potential field gradient. ;in, The equalization current setting, in amperes (A); This is the upper limit of the rated current of the equalization converter, in amperes (A), with a value of 5A. This is the potential-to-current conversion coefficient, in amperes (A), with a default value of 2.0A, which is adjusted by the strategy evolution module. The magnitude of the local potential field gradient is dimensionless. denoted as the channel BHI conduction coefficient, which is dimensionless.

[0150] The equilibrium direction is determined by the sign of the potential field gradient: When the voltage is greater than 0, energy flows from module i to module j. The converter duty cycle is calculated using the following formula: ;in, This represents the PWM duty cycle, which is dimensionless and ranges from 0.1 to 0.9. Target output voltage (in V); Input voltage (in V); The converter efficiency is dimensionless.

[0151] The maximum equilibrium duration during a single token holding period is (Default 500ms). Single-cycle energy equalization. The calculation is as follows = ;in, The average voltage (in V) during the equalization period.

[0152] Step 9, Real-time Evaluation of Equilibrium Effect: After each round of full-loop token transmission (all modules take turns holding the token once), the system calculates the global equilibrium effect index: ;in, Let SOC be the standard deviation of the system, which is dimensionless.

[0153] Simultaneously calculate the equilibrium efficiency index for this round: ;in, Efficiency is a balanced efficiency indicator, expressed as % / Wh; This represents the change in the standard deviation of SOC before and after this round of equilibrium (in percentage points). This represents the total balanced energy consumption for this round (in Wh).

[0154] Step 10, Equilibrium Termination Determination: The current equilibrium cycle terminates when any of the following conditions are met:

[0155] (1) Potential field convergence: The maximum value of the potential field gradient is lower than the convergence threshold (default 0.5, corresponding to a SOC deviation of about 0.5%).

[0156] (2) End of Balanced Window: The balanced window period marked by the working condition prediction module ends.

[0157] (3) Equilibrium timeout: The duration of a single equilibrium cycle exceeds (default 60 minutes).

[0158] (4) Abnormal interruption: The abnormal detection module triggers potential field isolation.

[0159] Step 11, Anomaly Detection and Potential Field Isolation: The anomaly detection module calculates the comprehensive anomaly score for each module in real time. ;in, The anomaly score for the i-th module is dimensionless. This is the weight for the rate of change of internal resistance, dimensionless, default value 0.4; The time-varying rate of change of the module's internal resistance (unit: mΩ / h); Reference internal resistance change rate (unit: mΩ / h, default: 0.5mΩ / h); The weight for temperature deviation is dimensionless and defaults to 0.35. Module temperature (in °C); Temperature warning value (unit: °C, default 45°C); This is the upper limit of temperature (unit: °C, default 55°C); This is a communication quality weight, dimensionless, with a default value of 0.25. The packet loss rate is dimensionless.

[0160] when Exceeding the isolation threshold (Default 0.7) When potential field isolation is performed, the potential energy of the module is isolated. Forced zeroing is equivalent to "removing" it from the potential field. The potential field is automatically redistributed among the remaining modules without explicit path replanning. If the module is holding a token, the token is forcibly passed to the next active module.

[0161] Step 12, Strategy Evolution and Version Management: After each round of balancing, the strategy evolution and versioning module performs the following operations:

[0162] (6) Record the current strategy version and its corresponding equilibrium efficiency .

[0163] (7) Stored in a circular buffer, buffer capacity (Default 20).

[0164] (8) From the most recent Choose from the following versions (5 by default): The highest version As a benchmark.

[0165] (9) In Generate new parameter combinations within the parameter space neighborhood. : ;in, The value of parameter p in the new version; The value of parameter p in the baseline version; The exploration range of parameter p is dimensionless and defaults to 0.1. for[ A uniform random number over an interval.

[0166] (10) Rollback determination: If continuous After the third (default 3) update All below of Then it will automatically roll back to The parameter snapshot is completed within one control cycle (1 second).

[0167] Step 13, BHI Periodic Updates: After each round of equalization is completed, the BHI evaluation module updates the BHI for each used equalization channel.

[0168] (7) Measure the actual transfer efficiency of this round. = / .

[0169] (8) The efficiency history sequence is appended to channel (i,j).

[0170] (9) Fit a linear trend to the historical efficiency sequence and update .

[0171] (10) Measure the channel contact resistance (calculated by voltage drop during no-load / light-load switching of the equalization converter), and update. .

[0172] (11) Cumulative energy += ,renew .

[0173] (12) Recalculate and .

[0174] Step Fourteen, Operating Condition Prediction Model Update: Every morning (or after each complete charge-discharge cycle), the operating condition prediction module updates its historical power database: adding the power curve of the latest day to the sliding window and removing the data of the oldest day. If the prediction accuracy (the overlap rate between the prediction window and the actual low-load period) is lower than the set threshold (default 70%), the number of backtracking days is automatically increased. (Maximum 30 days) to improve statistical stability.

[0175] Step 15, Equilibrium Result Recording and Reporting: After each round of equilibrium is completed, the system records the following information and reports it to the operation and maintenance platform: Potential field distribution snapshot (all modules) Value), token transmission trajectory and holding time of each module, BHI change trend, working condition prediction accuracy, strategy version number and rollback event, SOC standard deviation before and after balancing, and balancing efficiency indicators. Total balanced energy consumption.

[0176] Example: Potential field driven equilibrium in large-scale energy storage power stations.

[0177] Scenario: A 100MWh lithium iron phosphate energy storage power station has 20 battery clusters, each cluster has 16 modules, and each module has 16 280Ah cells connected in series. After 18 months of operation, the standard deviation of SOC among the modules is 4.2%.

[0178] Inputs: Cell voltage (3.20V to 3.35V), module current (0 to 150A), temperature (25°C to 38°C). EKF online operation, with SOC estimation uncertainty for each module ranging from 0.005 to 0.015.

[0179] Processing procedure:

[0180] Steps one through three: The 16 module controllers acquire and run EKF data at 1-second intervals, outputting SOC and uncertainty. The average SOC of each module within the cluster ranges from 49.7% to 56.5%, with a global average of 52.3%.

[0181] Step 4: Potential Field Construction. Taking Module 7 (SOC=56.5%, =0.008) as an example: =10.0·(0.565-0.523)·1 / (1+(0.008 / 0.02)²)=10.0·0.042·0.862=0.362. Taking Module 12 (SOC=49.7%, =0.012) as an example: =10.0·(0.497-0.523)·1 / (1+(0.012 / 0.02)²)=10.0·(-0.026)·0.735=-0.191. Potential gradient =(0.362-(-0.191))· =0.553·0.92=0.509.

[0182] Step 5: BHI Assessment. After 12 months of operation, channel (7,12) efficiency decreased from 94% to 91%, and contact resistance increased from 2.1mΩ to 2.8mΩ (maximum 5mΩ). BHI = 0.5·(91 / 94) + 0.3·(1-(2.8-2.1) / 5) + 0.2·(1-180 / 500) = 0.484 +

[0183] 0.258 + 0.128 = 0.870. =0.870²=0.757. (Using...) The gradient was then adjusted to 0.553·0.757=0.419.

[0184] Step Six: Identify the current time period (2:00 AM - 6:00 AM) as the balancing window. =15kW (3% of the rated 500kW) =8kW, window confidence level Conf=0.93.

[0185] Step 7: The token is passed among the 16 modules, with each round lasting 8 seconds. When module 7 holds the token, the potential gradient 0.419 is detected as greater than the threshold 0.1, and equalization is performed.

[0186] Step 8: Equalize the current =min(5,2.0·0.419·0.757)=min(5,0.634)=0.634A. Since the potential field drive is continuous rather than a single high-current operation, module 7 will continuously balance at approximately 0.634A for 500ms during each round of token transfer.

[0187] After running continuously for 4 hours (1800 rounds of token passing), the SOC of module 7 decreased from 56.5% to 53.1%, while the SOC of module 12 increased from 49.7% to 51.5%.

[0188] Step Nine: σSOC It decreased from 4.2% to 1.8%. η bal =2.4% / 1.2Wh=2.0% / Wh.

[0189] The equilibrium continued into the following night, σ SOC It further decreased to 0.6%.

[0190] Output: Within 72 hours, the system SOC standard deviation decreased from 4.2% to 0.6%, and the available capacity increased from 87% to 96%. Potential field-driven equilibrium eliminates the need for central path planning, with each module self-organizing to complete energy flow.

[0191] Technical effects: such as Figure 5 As shown, the potential field contour map illustrates the process by which the potential energy of each module gradually flattens out from "high and low fluctuations" during the equilibrium process.

Claims

1. A module-level distributed active load balancing method and system, characterized in that, The system includes: SOC Estimation and Uncertainty Quantification Module: This module is deployed in each module controller and is responsible for online SOC estimation of each cell within the module, and synchronously quantifying the uncertainty of the SOC estimation. The input of this module is the sampled values ​​of each cell voltage, module current and temperature, and the output is the estimated SOC value of each cell, the SOC uncertainty, the average SOC of the module and the SOC deviation vector between modules. The uncertainty information is reported along with the SOC deviation and used for the modulation of the potential field strength. Equilibrium Potential Field Construction Module: This module maps the SOC deviation of all modules in the system to a continuous equilibrium potential field Φ(x), where x is the position coordinate of the module in the equilibrium topology; this module operates in a distributed manner, that is, each module controller calculates the local potential field gradient based on the local SOC deviation and the potential energy information of neighboring modules, and independently determines the direction and intensity of energy inflow or outflow. Balanced Channel Health Assessment Module: This module is responsible for online assessment of the health status of each balanced channel. This module is updated regularly, and during the update, the newly measured channel efficiency is fitted with the historical efficiency sequence to determine the trend. Operating Condition Coupling Prediction Module: This module is responsible for pattern recognition based on historical power curves to predict the charging and discharging operating conditions of the system within a future time window. The operating condition prediction adopts a power pattern matching method based on sliding window statistics. The output of this module includes the expected start time, expected duration and confidence level of the next equalization window. Virtual Balanced Token Management Module: This module implements decentralized balanced permission management based on virtual tokens; The three-level equilibrium scheduling module is responsible for determining the equilibrium level that needs to be activated based on the potential field gradient strength and the equilibrium window status. Potential field gradient equalization execution module: This module is responsible for controlling the bidirectional equalization converter to perform energy transfer based on the local potential field gradient and token holding state; Anomaly detection and potential field isolation module: This module monitors the operating status of each module and equalization channel in real time, and isolates the abnormal module from the potential field when an anomaly is detected. Strategy Evolution and Versioning Module: After each round of equilibrium is completed, this module calculates the equilibrium efficiency index ηbal and stores all current equilibrium strategy parameters and equilibrium efficiency index ηbal as a strategy version. Balance Result Recording and Reporting Module: This module records complete information for each round of balance, including potential field distribution snapshot, token passing trajectory, BHI changes, working condition prediction accuracy, strategy version number, and balance efficiency indicators. The recorded information is reported to the operation and maintenance platform to generate balance trend analysis and system health assessment reports. The distributed active load balancing method is as follows: S1, Multi-channel data acquisition and preprocessing: Each module controller synchronously acquires the terminal voltage of all cells in the module, the total current of the module and the temperature of each temperature measuring point through a high-precision ADC at a fixed period. The acquired raw data is filtered by median to remove pulse noise and then filtered by moving average to smooth high-frequency fluctuations. S2, SOC estimation and uncertainty are quantified simultaneously: Each module controller estimates the operating SOC of each cell in the module based on the filtered voltage and current data using an extended Kalman filter; S3, Module Average SOC and Uncertainty Convergence: Each module controller calculates the module average SOC and average uncertainty; S4, Equilibrium Potential Field Construction: The equilibrium potential field construction module maps the SOC deviation of each module into a potential field; S5, Balanced Channel Health Assessment: For each balanced channel (i,j), the balanced channel health assessment calculates its health index; S6, Operating Condition Coupling Prediction: The operating condition coupling prediction module identifies the equalization window based on historical power curves; S7, Virtual Balance Token Transmission and Global State Broadcast: Virtual balance tokens are transmitted between modules in logical ring order; S8, Potential field gradient-driven equilibrium execution: Modules holding tokens perform equilibrium based on the local potential field gradient; S9, Real-time evaluation of balance effect: After each round of full-loop transmission of the token, the system calculates the global balance effect; S10, Equilibrium Termination Decision; The equilibrium termination decision logic is to terminate the current equilibrium cycle when any of the following conditions are met: (1) Potential field convergence: The maximum value of the gradient of the potential field is lower than the convergence threshold; (2) End of Balanced Window: The balanced window period marked by the working condition prediction module ends; (3) Equilibrium timeout: The duration of a single equilibrium cycle exceeds; (4) Abnormal interruption: The abnormal detection module triggers potential field isolation; S11, Anomaly Detection and Potential Field Isolation: The anomaly detection module calculates the comprehensive anomaly score of each module in real time; S12, Strategy Evolution and Version Management: After each round of balancing is completed, the strategy evolution and versioning module performs the following steps: recording the current strategy version and its corresponding balancing efficiency, storing the current strategy version and its corresponding balancing efficiency in a circular buffer, selecting the highest version as the baseline, generating a new parameter combination in the parameter space neighborhood of the highest version, and rollback determination. S13, BHI Regular Updates: After each round of balancing, the health assessment of the balancing channels updates the BHI for each used balancing channel; the BHI regular update process is as follows: (1) Measure the actual transfer efficiency of this round. = / ; (2) The efficiency history sequence is appended to channel (i,j); (3) Fit a linear trend to the historical efficiency sequence and update ; (4) Measure the channel contact resistance and update. ; (5) Cumulative energy += ,renew ; (6) Recalculate and ; S14, Operating condition prediction model update: Every day at dawn, the operating condition prediction module updates its historical power database; S15, Equilibrium Result Recording and Reporting: After each round of equilibrium, the system records the potential field distribution snapshot, token transmission trajectory and holding time of each module, BHI change trend, operating condition prediction accuracy, strategy version number and rollback event, SOC standard deviation before and after equilibrium, and equilibrium efficiency indicators. The total balanced energy consumption information is then reported to the operation and maintenance platform. In the equilibrium potential field construction module, the construction rule of the potential field is: modules with SOC higher than the global average are given positive potential energy, and modules with SOC lower than the global average are given negative potential energy. In step S3, the formulas for calculating the module's average SOC and average uncertainty are as follows: ; ;in, Let be the average SOC of the i-th module; The number of battery cells in the module, a positive integer, typically 16; This represents the module's average SOC uncertainty.

2. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S2, the state equation of the extended Kalman filter is based on a first-order RC equivalent circuit model: ;in, For state vectors, This is the SOC estimate for the k-th sampling period. is the RC polarization voltage; A is the state transition matrix, where B is the input matrix. , ; This refers to the charging and discharging current. This is the process noise vector; The sampling period; Polarization resistor; Polarizing capacitor; Coulomb efficiency; This refers to the nominal capacity.

3. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S4, the potential energy is defined as: ;in, Let m be the potential energy value of the m-th module; This is the potential field gain coefficient, with a default value of 10.0, which is adjusted by the strategy evolution module. Average SOC of the module; The global average SOC; The uncertainty modulation function is defined as: ;in, The module's average SOC uncertainty; For reference uncertainty, the default value is 0.02; when hour When it is close to 1, hour Close to 0.

4. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S5, the health index of the balanced channel is calculated: ;in, The balance health of channel (i,j) is represented by a value between 0 and 1. The efficiency decay weight is set to 0.5 by default. = / , which is the ratio of the current efficiency to the initial efficiency; This is the contact resistance weight, defaulting to 0.3; , is the normalized complement of the contact resistance degradation; The current contact resistance is... The initial contact resistance, Maximum permissible contact resistance; The cumulative energy weight is 0.2 by default. , is the complement of the ratio of cumulative equilibrium energy to channel design lifetime energy; Accumulated energy transfer for the channel, Total energy transferred over the channel's design lifetime.

5. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S6, the operating condition coupling prediction module identifies the equalization window based on the following historical power curves: (1) Extracting the past The system total power curve at the same time on the same day ; (2) Calculate the mean and standard deviation of power for the predicted period: ; in, To predict the historical average power for a given period; The historical power standard deviation; For predicted future time points; The default number of days for historical data lookup is 7 days. (3) Criteria for determining the equilibrium window: ;in, This is a Boolean flag for the equalization window. This is the system's rated power. This is the window power ratio threshold, with a default value of 0.

2. This is the power stability ratio threshold, with a default value of 0.

1. (4) Calculate the window confidence level: ;in, The confidence level for the window, ranging from 0 to 1.

6. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S7, the token data structure includes the following fields: (1) token_gen: Token algebra number, a positive integer; (2) holder_id: The current holder module number, a positive integer; (3) soc_global_snapshot: Global SOC mean snapshot, dimensionless; (4) field_strength_max: The maximum value of the global field gradient, dimensionless; (5) window_active: Whether the current window is in a balanced window period, Boolean; (6) strategy_version: Current strategy version number, a positive integer; The token holding module executes the following logic during the holding period: (1) Update the global SOC snapshot in the token; (2) Determine whether this module needs to perform load balancing; (3) If balancing is required and the current balancing window is in progress, perform the balancing operation in step eight; (4) If balancing is not required or balancing has been completed, pass the token to the next module in the logical ring; Token loss recovery mechanism: Each module maintains a token timeout timer. If no token message is detected on the CAN bus within the timeout period, the active module with the smallest number regenerates a new token and restarts the loop from its own position.

7. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S8, the magnitude of the equilibrium current is proportional to the magnitude of the potential field gradient: ;in, The equalization current setting, in amperes (A); The upper limit of the rated current of the equalization converter is set to 5A; This is the potential-to-current conversion coefficient, with a default value of 2.0A, which is adjusted by the strategy evolution module. This represents the magnitude of the local potential field gradient. The BHI conduction coefficient of the channel; The equilibrium direction is determined by the sign of the potential field gradient: When the voltage is greater than 0, energy flows from module i to module j. The converter duty cycle is calculated using the following formula: ;in, This represents the PWM duty cycle, ranging from 0.1 to 0.

9. The target output voltage; Input voltage; For converter conversion efficiency; The maximum equilibrium duration during a single token holding period is Single equilibrium energy The calculation is as follows = ;in, This represents the average voltage during the equalization period.

8. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S9, the metrics for the global equilibrium effect are calculated: ;in, The standard deviation of the system's SOC is given; the equilibrium efficiency index for this round is also calculated. ;in, To balance efficiency indicators; This represents the change in the standard deviation of SOC before and after this round of equilibrium. This represents the total balanced energy consumption for this round.

9. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S11, the anomaly detection module calculates the comprehensive anomaly score for each module in real time: ;in, For the anomaly score of the i-th module; Weighted by the rate of change of internal resistance, default value is 0.4; The time rate of change of the module's internal resistance; Used as a reference rate of change in internal resistance; The weight for temperature deviation is 0.35 by default. For module temperature; This is the temperature warning value; This is the upper limit of the temperature range; This is the communication quality weight, with a default value of 0.

25. For communication packet loss rate; when Exceeding the isolation threshold At that time, perform potential field isolation: isolate the potential energy of the module. Forced zeroing is equivalent to "removing" it from the potential field. The potential field is automatically redistributed among the remaining modules without explicit path replanning. If the module is holding a token, the token is forcibly passed to the next active module.

10. The module-level distributed active load balancing method and system according to claim 1, characterized in that: In step S12, the strategy evolution and versioning module performs the following operations: (1) Record the current strategy version and its corresponding equilibrium efficiency ; (2) Stored in a circular buffer, buffer capacity ; (3) From the most recent Choose from the following versions (5 by default): The highest version As a benchmark; (4) In Generate new parameter combinations within the parameter space neighborhood. : ;in, The value of parameter p in the new version; The value of parameter p in the baseline version; The exploration range for parameter p, with a default value of 0.1; for[ Uniformly distributed random numbers over an interval; (5) Rollback determination: If continuous After the next update All below of Then it will automatically roll back to The parameter snapshot is completed within one control cycle.

Images