Differential equalization control method for storage batteries of compact power supply system

Abstract

The invention discloses a differential equalization control method for a storage battery of a compact power supply system, and the method comprises the steps: obtaining the original electrical quantity and environment quantity of a battery, and determining the state and environment characteristic quantity representing the operation state and microenvironment characteristics of the battery according to the original electrical quantity and environment quantity; based on the characteristic quantity, generating a thermal steady state mark and a prospective dew point crossing mark; the two marks are combined to judge the balance permission state, and security gating of first permission is constructed; and when the equalization permission state is true, a differentiated final equalization parameter is decided for the permitted battery by solving the constraint optimization problem considering the microenvironment risk. According to the method, a decision process of first permission and then optimization is constructed, and the safety, accuracy and effectiveness of balance control under complex working conditions are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent battery management, and in particular to a method for differential equalization control of batteries in a compact power system. Background Technology

[0002] In compact power systems, battery packs are typically composed of numerous individual cells connected in series and parallel. However, due to slight differences in manufacturing processes, inconsistencies in operating environments, and the randomness of charging and discharging, inconsistencies inevitably arise between individual cells, primarily manifested in differences in voltage, capacity, internal resistance, and state of charge (SOC). This inconsistency severely restricts the overall performance of the battery pack, causing the usable capacity to be determined by the weakest link in the chain—the worst-performing cell. More seriously, long-term accumulated inconsistencies can lead to overcharging or over-discharging of some cells, accelerating battery aging. Therefore, actively eliminating or mitigating this inconsistency through battery balancing control technology is of crucial technical significance and application value for ensuring the safety of compact power systems, extending their cycle life, and improving their energy utilization efficiency.

[0003] Currently, the core logic of most battery balancing control technologies widely used in the industry is based on the direct monitoring of electrical parameter differences between individual battery cells. A mainstream approach involves pre-setting a fixed threshold for voltage or state of charge (SOC) difference in the Battery Management System (BMS). During system operation, the BMS periodically checks the terminal voltage or estimated SOC value of all cells within the battery pack. Once the voltage or SOC difference between any two cells exceeds the preset threshold (e.g., a voltage difference greater than 20 millivolts), the balancing control module is activated. Depending on the topology of the balancing circuit, this module performs dissipative discharge (passive balancing) on ​​the cell with the higher SOC parameter, or transfers energy from the cell with the higher SOC parameter to the cell with the lower SOC parameter (active balancing), until the parameter differences between the cells fall back within the preset tolerance range. This on / off control strategy based on a single electrical quantity threshold is widely deployed and applied in many existing compact power systems due to its intuitive principle, simple implementation, and low computational overhead, forming the mainstream technological foundation in the current field of battery balancing control.

[0004] However, the traditional equalization control strategy that relies solely on static electrical quantity thresholds has increasingly revealed its inherent technical defects when facing complex and ever-changing actual working environments. This is mainly reflected in the fact that its decision-making mechanism fails to fully couple the thermodynamic dynamic process of the battery with the dynamic risks of the microenvironment, resulting in inappropriate timing, inaccurate triggering, and uneven dosage of equalization control. Summary of the Invention

[0005] The purpose of this invention is to provide a method for differential equalization control of batteries in a compact power system, so as to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution: A method for differentiated balancing control of batteries in a compact power system, including:

[0007] Obtain the original electrical and environmental quantities of the battery, and determine the battery state and environmental characteristic quantities that characterize the battery's operating state and microenvironmental properties accordingly.

[0008] Based on battery state and environmental characteristics, generate thermal steady-state indicators and dew point crossing indicators;

[0009] The combined thermal steady-state indicator and dew point crossing indicator are used to determine the equilibrium permissible state.

[0010] When the equilibrium permitting state is true, the final equilibrium parameters are determined by solving a constrained optimization problem that takes into account micro-environmental risks, based on the battery state and environmental characteristics.

[0011] Beneficial effects: This invention establishes a decision-making process of prior approval followed by optimization, which improves the safety, accuracy, and effectiveness of balanced control under complex operating conditions. Attached Figure Description

[0012] Figure 1 A flowchart illustrating the steps of a battery differential balancing control method for a compact power system provided in this application embodiment.

[0013] Figure 2 A flowchart illustrating the steps for determining the final equilibrium parameters provided in this application embodiment.

[0014] Figure 3 A flowchart illustrating the steps for establishing and optimizing the objective function provided in this application embodiment.

[0015] Figure 4 A flowchart illustrating the steps for applying constraints in an embodiment of this application. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0018] The study revealed the following problems with existing solutions: Inappropriate timing due to neglecting thermal hysteresis. After high-rate operation such as rapid charging and discharging, compact power systems accumulate significant heat inside the battery, causing the core temperature to be much higher than the surface temperature. Even after the current stops, due to the inherent delay in heat conduction from the core to the surface, the surface temperature and terminal voltage may have already stabilized, but the battery interior remains in a thermally unstable state where heat has not been fully dissipated. Traditional methods may mistakenly determine that the battery is ready for balancing based on the increasingly uniform voltage at this point, thus initiating balancing operations. This additional heat source applied while the thermal gradient is still significant exacerbates the non-uniformity of the internal temperature field, accelerates material aging, and may even create potential safety hazards such as localized overheating. Furthermore, there is the problem of inaccurate triggering due to ignoring changes in ambient humidity. In environments with large diurnal temperature variations or high humidity, when the battery surface temperature drops to near the air dew point due to environmental changes, tiny, imperceptible water droplets can easily condense on its surface or metal terminals. Condensation can create unintended conductive paths, potentially causing transient and abnormal voltage spikes, resulting in a false voltage difference exceeding a preset threshold. Traditional methods cannot identify this artifact and will initiate invalid equalization operations based on this erroneous signal. More dangerously, energizing terminals in a humid environment significantly accelerates electrochemical corrosion, potentially leading to micro-short circuits in the long run and threatening system electrical safety. There is also the issue of uneven dosage due to a lack of differentiated consideration. Within a compact battery pack, the microenvironment (especially heat dissipation conditions) of cells in different locations (such as near heat dissipation ducts or in heat dissipation dead zones) varies significantly. Traditional methods employ a one-size-fits-all equalization strategy, applying the same equalization current or duration to all cells. This inevitably leads to excessive temperature rise in poorly heated cells during equalization, while well-heated cells may be under-equalized. Over long-term operation, this non-differentiated equalization method amplifies rather than reduces inconsistencies between cells, contradicting the original intention of equalization control.

[0019] To better understand the technical background and the technical problem to be solved in this application, a common battery balancing control method in the prior art is first introduced. In traditional battery management systems, the triggering logic for balancing control usually relies solely on monitoring the voltage or state of charge (SOC) differences between individual battery cells. Specifically, the system presets a voltage difference threshold, such as 20mV. During system operation, the battery management unit (BMS) periodically detects the terminal voltage of all individual cells in the battery pack. When the voltage difference between any two cells exceeds the preset 20mV threshold, the balancing control circuit is activated, discharging the cell with the higher voltage or transferring energy from the cell with the higher voltage to the cell with the lower voltage, until the voltage difference falls back within the threshold range.

[0020] However, this control strategy, which relies solely on a single electrical quantity threshold, has significant technical drawbacks in complex real-world operating environments. For example, after rapid charging or high-current discharging of a battery pack, a significant temperature rise occurs inside the battery, with its core temperature (T0) reaching a certain level. core The temperature is much higher than the surface temperature (T) that can be measured by external sensors. surf After the charging and discharging process is completed, due to the delayed heat conduction from the core to the surface (i.e., thermal hysteresis), the surface temperature will drop first, while the core temperature remains at a higher level. At this time, if the system judges the battery state solely based on the stabilized terminal voltage, it may mistakenly assume the battery has entered a suitable quiescent state for equalization, thus initiating equalization. This equalization, performed before heat has completely dissipated and the internal thermal gradient remains large, generates additional heat, exacerbates the unevenness of the battery's internal temperature, and may even lead to localized overheating, accelerating battery aging and posing safety hazards. In environments with large diurnal temperature differences or high humidity, especially when a vehicle is moved from a warm interior to a cold exterior, or in coastal or rainy areas, condensation easily forms inside the compact power system. When the battery surface temperature drops to near or below the dew point temperature (T0) of the surrounding air due to the decrease in ambient temperature... dewWhen a battery is energized, tiny water droplets can condense on its surface or metal terminals. These droplets can form unexpected conductive paths, potentially causing transient voltage spikes or measurement deviations, resulting in a false voltage difference exceeding a preset threshold. Traditional methods initiate equalization based on this erroneous signal, which is not only ineffective but also exacerbates electrochemical corrosion of the terminals in humid environments, potentially leading to micro-short circuits and threatening the safety of the entire power system. Within the same battery pack, due to different installation locations (e.g., near heat dissipation ducts or in heat dissipation dead zones), the microenvironment (especially temperature and heat dissipation conditions) of different individual cells varies significantly. Traditional methods employ a one-size-fits-all equalization strategy, applying the same equalization current or equalization time to all cells to be equalized. This can cause cells in poorly ventilated locations to overheat during equalization, while cells in well-ventilated locations may be under-equalized. Over time, this equalization method, which ignores microenvironmental differences, amplifies rather than reduces inconsistencies between cells.

[0021] In summary, existing technologies fail to fully consider the complexity of the battery's thermal dynamics and microenvironment, resulting in inappropriate timing, inaccurate triggering, and uneven dosage in the equilibrium decision-making process, thus failing to achieve safe and efficient equilibrium control.

[0022] like Figure 1 As shown, a method for differentiated equalization control of batteries in a compact power system is proposed, including the following steps:

[0023] Obtain the original electrical and environmental quantities of the battery, and determine the battery state and environmental characteristic quantities that characterize the battery's operating state and microenvironment based on the original electrical and environmental quantities.

[0024] It can also be used to obtain the original electrical and environmental quantities of the battery, preprocess them to obtain the processed original electrical and environmental quantities, and determine the battery state and environmental characteristic quantities that characterize the battery's operating state and microenvironmental characteristics based on the processed original electrical and environmental quantities.

[0025] In this embodiment, the system first collects a series of raw data using sensors deployed on or around the battery. Exemplarily, electrical quantities mainly include the terminal voltage of each individual battery cell and the total current of the battery pack. Environmental quantities mainly include the surface temperature of each individual battery cell or its vicinity and the relative humidity. This raw data may come from different acquisition channels with different sampling rates and timestamps. Therefore, it is necessary to perform time synchronization alignment and necessary filtering and noise reduction on this data to form a unified data foundation. Based on this processed foundational data, a series of more informative feature quantities are derived through calculation or model estimation. For example, the dew point temperature is calculated using thermodynamic formulas, the core temperature of the battery is estimated by establishing an online thermal model, and the difference between the surface temperature and the dew point temperature is calculated. These feature quantities together constitute a comprehensive description of the current battery state and microenvironment.

[0026] Based on battery state and environmental characteristics, thermal steady-state indicators and dew point crossing indicators are generated.

[0027] Specifically, before executing the equilibrium decision, two independent, physically safe gating checks are established. The first is a thermal steady-state indicator, designed to determine whether the battery is in a thermodynamically stable state. Only when the internal heat exchange of the battery tends to equilibrium, the difference between the core temperature and the surface temperature stabilizes within a small range, and there are no large charging / discharging current disturbances, can the measured electrical quantities truly reflect the battery's internal state. Optionally, a finite state machine mechanism is used to determine the thermal stabilization process; a true thermal steady-state indicator is generated only when the state machine enters an equilibrium state. The second is a dew point crossing indicator, designed to mitigate the risks associated with condensation. Optionally, by proactively predicting condensation risks and combining this with monitoring actual environmental parameters, it is determined whether the battery is currently in or about to enter a dangerous window period where condensation may occur. Once such a risk is detected, the system generates a true dew point crossing indicator and maintains it for a period until the risk is completely eliminated.

[0028] The combined thermal steady-state indicator and dew point crossing indicator are used to determine the equilibrium permissible state.

[0029] In this embodiment, the logical determination steps are provided. Equalization operations can only be considered under the premise of absolute safety. Therefore, it can be stipulated that the equalization permission state of the corresponding battery is determined to be true only when the thermal steady-state flag is true (indicating that the battery's thermal state is stable) and the dew point crossing flag is false (indicating no risk of condensation). If either condition is not met, such as the battery being in a thermally unstable state or having a risk of condensation, then the equalization permission state of the battery is false, and subsequent equalization operations will be forcibly prohibited.

[0030] When the equilibrium permitting state is true, the final equilibrium parameters are determined by solving a constrained optimization problem that takes into account micro-environmental risks, based on the battery state and environmental characteristics.

[0031] Alternatively, it can determine whether the equilibrium permit state is true. If it is, based on the battery state and environmental characteristics, it determines the final equilibrium parameters by solving a constrained optimization problem that takes into account micro-environmental risks. Otherwise, it pauses the equilibrium operation and records the reason.

[0032] In this embodiment, instead of a fixed balancing strategy, all batteries that have been granted balancing permission are transformed into an online optimization problem. This optimization problem comprehensively considers the expected benefits and potential risks of balancing. Specifically, the system constructs an optimization objective function that aims to maximize the benefits of balancing (e.g., minimizing battery inconsistencies after balancing) while minimizing the risks arising from the balancing operation in the current microenvironment (e.g., the higher the risk of a battery, the greater the cost of applying balancing). Optionally, the solution process is also subject to a series of strict constraints, such as the temperature rise caused by balancing not exceeding a safety limit and the total balancing energy not exceeding the system's heat dissipation capacity. By solving this constrained optimization problem, an optimal combination of balancing parameters can be determined for each permitted battery, including, for example, a voltage difference trigger threshold, a balancing current limit, and a single balancing duration. This replaces the traditional one-size-fits-all strategy and achieves differentiated, adaptive, and risk-aware fine-grained balancing control for each battery.

[0033] This embodiment establishes a decision-making process of granting permission first and then optimizing, which improves the safety, accuracy and effectiveness of battery equalization control.

[0034] In one possible implementation, determining battery state and environmental characteristics includes data acquisition and preprocessing steps. Specifically, raw data streams are acquired using sensors deployed at different locations within the battery pack. This data includes, but is not limited to, the raw surface temperature of each or every group of battery cells, the raw relative humidity of the surrounding environment, the raw terminal voltage of each cell, and the raw total current of the battery pack. Because the sampling clocks of each sensor channel may deviate, the acquired data is asynchronous in time. To establish a unified data benchmark, the system first performs clock synchronization and resampling. Specifically, using a high-precision system reference clock as a benchmark, the timestamps of each channel are calibrated using a timestamp drift correction algorithm. All data streams are resampled to a fixed standard sampling period, such as 10 Hz. During the resampling process, piecewise linear interpolation or bounded spline interpolation can be used to fill in the data points, and the interpolated data is marked to distinguish it from the original acquired data.

[0035] Optionally, after time alignment, the data undergoes denoising and anomaly suppression. For example, for relatively stable physical quantities such as temperature and humidity, a sliding window mid-range filter combined with a first-order low-pass filter can be used to suppress spike noise. For signals such as voltage, which may exhibit rapid dynamics, methods such as Hampel filters can be used to remove glitches while preserving their true dynamic characteristics. After the above processing, the time-aligned and low-noise filtered surface temperature, filtered relative humidity, filtered terminal voltage, and aligned battery pack current are obtained.

[0036] In another possible implementation, to quantify the reliability of each sensor data channel and determine battery state and environmental characteristics, the method also includes a step of generating and utilizing channel confidence labels, specifically:

[0037] Evaluate the health indicators of any data channel in the raw electrical and environmental quantities. The health indicators include at least: the percentage of missing data within a preset time window, the number of times data saturation occurs (i.e., the number of times the data value reaches the upper or lower limit of the sensor range), and the noise level of the signal.

[0038] The noise level can be estimated online, for example, by calculating the root mean square error of the signal and its median filtering result within a short sliding window.

[0039] The health indicators are normalized to generate channel confidence labels that characterize the reliability of the data channel.

[0040] In other words, the health indicators are normalized and then fused into channel confidence labels using a weighted function.

[0041] An example calculation formula is as follows: Confidence = clip(1 - a * Missing) Rate -b*Saturation Count -c*Noise Norm (0, 1); where Confidence is the final channel confidence label; Missing Rate Saturation is the normalized missing data rate. Count Noise is the normalized saturation number. Norm is the normalized noise level; a, b, and c are preset, calibrable weighting coefficients used to adjust the influence of different health indicators on the final confidence level; the clip function is used to limit the calculation results to the interval [0, 1]. The channel confidence label is a value between 0 and 1; the closer the value is to 1, the higher the reliability of the data channel.

[0042] When generating dew point crossing markers, the safety threshold for condensation risk assessment is dynamically adjusted based on the channel confidence label; and the channel confidence label is used as one of the inputs for constructing the microenvironment risk index.

[0043] Furthermore, to enable confidence labels to dynamically reflect real-time changes in signal quality, a confidence fusion and update mechanism is included after generating channel confidence labels. Specifically, this involves periodically estimating the signal-noise variance of the data channels online, for example, by analyzing the high-frequency components of the signal. Based on the signal-noise variance, the channel confidence labels are fused and updated according to preset weights. Specifically, the latest estimated noise variance is used as an independent indicator reflecting immediate signal quality, and existing channel confidence labels are updated according to preset fusion weights. This update can take the form of an exponential moving average, allowing the confidence labels to reflect both long-term channel health and rapid response to short-term signal quality degradation.

[0044] Optionally, the steps of determining battery state and environmental characteristics may further include: estimating the noise variance of any data channel online. The online-estimated noise variance can also be used to adaptively adjust the parameters of the data filtering algorithm. For example, the noise variance can be transformed into a suggested smoothing window width for subsequent median filtering or low-pass filtering steps using a preset mapping relationship (such as a lookup table or piecewise function). When the noise variance is large, the filtering window width is automatically increased to enhance the smoothing effect; conversely, the window width is decreased to retain more signal details.

[0045] In another possible implementation, the battery's core temperature is a key parameter for assessing its internal state and safety, but it is often difficult to measure directly. The steps for determining the battery's state and environmental characteristics also include a process for estimating the core temperature, which involves applying an online thermal model and recursively deriving the core temperature estimate in real time based on the filtered surface temperature and aligned battery pack current as input.

[0046] Preferably, the online thermal model can be a first-order or second-order RC (resistor-capacitor) equivalent circuit model. Taking a first-order RC model as an example, its physical process can be described by the following differential equation: C th *d(T core ) / dt=(T surf -T core ) / R th +k heat *|I pack |;where T core The core temperature to be estimated; T surf The measurable filtered surface temperature; I pack The current of the aligned battery pack; C th R th and kheat T represents the battery's equivalent heat capacity, equivalent thermal resistance, and heat generation coefficient, respectively; t represents time. In embedded systems, the above continuous-time model is discretized into a recursive form for easier calculation: T core [k+1]=α*T core [k]+(1-α)*T surf [k]+β*|I pack [k]|; where k is the discrete time step; α and β are related to the model parameters R. th C th k heat And the coefficient related to the sampling period Δt. Through this recursive formula, the system can update the estimated core temperature in real time during each sampling period.

[0047] Optionally, the thermal model parameters are specifically: equivalent thermal resistance R th The value ranges from 1.0 to 2.0 K / W, and the equivalent heat capacity C th The value ranges from 500 to 1200 J / K, and the heat production coefficient k heat The value range is 0.001-0.01 KW / A 2 Discretization coefficient α = exp(-Δt / (R) th ·C th ), where Δt is the sampling period, typically 0.1 seconds; β = k heat ·Δt / C th .

[0048] Based on the model residual between the core temperature estimate and the filtered surface temperature, the model parameters of the online thermal model are corrected online.

[0049] In this embodiment, to address the problem of model parameter mismatch caused by battery aging or changes in operating conditions, the model parameters of the online thermal model are corrected online. Specifically, the system identifies and corrects the model parameters online based on the model residual between the core temperature estimate and the filtered surface temperature. Preferably, recursive least squares (RLS) with a forgetting factor can be used. To avoid insufficient system excitation (e.g., when the battery is in a static state for a long time, I... pack Technical problems such as parameter drift or divergence in identification results (approaching zero) occur. The parameter correction procedure only performs the correction when preset excitation conditions are met (e.g., |I pack The system is triggered only when the amplitude of the signal exceeds a certain threshold (e.g., C / 20 of the rated capacity) for a sustained period. Simultaneously, physical boundaries are set for the identified parameter values, and amplitude limiting is applied to ensure they remain within a reasonable physical range. As an optional implementation, when the battery's thermal dynamics are complex or the thermal hysteresis effect is significant, the system can automatically select or switch to a second-order RC model to improve the accuracy of core temperature estimation.

[0050] Furthermore, after determining the battery state and environmental characteristics, a unified Micro-Environment Risk Index (MRI) can be constructed to quantify the comprehensive risk level faced by a single battery in its current microenvironment. This index is obtained by weighted synthesis of multiple characteristics. Specifically, constructing the MRI involves: resolving filtered surface temperature, filtered relative humidity, the gap between surface temperature and dew point, and a core temperature estimate from the battery state and environmental characteristics. The resolved surface temperature, relative humidity, gap, and core temperature estimate are then weighted and synthesized to generate the MRI.

[0051] An exemplary synthesis formula is as follows: MRI = w1 * f1(T core T surf )+w2*f2(Gap)+w3*f3(RH)+w4*f4(Confidence); where MRI is the final risk index; w1 to w4 are calibrable weighting coefficients; f1 to f4 are the corresponding risk contribution functions. f1(T core T surf The value of f(x) is used to characterize thermal risk; for example, it can be the absolute value of the core temperature or a function of the temperature difference between the core and the surface. The higher the temperature, the greater the risk. f2(Gap) is used to characterize condensation risk; for example, max(0, Gap) = f(x) + ... threshold -Gap), meaning the smaller the Gap value (closer to the dew point), the greater the risk. Gap represents the difference between surface temperature and dew point temperature. f3(RH) characterizes humidity risk and can be a function of relative humidity. Higher humidity indicates greater risk. RH represents relative humidity. f4(Confidence) characterizes data quality risk. For example, 1-Confidence means that the lower the data confidence level, the less accurate the system's understanding of the state, and the higher the risk should be considered.

[0052] In another exemplary embodiment, the microenvironment risk index is specifically calculated as follows: MRI i =0.2×|T core -T surf |+1.0×max(0,1.0-Gap)+0.5×RH+2.0×(1-Confidence); where the temperature difference term has a weight of 0.2 reflecting the contribution of thermal risk, the gap term has a weight of 1.0 reflecting the high priority of condensation risk, the humidity term has a weight of 0.5 indicating the impact of environmental humidity, and the confidence term has a weight of 2.0 emphasizing the importance of data quality for risk assessment.

[0053] This embodiment can extract and generate a standardized, information-rich set of battery state and environmental characteristics that include reliability metrics from raw, multi-source sensor data, providing a solid data foundation for subsequent thermal steady-state gating, dew point crossing gating, and risk budget-based equilibration.

[0054] According to one aspect of this application, in a compact power system, the recovery of the internal thermal state of a battery after experiencing thermal disturbances such as high-current charging and discharging is a gradual process, directly affecting the reliability of the equalization decision. To address this issue, a thermal hysteresis steady-state gating (TLS) mechanism is introduced. Furthermore, a thermal steady-state flag is generated, including:

[0055] Construct a finite state machine that includes at least three states: thermal instability, thermal stabilization, and equilibrium, to model the thermal stabilization process of the battery.

[0056] Alternatively, a finite state machine can be constructed that includes three states: thermal instability, thermal stabilization, and equilibrium, to model the thermal stabilization process of the battery.

[0057] Specifically, the thermally unstable state is the system's default initial state, or a state forcibly entered after detecting a significant thermal disturbance (such as a large current surge). In this state, there is a large or uncertain gradient between the battery's core temperature and surface temperature, and the heat flow is not yet balanced. Therefore, any equalization operation is strictly prohibited, and the thermal stability flag is false. In the thermally pending-stable state, as the thermal disturbance weakens and the battery begins to stabilize, the state machine transitions from the thermally unstable state to this transitional state. In this state, the system continuously monitors the battery's thermal dynamics, but equalization operations are still prohibited because the sustainability of the stabilization trend is yet to be confirmed. Only when the battery's thermal stability trend is confirmed, and this stable state persists for a preset period of time, does the state machine transition from the thermally pending-stable state to the equalization-permitted state, i.e., the final state. The thermal stability flag is set to true if and only if the finite state machine is in the equalization-permitted state, indicating that equalization operations are safe and the data is reliable at this time.

[0058] Based on battery state and environmental characteristics, determine whether the thermal stability trend condition is met.

[0059] Specifically, the core driving force for the state machine's transitions between the three states mentioned above comes from the periodic determination of the thermal stability trend condition. Optionally, determining whether the thermal stability trend condition is met includes: within a preset time window, resolving the core temperature estimate, the filtered surface temperature, and the aligned battery pack current from battery state and environmental characteristics; jointly verifying at least three criteria: the rate of change of the core temperature estimate, the temperature difference between the core temperature estimate and the filtered surface temperature, and the amplitude of the aligned battery pack current; optionally, the rate of change of the core temperature estimate can be: calculating the absolute value of the maximum rate of change of the core temperature estimate within the time window; the core and surface temperature difference can be: calculating the absolute value of the temperature difference between the core temperature estimate and the filtered surface temperature; the amplitude of the battery pack current can be: evaluating whether the amplitude of the aligned battery pack current is within a sufficiently small range. The thermal stability trend condition is deemed met if and only if the (three) criteria simultaneously satisfy their respective stability thresholds within the time window. For example, these stability thresholds can be set as follows: the core temperature change rate is no greater than 0.02-0.10 degrees Celsius / second, the core-to-surface temperature difference is no greater than 1.0-2.5 degrees Celsius, and the current amplitude is no greater than C / 30 of the battery's rated capacity.

[0060] Transitions are performed between states of the finite state machine based on whether the thermal stability trend condition is met and the duration of the thermal stability trend condition being met.

[0061] Optionally, to ensure the smoothness and robustness of state transitions and avoid frequent state switching due to signal noise or instantaneous fluctuations, the transition process between states in the finite state machine includes hysteresis and timer mechanisms: First, a hysteresis threshold is configured for state transitions, making the thermal stability tendency condition required to enter a more stable state stricter than the condition for exiting that stable state. For example, transitioning from a thermally unstable state to a thermally stable state may require a core-to-surface temperature difference of less than 1.5 degrees Celsius; while the triggering condition for returning from a thermally stable state to a thermally unstable state may be relaxed to a temperature difference greater than 2.0 degrees Celsius. This creates a stable state that is easy to exit but difficult to enter, effectively suppressing jitter near the critical point. Second, a minimum dwell time is set for state transitions, stipulating that a transition from the current state to the next state is only permitted if the thermal stability tendency condition is met for more than the corresponding minimum dwell time. For example, even if the thermal stability tendency condition is met, the state machine must remain in the thermally stable state for at least τ seconds. quiet It takes seconds to transition to a state that can achieve equilibrium.

[0062] The thermal steady-state flag is set if and only if the finite state machine is in an equilibrium state; otherwise, the thermal steady-state flag is cleared.

[0063] Furthermore, it also includes advanced designs for dynamic self-tuning and safety interruption. Specifically, the stability threshold is a weighted threshold for dynamic self-tuning based on environmental state labels and channel confidence labels determined from battery state and environmental characteristics.

[0064] Specifically, the system can preset a set of baseline thresholds and adjust these thresholds in real time based on the values ​​of environmental state labels (e.g., rapid cooling or stabilization) and channel confidence labels, using a lookup table or weighting function. For example, when the confidence of a temperature sensor channel is too low, the system will automatically tighten (reduce) the stabilization thresholds for temperature difference and rate of change to compensate for the risks caused by data uncertainty.

[0065] Furthermore, the transition steps between the three states are also subject to independent safety interruption rules. These rules stipulate that when the core temperature estimate or battery pack current, derived from the battery state and environmental characteristics, exceeds a preset safety limit, the current state of the finite state machine is forced to immediately revert to the thermally unstable state, regardless of the current state. This is the highest priority safety mechanism, independent of conventional state transition logic, ensuring that the equalization permission is immediately and unconditionally revoked under any extreme conditions (such as battery overheating or a sudden current surge caused by an external short circuit).

[0066] This embodiment uses a three-state finite state machine with hysteresis, timer, dynamic threshold and safety interruption mechanism to accurately generate thermal steady state flag, providing a reliable safety premise for subsequent equilibrium decision-making.

[0067] In an optional implementation, generating a thermal steady-state flag further includes: obtaining the surface temperature change rate from battery state and environmental characteristics, and estimating the second-order rate of change of surface temperature based on the filtered surface temperature contained in the battery state and environmental characteristics; generating a thermal steady-state flag when the surface temperature change rate and the second-order rate of change of surface temperature are both below their respective preset stability thresholds.

[0068] Specifically, the system obtains the first derivative of the surface temperature, i.e., the rate of change of surface temperature, from the filtered surface temperature time series through differential calculation or linear regression fitting within a short sliding window. Using sliding window linear regression, compared to simple differential calculation, better suppresses noise interference. Based on the obtained first derivative sequence, differential calculation is performed again, or a second-order central difference is directly performed on the original filtered surface temperature sequence to estimate the second-order rate of change of surface temperature, which physically corresponds to the acceleration of temperature change. The absolute value of the surface temperature rate of change is compared with a preset rate threshold (e.g., 0.05 degrees Celsius / second), and simultaneously with a preset acceleration threshold. Only when the absolute values ​​of both indicators are simultaneously below their respective stability thresholds does the system determine that the battery has reached a thermally stable state and generate a true thermally stable state flag; otherwise, the flag is false.

[0069] In this embodiment, since an object in a thermodynamically stable state not only changes its temperature slowly (low rate), but its temperature change trend should also be stable (low acceleration). By simultaneously constraining the first and second derivatives of the surface temperature, cases where the instantaneous rate of change may be zero (such as when the temperature reaches a local extreme point) but the overall process is still undergoing drastic change can be effectively filtered out, thereby improving the accuracy of the judgment.

[0070] Optionally, to further improve the stability of the scheme, hysteresis control can be introduced for determining the rate threshold and acceleration threshold. That is, the threshold for entering the steady state from the unsteady state is set more strictly than the threshold for exiting the steady state. It should be emphasized that although the internal implementation logic of this embodiment differs from the method based on core temperature and state machine, the function and interface of its final output thermal steady-state flag are completely consistent and interchangeable throughout the entire control method, ensuring the portability and applicability of this application across different hardware configurations and computing platforms.

[0071] According to one aspect of this application, generating a dew point crossing mark includes:

[0072] The difference between surface temperature and dew point temperature (Gap) and the trend of environmental changes are determined from battery state and environmental characteristics.

[0073] Specifically, based on the filtered surface temperature (denoted as T) surf The system calculates the dew point temperature (denoted as T) under the current environment using the Magnus approximation formula, based on the filtered relative humidity (RH) and the current ambient temperature. dew A widely applicable formula within the temperature range of -40 to 50 degrees Celsius is as follows: γ=(a*T surf ) / (b+T surf )+ln(RHcalc );T dew = (b*γ) / (a-γ); where a and b are empirical constants, which can take values ​​of 17.62 and 243.12 respectively, applicable to a temperature range of -40℃ to 50℃; ln is the natural logarithm function, and γ is an intermediate variable used to simplify the dew point temperature calculation formula, RH calc The relative humidity (RH) is robustly processed. To ensure the stability of the calculated values, especially when the sensor may output extremely low humidity values, the input relative humidity (RH) is robustly processed. calc =max(RH, 0.01), using lower limit clamping to avoid calculation failures caused by taking the logarithm of zero or negative values. The key criterion—the difference between surface temperature and dew point temperature, Gap, is calculated, i.e., Gap = T. surf -T dew This value intuitively reflects the safety margin of the surface temperature from the condensation point.

[0074] Based on the gap difference and environmental change trends, the risk of condensation is predicted and a condensation risk warning is generated.

[0075] In other words, based on the gap and environmental change trends, the risk of condensation can be predicted in a forward-looking manner, and a condensation risk warning can be generated.

[0076] Optionally, to achieve proactive protection, a condensation risk warning is generated, including: applying a pre-configured short-term prediction model to estimate the predicted value of the Gap difference for the next time step based on the historical sequence of the Gap difference; analyzing the temperature change trend and humidity change trend from the environmental change trend; and generating a condensation risk warning when the predicted value of the Gap difference is lower than the preset condensation warning threshold, and the temperature change trend shows a cooling trend while the humidity change trend shows an increasing trend.

[0077] In this embodiment, optionally, the system uses an exponential moving average (EMA) or a first-order autoregressive (AR) model as a short-term prediction model to perform a one-step prediction of the time series of the Gap value, thereby obtaining the predicted value of Gap. pred Simultaneously, the system uses the calculated rates of surface temperature change and relative humidity change to determine the current trends in temperature and humidity. A high-risk indicator is a decreasing temperature (negative rate) and increasing humidity (positive rate). When gaps... pred When the predicted value is lower than the preset condensation warning threshold (e.g., 0.2-1.0 degrees Celsius) and occurs simultaneously with a cooling and humidification trend, the system will generate a condensation risk warning signal.

[0078] In an optional embodiment, to further enhance the robustness and adaptability of the early warning mechanism, channel confidence labels can also be used. The use includes: when prospectively predicting condensation risk, using channel confidence labels as weights to weight the features used for prediction; and when the value of the channel confidence label is lower than a preset confidence threshold, automatically amplifying the safety threshold used to determine the condensation risk early warning.

[0079] In this embodiment, during feature weighting, features such as the gap value or its rate of change input into the prediction model are first multiplied by the channel confidence level of their corresponding source (e.g., a temperature or humidity sensor). This means that when the data quality of a sensor deteriorates, its influence in the prediction model will also decrease accordingly. In the safety threshold amplification mechanism (or confidence barrier), if the confidence label of the humidity sensor is below a threshold value (e.g., 0.7), the system will automatically raise the trigger threshold for condensation warnings from 0.5 degrees Celsius to 1.0 degrees Celsius, and may require a longer duration of the dangerous trend before triggering the warning. This makes the system more conservative when the situation is unclear, thus ensuring safety.

[0080] When the condensation risk warning is true, or the real-time value of the gap is lower than the preset trigger threshold, the dew point crossing flag is set and the drying timer is started.

[0081] In this embodiment, the logic for generating and clearing the dew point crossing marker integrates the aforementioned predictive warning and reactive testing. A dual-evidence fusion trigger mechanism (triggered when either condition is met) ensures comprehensive protection. Once the marker is set, a configurable drying timer (e.g., 10-30 minutes) is activated or reset.

[0082] Once the gap value returns to the preset release threshold and the relative humidity drops and continues until the drying timer expires, the dew point crossing flag is cleared.

[0083] In this embodiment, the clearing of the flag follows a more stringent recovery logic with double hysteresis. For example, clearing the dew point crossing flag specifically involves: performing the clearing operation only when both conditions are met simultaneously—the gap has recovered to a preset release threshold higher than the trigger threshold and the relative humidity has decreased—and this condition persists until the drying timer expires. Here, the release threshold (e.g., 1.0-2.0 degrees Celsius) is set higher than the trigger threshold, forming the first hysteresis; while the setting of the drying timer requires the safe state to remain stable for a period of time, constituting the second hysteresis. Only when all three conditions (gap recovery, humidity decrease, and timer expiration) are met is the flag cleared, and only then can the equalization operation be allowed.

[0084] In this embodiment, the system generates a clear and reliable dew point crossing indicator as a hard gating signal, which forces a freeze-equalization operation within a time window where there is any risk of condensation, thereby effectively avoiding false triggering and potential electrochemical risks.

[0085] In another optional implementation, generating a condensation risk warning further includes: constructing a feature set for a probabilistic model from battery state and environmental features; inputting the feature set into a pre-trained probabilistic model to estimate the probability of condensation occurrence; and generating a condensation risk warning when the probability of condensation occurrence exceeds a preset cost-sensitive threshold.

[0086] Specifically, a feature set is constructed for the probabilistic model. This feature set may include, but is not limited to, the current Gap value, the rate of change of the Gap value over a past period, the current filtered relative humidity, the rate of change of the relative humidity, the current filtered surface temperature, and the rate of change of the surface temperature. Optionally, historical values ​​of these features (such as the mean, variance, and other statistics over the past few minutes) can also be included in the feature set to provide richer dynamic information. During the product development phase, a large amount of data containing both normal operating conditions and condensation-affected operating conditions is collected experimentally. This data is labeled and then used to train a probabilistic model, i.e., a classification model, such as Logistic Regression or Naive Bayes classifiers. This model learns the mapping relationship from the input feature set to the binary result of whether condensation occurs. At runtime, the model receives the real-time constructed feature set as input, and its output is the probability of condensation occurrence between 0 and 1. Unlike a fixed threshold, this embodiment uses a cost-sensitive threshold. In battery management applications, the safety risks posed by missed condensation (i.e., failure to predict impending condensation, a false negative) far outweigh the equilibrium efficiency losses caused by false positives (i.e., issuing a warning when condensation is not expected, a false positive). Therefore, the threshold setting tends to favor false positives over missed condensation. For example, even if the model outputs a condensation probability of only 0.6, it may still exceed the cost-sensitive threshold, thus generating a condensation risk warning. The specific value of this threshold can be quantified by analyzing the costs of different decision-making errors. It should be noted that the condensation risk warning signal generated by the probabilistic model in this embodiment is functionally equivalent to the warning signal generated by the physical model. It will be used as input to a unified gating flag generation and recovery logic (i.e., dual-evidence fusion triggering and drying timer recovery).

[0087] This embodiment can learn and capture more complex and nonlinear correlation patterns from data than simple physical rules, potentially achieving higher prediction accuracy in certain specific scenarios. The trade-off is the need for prior data acquisition and model training, and relatively higher computational resource requirements during runtime. The use of physical and probabilistic models provides this application with flexible options for different application scenarios.

[0088] According to one aspect of this application, within a specific scheduling cycle, when the equilibrium permit status of one or more battery cells is determined to be true (i.e., the thermal steady-state flag is true and the dew point crossing flag is false), the system does not immediately adopt a preset, fixed strategy for equilibrium, but instead initiates a multi-stage, optimization-based decision-making process.

[0089] Optionally, after determining the equilibrium permitting state and before solving the constrained optimization problem, the method further includes: performing conflict arbitration on all candidate units that have obtained the equilibrium permitting state according to the preset bus current constraint, parallel quantity constraint and thermal margin constraint, and generating an arbitration candidate set; and using the arbitration candidate set as the object of solving the constrained optimization problem.

[0090] In this embodiment, since the physical capabilities (such as total power and heat dissipation capacity) of the equalization circuit are limited, it is usually impossible to perform equalization on all eligible batteries at the same time. Therefore, an arbitration process is required, specifically as follows: A dynamically prioritized priority queue is constructed based on the urgency of equalization for each candidate cell. The ranking criteria can be multi-dimensional, for example, prioritizing equalization of the cell with the largest voltage deviation, or prioritizing equalization of the cell with the lowest microenvironment risk index (MRI) to achieve risk avoidance. Starting from the top of the priority queue, cells are selected one by one, and it is calculated in real time whether adding them to the current equalization task would violate the system's global constraints. These constraints include: Bus current constraint: The sum of the equalization currents of all cells being equalized simultaneously must not exceed the maximum current carried by the equalization circuit bus. Parallel quantity constraint: The number of cells being equalized simultaneously must not exceed the maximum number of channels in the hardware design. Thermal margin constraint: The sum of the estimated heat generation power of all cells being equalized simultaneously must not exceed the remaining heat dissipation capacity of the current battery pack (i.e., total heat dissipation power minus the power generated by the charging and discharging of the battery pack). The system will continuously select cells from the queue until adding any more cells would violate any of the above constraints. At this point, all the selected individuals together constitute the candidate set after arbitration, which serves as the sole object for the next stage of optimization.

[0091] Furthermore, after determining the specific entities participating in this round of optimization, the system needs to set reasonable initial boundaries for the optimization problem to be constructed. Optionally, after constructing the microenvironment risk index and before establishing the optimization objective function, the system further includes: generating an initial value for the voltage difference trigger threshold by applying a monotonically increasing mapping based on the microenvironment risk index; and generating an initial value for the equilibrium current upper limit by applying a monotonically decreasing mapping based on the microenvironment risk index; wherein the initial values ​​for the voltage difference trigger threshold and the equilibrium current upper limit are used as input boundaries to constrain the optimization problem.

[0092] In this embodiment, the abstract risk level (MRI) is transformed into specific, guiding control parameter boundaries. The underlying logic is: the higher the risk, the more conservative the equilibrium behavior should be. For example, the initial value mapping of the voltage difference trigger threshold (monotonically increasing): V thinitial =f inc (MRI); where f inc It is a monotonically increasing function, which can be a piecewise linear function or a smooth nonlinear function. This means that the higher the MRI of a single cell, the larger the voltage difference required to trigger equalization, thus avoiding unnecessary, small-amplitude equalization adjustments to cells in a high-risk state. Initial value mapping of the equalization current upper limit (monotonically decreasing): I maxinitial =f dec (MRI); where f dec It is a monotonically decreasing function. This means that the higher the MRI of a single cell, the smaller the maximum equalization current that can be applied to it, thereby directly limiting the heat generation during the equalization process. These generated initial values ​​are then subjected to system-level safety upper and lower bounds to ensure they are within the absolute limits allowed by the hardware, and then used as boundary conditions for the optimization problem.

[0093] Furthermore, the decision variables for solving the constrained optimization problem are defined as a ternary combination consisting of the voltage difference trigger threshold, the upper limit of the equilibrium current, and the duration of a single equilibrium step. For each individual i in the candidate set after arbitration, its optimal equilibrium parameter {Voltage} needs to be solved. threshold_i I balmax_i T slot_i}

[0094] like Figure 2 As shown, in an exemplary embodiment, the final equilibrium parameters are determined by: constructing a microenvironment risk index based on battery state and environmental characteristics; and establishing an optimization objective function with the joint objective of minimizing the proxy quantity of the imbalance after equilibrium and minimizing the risk penalty term composed of the product of the microenvironment risk index and the equilibrium dose.

[0095] Optionally, such as Figure 3As shown, an optimization objective function is established, including: quantifying the expected equilibrium benefit brought about by the equilibrium dose, wherein the expected equilibrium benefit is used to characterize the expected reduction in battery imbalance after equilibrium; evaluating the risk cost generated by the equilibrium dose in a predetermined microenvironment, wherein the risk cost is determined by multiplying the microenvironment risk index, the equilibrium dose and a preset risk weight coefficient; and combining the expected equilibrium benefit and the risk cost to construct the optimization objective function, thereby forming an explicit trade-off between the equilibrium benefit and the risk cost.

[0096] Specifically, a representative objective function takes the following form: minΣ i (ΔV hat_i (I i T i )) 2 +λ*Σ i MRI i *(I i *T i The structure of this objective function reflects the trade-off between return and risk, where the first term (ΔV) hat_i (I i T i )) 2 It is a proxy for the expected equilibrium return, ΔV hat_i It is the predicted value of the voltage difference after equalization, which can be linearly approximated as ΔV. initial_i -K i *I i *T i K i It is a coefficient related to the battery's internal resistance and the SOC-OCV curve, ΔV initial_i Let be the initial voltage difference of the i-th battery cell before the equalization operation begins. Minimizing the sum of squares of this term is equivalent to minimizing the voltage variance of the entire battery pack after equalization, i.e., achieving the best equalization effect. The second term is MRI. i *(I i *T i (), is the risk cost, which is the microenvironment risk index MRI. i With equal dose (current I) i With duration T i Multiplying the products of (MRI and MRI) together constitutes the risk penalty. This means that towards a high-risk (MRI) level... i Injecting any equilibrium dose of a single monomer with a large value will significantly penalize the objective function. λ is a preset risk weighting coefficient used to adjust the designer's preference between pursuing balanced efficiency and avoiding microenvironmental risks.

[0097] Apply constraints, which include at least: a hard constraint that forces the equalization current to zero based on the state of the dew point crossing mark, and a thermal safety constraint that ensures the upper limit of the temperature rise caused by the equalization dose does not exceed a preset value.

[0098] Specifically, while minimizing the above objective function, the solution process must satisfy a series of strict constraints. For example, such as... Figure 4 As shown, the imposed constraints include: Establishing a hard condensation constraint: This constraint stipulates that when the dew point crossing flag of any battery is true, the equalization current decision variable allocated to that battery remains constant at zero. In implementation, since cells holding a true dew point flag have already been filtered during the permission determination stage, this constraint ensures the completeness of the optimizer's internal logic. Establishing a thermal safety constraint: This constraint maps the equalization dose to an upper bound of the battery core temperature rise using a pre-configured online thermal model, requiring that this upper bound of the temperature rise must not exceed a preset thermal safety threshold. Specifically, using the online thermal model, the equalization dose (I0) can be predicted. i T i The additional temperature rise ΔT caused by this core_i This constraint requires ΔT core_i ≤Θ safe , where Θ safe This is a preset safe temperature rise threshold (e.g., 5-8 degrees Celsius). For ease of calculation, ΔT is usually... core_i The nonlinear relationship between dose and radiation is linearized online or approximated quadratically. Dose budget constraint: This constraint requires that the sum of the equalization doses of all participating batteries must not exceed the current dose budget determined based on the system's heat dissipation capacity. Its mathematical form is Σ i I i *T i ≤E budget Boundary constraints for decision variables: Each decision variable must be within its feasible region, for example, 0 ≤ I. i ≤I maxinitial_i , 0≤T i ≤T maxslot E, etc. budget For the current dose budget, I maxinitial_i For the initial maximum equalization current, T maxslot The goal is to maximize the equilibrium time. This optimization problem can typically be formulated as a quadratic programming (QP) problem, which can be solved within each scheduling cycle (e.g., hundreds of milliseconds) using an efficient embedded solver. As an alternative, when computational resources are extremely limited, less computationally intensive algorithms such as iterative projective gradient descent can be used to obtain an approximate optimal solution.

[0099] Solve for the optimal solution to the objective function under constraints to obtain the final equilibrium parameters.

[0100] In this embodiment, after the solution is completed, the system performs a feasibility repair, which checks whether the solution has slightly exceeded the limit due to numerical errors and projects it back to the feasible region. Finally, a set of optimal, directly executable equilibrium parameters {Voltage} is generated for each participating unit. threshold_i I balmax_i T slot_i}, where Voltage threshold_i I is the voltage difference trigger threshold. balmax_i To balance the upper limit of current, T slot_i This is the duration of a single equalization cycle.

[0101] This embodiment elevates the equilibrium decision-making process from a simple threshold judgment to a refined multi-objective optimization process that balances benefits and risks within a safe boundary, thereby achieving true differentiated and adaptive control.

[0102] In a specific embodiment, assuming that during a certain scheduling cycle, the system obtains the following set of stable, preprocessed battery state and environmental characteristics for cell i: filtered surface temperature (T surf Temperature: 25.0℃; Filtered relative humidity (RH): 85% (i.e., the value of 0.85 is used for calculation); Aligned battery pack current (I pack -2.0A (indicating the battery pack is in a low-current discharge state); core temperature estimate (T) obtained from the online thermal model. core ): 26.5℃; Initial voltage difference (ΔV) between the single cell and the average voltage of the battery pack. initial ): 25mV; Overall channel confidence label: 0.95 (indicating good data quality); Identified thermal model parameters for this unit: equivalent thermal resistance R th =1.5K / W, equivalent heat capacity C th =800 J / K. Magnus formula constants: a = 17.62, b = 243.12. To ensure calculation robustness, a lower limit clamp is applied to the relative humidity: RH calc =max(0.85, 0.01) = 0.85. Calculate the intermediate variable γ: γ = (17.62 * 25.0) / (243.12 + 25.0) + ln(0.85) = 1.6375 - 0.1625 = 1.475. Calculate the dew point temperature T. dew :T dew = (243.12 * 1.475) / (17.62 - 1.475) = 358.6 / 16.145 = 22.21℃. Calculate the difference Gap: Gap = T surf -T dew=25.0 - 22.21 = 2.79℃. A weighted formula is used, and a set of exemplary weights and parameters are set: w1 = 0.2, w2 = 1.0, w3 = 0.5, w4 = 2.0, representing the Gap threshold related to condensation risk. threshold =1.0℃. Microenvironmental risk index MRI = w1*|T core -T surf |+w2*max(0, Gap) threshold -Gap)+w3*RH+w4*(1-Confidence); Substituting the values: MRI=0.2*|26.5-25.0|+1.0*max(0,1.0-2.79)+0.5*0.85+2.0*(1-0.95)=0.2*1.5+1.0*0+0.425+2.0*0.05=0.3+0.425+0.1=0.825. This MRI value (0.825) quantifies the overall risk level of the current individual.

[0103] Assume the system calculates the core temperature change rate |dT within the current time window. core / dt|=0.03℃ / s, core-surface temperature difference |T core -T surf |=1.5℃, current amplitude|I pack |=2.0A. Compare with the set stability thresholds: 0.03℃ / s≤0.1℃ / s (satisfied), 1.5℃≤2.5℃ (satisfied), 2.0A≤C / 30 (assuming C / 30 is 5A, satisfied). Since all conditions are met, and assuming the satisfied state has lasted longer than the minimum dwell time, the state machine enters an equilibrium state. Thermal steady-state flag. TLS =True. The current real-time Gap value is 2.79℃, far exceeding the set trigger threshold (e.g., 1.0℃). Assuming that environmental trend predictions indicate no risk of a rapid decrease in Gap in the near future, the condensation risk warning is false. Dew point crossing flag. DCI =False. Final permission decision due to Flag. TLS True and Flag DCI If the result is false, the equilibrium permission state of individual unit i is determined to be true. This unit can then proceed to the subsequent equilibrium parameter optimization stage.

[0104] Suppose that individual i passes the conflict arbitration and becomes a member of the post-arbitration candidate set. The decision variable is the single equilibrium duration T. i (seconds) and upper limit of equalization current I i (Ampere). The objective function is: min(ΔV) initial_i -K i *I i *T i / 3600)2 +λ*MRI i *(I i *T i / 3600). Wherein, ΔV initial_i =0.025V; Assuming voltage-electricity coefficient K i =50V / Ah; Risk weight λ=0.5; MRI i =0.825. The objective function is specified as: min(0.025-(50 / 3600)*I i *T i ) 2 +0.5*0.825*(I i *T i / 3600). Hardware constraint: 0≤I i ≤1.0A (assuming maximum hardware current is 1A); 0≤T i ≤600s (assuming a maximum single duration of 600s). Thermal safety constraint: ΔT core_i ≤Θ safe (Let Θ) safe =5℃). The temperature rise can be approximated by ΔT. core_i ≈R th *(k heat *|I pack |+I i 2 *R internal For simplification, another linearized form ΔT is used here. core_i ≈C heat *(I i *T i / 3600), C heat Let I be the dose-temperature rise coefficient, assumed to be 12℃ / Ah. The constraint then becomes: 12*(I i *T i / 3600)≤5, that is, I i *T i ≤1500 A·s. Dose budget constraint: Assume the total dose budget for this cycle is E. budget =0.2Ah = 720A·s. The constraint is I. i *T i ≤720. The above includes the objective function and constraints (0≤I). i ≤1.0, 0≤T i ≤600, I i *T i The optimization problem (≤720) is handled by a solver. The solver, after weighing the trade-off between maximizing the reduction of the 25mV voltage difference and minimizing the penalty imposed by the risk exponent of 0.825, obtains a set of optimal solutions. One possible solution is: I i =0.8A,T i=900s. But at this time T i This exceeds the maximum duration constraint for a single dose. Considering all constraints, especially the dose budget constraint I... i *T i ≤720, a more realistic optimal solution might be: final equilibrium parameter: upper limit of equilibrium current I i =1.0A, single equilibrium duration T i =720s. However, this duration exceeds the maximum duration for a single test, so the final solution will fall on the boundary, for example: the upper limit of the equilibrium current I. i =1.0A, single equilibrium duration T i =600s. At this point, the total dose is 600 A·s, satisfying all constraints.

[0105] According to one aspect of this application, the thermal hysteresis steady-state gating step further includes: introducing a graded timer and a dynamic hysteresis threshold to further improve the stability of the state transition.

[0106] In this embodiment, when constructing a three-state finite state machine, the three states are specifically defined as thermal instability, thermal stabilization, and equilibrium. Thermal instability indicates that the battery is in a thermal dynamic process, where there is a significant gradient between the core temperature and the surface temperature; thermal stabilization is a transitional state, indicating that the thermal disturbance has weakened but stability still needs to be confirmed; equilibrium indicates that the battery has been confirmed to have entered a thermally stable state. Specifically, the state transition adopts a graded timing mechanism. The transition from thermal instability to thermal stabilization requires that the thermal stability tendency condition be met for a duration exceeding τ. quiet , where τ quiet This represents the initial settling time, ranging from 30 to 60 seconds. The transition from thermally stable to potentially equilibrium requires conditions to continue satisfying a longer settling time τ. gap , τ gap The value range is 60-180 seconds. This tiered design avoids misjudgments caused by brief periods of stability. Regarding hysteresis threshold settings, the system sets differentiated entry and exit thresholds for the core and surface temperature differences. For example, the requirement for transitioning from a thermally stable state to an equilibrium state is |T|. core -T surf |≤1.5℃, while the triggering condition for the return to thermal stability from equilibrium is |T core -T surf The threshold for the core temperature change rate is 2.0℃. Similarly, the entry threshold is set to 0.02℃ / s, and the exit threshold is set to 0.05℃ / s. This hysteresis band design creates a temperature difference buffer of 0.5℃ and a rate buffer of 0.03℃ / s.

[0107] Optionally, the system can also dynamically adjust the width of the hysteresis band according to the historical operation data of the battery. When it is detected that the battery pack is in a complex working condition of frequent charging and discharging, the hysteresis band is automatically expanded to 0.8 - 1.0 °C; in a relatively stable floating charge working condition, it can be appropriately narrowed to 0.3 - 0.4 °C to improve the response speed while ensuring stability.

[0108] According to another aspect of the present application, the dew point crossing event gating can also be: strengthening the condensation risk warning mechanism and introducing parametric designs of trend consistency judgment and drying timer.

[0109] In this embodiment, when generating a condensation risk warning, the system adopts a trend consistency judgment rule. The specific implementation is: dT surf / dt < -ε T and dRH / dt > ε RH and Gap pred ≤ Δ gap ; where ε T is the cooling rate threshold, with a value of 0.01 - 0.03 °C / s; ε RH is the humidification rate threshold, with a value of 0.1 - 0.3 % / s; Δ g ap is the warning threshold, with a value of 0.2 - 1.0 °C. Only when all three conditions are met simultaneously is it determined that there is a high condensation risk. This combined judgment avoids false triggering of a single indicator. The initial value of the drying timer timer dry is dynamically set according to the ambient humidity. When the relative humidity RH > 90%, timer dry is set to 30 minutes; when 70% < RH ≤ 90%, it is set to 20 minutes; when RH ≤ 70%, it is set to 10 minutes. This hierarchical design ensures sufficient drying time in a high-humidity environment.

[0110] Furthermore, the system implements a recovery mechanism for the dual hysteresis characteristics. The first hysteresis is reflected in the Gap threshold: the trigger threshold Δ g ap is set to 0.5 °C, and the release threshold Δ r ec is set to 1.5 °C, forming a safety margin of 1.0 °C. The second hysteresis is reflected in the humidity condition: it is considered that the humidity fall condition is met only when the relative humidity drops more than 10% from the peak. For example, if the humidity once reached 85%, it needs to drop below 75% to meet the recovery condition. Optionally, in application scenarios with seasonal alternation or large diurnal temperature differences, the system can introduce a time period factor to correct the threshold. During the high condensation risk period from 2 to 6 am, the trigger threshold Δ g ap is automatically increased by 0.2 - 0.3 °C; during the noon period, it can be appropriately relaxed by 0.1 - 0.2 °C.

[0111] According to another aspect of this application, the steps for solving the constrained optimization problem can also be: refining the optimization solution of the risk budget adaptive allocation, and clarifying the construction of the objective function and the selection of the solution algorithm.

[0112] For example, the objective function is constructed in minimization form: min J = Σ i [(ΔV i after ) 2 ]+λ·Σ i [MRI i Dose i ]; where ΔV i after =ΔV i before -K i ·I bal_i ·T slot_i / 3600 represents the expected voltage difference after equalization; K i ΔV is the voltage-to-charge conversion factor for the i-th cell, in V / Ah, with a value ranging from 40 to 60. i before The expected voltage difference before equalization; Dose i =I bal_i ·T slot_i / 3600 is the equilibrium dose, in Ah; λ is the risk trade-off coefficient, ranging from 0.1 to 5.0; MRI i Let λ be the microenvironmental risk index for the i-th cell. Optionally, the risk trade-off coefficient λ can be dynamically adjusted based on the application scenario. In applications with extremely high reliability requirements, such as data centers, λ is set to a larger value (3.0-5.0) to prioritize safety; in applications with high balancing efficiency requirements, such as power tools, λ is set to a smaller value (0.1-0.5) to pursue rapid balancing. The system can also dynamically adjust the value of λ based on the state of health (SOH) of the battery pack; the lower the SOH, the larger the value of λ. Regarding the choice of solution algorithm, when the number of candidate cells does not exceed 8, the quadratic programming (QP) algorithm is used, employing the interior-point method or the effective set method to complete an accurate solution within 200ms. When the number of candidate cells exceeds 8 or computational resources are limited, the gradient projection iterative algorithm is switched to obtain an approximate optimal solution through 10-20 iterations, with the time of each iteration controlled within 10ms.

[0113] Furthermore, the system linearizes the temperature rise constraint to facilitate the solution. The nonlinear temperature rise model ΔT... core_i =f(I bal_i T slot_i Perform a Taylor expansion near the operating point, retaining the first-order terms: ΔT core_i ≈α i ·I bal_i +βi ·T slot_i ;where α i =R th_i ·R internal_i ,β i =k heat_i ·I bal_i 0 / C th_i I bal_i 0 R is the current value at the expansion point. th_i Let R be the thermal resistance of the i-th cell from its core to the environment. internal_i Let k be the resistance along the current flow path of the i-th battery cell. heat_i Let C be the heat generation rate of the i-th cell under unit current. th_i Let I be the heat capacity of the i-th battery cell. bal_i T represents the equalization current value allocated to the i-th battery cell in the current scheduling cycle. slot_i The equalization operation duration allocated to the i-th battery cell in the current cycle. Optionally, when the system detects an MRI of a certain cell... i When the current is greater than 2.0, its equalization current upper limit I can be directly set. balmax_i Limit to 30% of the maximum hardware value, balancing time T slot_i The time limit is set to 300 seconds, thus implementing risk control before entering the optimization solution.

[0114] According to another aspect of this application, the data acquisition and preprocessing steps can also be: to further implement the data repair and environmental status label generation mechanism.

[0115] Specifically, in terms of data repair, the system uses a bounded spline interpolation algorithm to handle missing data. The specific implementation is: S(t)=Σ k [a k ·φ k (t)],subject to:L min ≤S(t)≤L max Where S(t) is the interpolation function; φ k (t) is a B-spline basis function; a k L represents the spline coefficients to be solved; min and L max For physical boundary constraints. For example, for a relative humidity signal, L min =0%, L max =100%; for surface temperature, L min = -40℃, L max =85℃. Solve for the coefficient a using the least squares method. kSimultaneously satisfying boundary constraints. Environment state labels are generated via a rule engine. The system predefines the following rule set: if dT surf When dt < -0.05℃ / s and dRH / dt > 0.2% / s, generate a label for rapid cooling and humidification; if |dT surf If dT / dt| < 0.01℃ / s and |dRH / dt| < 0.05% / s, the label generation environment is stable; if dT surf When / dt>0.1℃ / s, a tag for rapid heating is generated. Each tag is associated with a set of parameter correction coefficients; for example, a rapid cooling and humidification tag will extend the time window for thermal stability determination by 1.5 times. For dynamic confidence assessment, the system uses a Kalman filter innovation sequence to evaluate signal quality. Innovation r k =y k -H·x k |k-1; Information covariance S k =H·P k |k-1·H T +R; where y k This is the measured value; x k |k-1 represents the one-step prediction value; H is the observation matrix; P k |k-1 is the prediction error covariance; R is the measurement noise covariance. T This is a transpose. When the statistical properties of the innovation sequence deviate from white noise, it indicates a decrease in signal quality, and the confidence label value is reduced accordingly. Furthermore, the system implements adaptive filtering parameter adjustment. The real-time estimated noise variance σ is used... noise 2 Mapped to filter time constant: τ filter =τ base ·(1+k adapt ·σ noise / σ ref ); where τ base The reference time constant, with a value ranging from 0.5 to 1.0 seconds; k adapt σ is an adaptive coefficient, ranging from 0.5 to 2.0; ref The standard deviation of the noise level is used as a reference. As noise increases, the filter automatically enhances its smoothing effect. Optionally, the system can maintain a long-term health record for each sensor channel, recording information such as the number of historical failures and drift trends, as a supplementary basis for confidence assessment.

[0116] According to another aspect of this application, the conflict arbitration procedure can also be: a specific implementation of the conflict arbitration mechanism for multi-unit equilibrium. The conflict arbitration employs a three-level constraint checking mechanism. The first level is the bus current constraint: Σ i (I bal_i )≤I bus_max ; where I bus_maxTo balance the maximum bus current, a typical value of 5-10A is required. The second level is for parallel channel constraints: count(active) cells )≤N channel ;where N channel The maximum number of parallel equalization channels supported by the hardware, typically 4-8; active cells This is the set of battery cells selected for balancing operations within the current cycle. `count()` is a function used to count the number of elements in the set. The third level is the thermal margin constraint: Σ i (P bal_i )≤P cooling -P operation ;where P bal_i =I bal_i 2 ·R bal For single-unit equalization power; P cooling P represents the total heat dissipation capacity of the system. operation R is the current operating heat production capacity. bal This represents the resistance value connected between the battery cell and the discharge path during passive balancing. The priority queue is constructed using a composite scoring mechanism. The priority score for each candidate cell is calculated as: Score i =w1·|ΔV i | / V ref +w2·(1 / MRI i )+w3·SOH i ;where |ΔV i | represents the absolute value of the voltage deviation; V ref For reference voltage, take 3.2V; MRI i Microenvironmental risk index; SOH i The health status of a single unit is defined by w1, w2, and w3, which are weighting coefficients satisfying w1 + w2 + w3 = 1. Typical values ​​are w1 = 0.5, w2 = 0.3, and w3 = 0.2, reflecting the relative importance of voltage deviation, risk avoidance, and health status in priority determination, respectively. During arbitration, the system maintains a real-time resource occupancy table. Each time a single unit is admitted to the execution list, the remaining bus current capacity I is immediately updated. remain =I bus_max -Σ(I allocated ); Number of remaining channels N remain =N channel -count(allocated); Remaining heat margin P remain =P cooling -P operation -Σ(P allocated ); where I allocated The allocated equalization current is P, where allocated is the set of allocated individual cells. allocatedThe system allocates equalization power. Arbitration terminates when any remaining resource is insufficient to accommodate the next candidate cell. For cells that fail arbitration, the system generates a diagnostic record: the record format is: Cell i: Equalization delayed due to [specific reason], will be reassessed after [estimated time]. Specific reasons include bus current saturation, insufficient number of channels, limited thermal margin, etc. These records are used for system optimization and fault analysis. Optionally, when certain high-priority cells (e.g., voltage deviation exceeding 50mV) fail arbitration, the system can trigger preemptive scheduling, suspending the currently executing low-priority equalization task to free up resources for high-priority cells.

[0117] In one embodiment of this application, the system establishes a simplified Kalman filter for each sensor channel. For the temperature sensor, the state equation adopts a first-order Markov model: x k =x k-1 +w k The observation equation is: y k =x k +v k ;where x k This represents the actual temperature value; y k The sensor measurement value; w k The process noise has a variance of Q; v k To measure noise, the variance is R. The recursive process of Kalman filtering includes two steps: prediction and update. The prediction step calculates a predicted value x. k|k-1 =x k-1 and prediction error covariance P k|k-1 =P k-1 +Q. Update steps to calculate new information r k =y k -x k|k-1 This is the difference between the measured value and the predicted value. The information covariance S k =P k|k-1 +R reflects the uncertainty in the prediction. Kalman gain K k =P k|k-1 / S k Used to weigh the reliability of predicted and measured values. The final state is updated to x. k =x k|k-1 +K k ×r k Signal quality assessment is based on the statistical properties of the innovation sequence. Ideally, the innovation sequence should be zero-mean white noise. The system calculates the mean μ of the innovation within the sliding window. r and variance σ r 2 When |μ r |>0.1×σ r When σ indicates the presence of a systematic bias; when σ r2 Significant deviation from the theoretical value S k When this occurs, it indicates a model mismatch. Based on these indicators, a quality factor Q between 0 and 1 is generated. factor =exp(-|μ r | / σ r -|log(σ r 2 / S k This quality factor is integrated with other health indicators to form the final channel confidence label. The specific integration formula is: Confidence = 0.6 × Q factor +0.2×(1-Missing Rate )+0.2×(1-Saturation Rate ); among which Missing Rate Data missing rate; Saturation Rate This represents the saturation rate. The weighting coefficient of 0.6 reflects the dominant role of real-time signal quality.

[0118] In some alternative implementations, an extended Kalman filter (EKF) can be used to handle the nonlinear sensor characteristics. For the nonlinear response of the temperature sensor, the observation equation is modified to y k =h(x k )+v k , where h(·) is a nonlinear function. Linearization allows for a more accurate assessment of signal quality.

[0119] In another embodiment of this application, adaptive adjustment of filter parameters based on noise variance is performed to achieve intelligent data preprocessing. Specifically, the real-time noise variance of each sensor channel is estimated. The standard deviation within a sliding window is used for estimation: σ noise 2 = (1 / N)×Σ(y) i –y*) 2 Where N is the window length, typically 20-50 sampling points; y i σ represents the original measured value; y* represents the mean within the window. To improve robustness, the median absolute deviation (MAD) method can be used: σ noise = 1.4826 × median(|y i -median(y)|), this method is insensitive to outliers. Based on the estimated noise variance, the filter parameters are dynamically adjusted. For the low-pass filter, the time constant is adjusted according to the following formula: τ filter =τ base ×(1+k adapt ×σ noise / σ ref ); where τ baseThe reference time constant, with a value ranging from 0.5 to 1.0 seconds; k adapt σ is an adaptive coefficient, ranging from 0.5 to 2.0; ref The reference noise standard deviation, obtained through offline calibration, is typically twice the sensor's nominal accuracy. When σ noise <0.5×σ ref At that time, the signal quality was considered excellent, τ filter =τ base Maintain a fast response. When 0.5×σ ref ≤σ noise ≤2×σ ref When σ increases linearly, the filter strength increases. noise >2×σ ref At that time, τ filter Reaching the upper limit of 3×τ base To suppress noise to the greatest extent possible. For the median filter, the window width W is also adaptively adjusted: W = W base +round(k w ×(σ noise / σ ref -1)); where W base The base window width is 3-5; k w The adjustment coefficient is set to 2-4; `round()` is the rounding function. The window width is limited to the range [3, 11] to avoid excessive smoothing. Furthermore, the system maintains a historical record of parameter adjustments to prevent frequent switching. An update is only performed when the difference between the new and old parameters exceeds 20%, and the minimum interval between two updates is 10 sampling periods. This hysteresis mechanism avoids parameter jitter.

[0120] In another embodiment of this application, the decision variable boundary of the optimization problem is dynamically generated based on the microenvironment risk index. Specifically, the mapping of the voltage difference trigger threshold adopts a three-segment piecewise linear function. When MRI... i When the value is less than 0.5, it indicates a low-risk state, and the voltage difference trigger threshold V is set. th_i =10mV, allowing for a small voltage difference to trigger equalization. When 0.5≤MRI i When V < 1.5, it falls within the medium-risk range. th_i =10 + 10 × (MRI) i -0.5)mV, linearly increasing the trigger threshold. When MRI i A value ≥1.5 indicates a high-risk status. th_i =20mV, requiring a significant voltage difference to trigger equalization. The mapping of the equalization current upper limit uses an inverse piecewise linear function. When MRI i When <0.5, the upper limit of the equalization current I max_i =I hardware_max The maximum hardware capability can be used, of which Ihardware_max This represents the maximum equalization current capability of the hardware. When 0.5 ≤ MRI i When I < 2.0, max_i =I hardware_max ×(1-0.5×(MRI i -0.5) / 1.5), linearly reducing the upper limit of current. When MRI i When ≥2.0, I max_i =0.5×I hardware_max Limited to 50% of hardware capabilities. The mapping of single-cycle equalization duration takes into account the heat accumulation effect. Base duration T base =600 seconds. Adjusted based on MRI: T slot_i =T base ×exp(-0.2×MRI i This exponential decay rapidly shortens the equilibrium time of high-risk cells. A lower limit of 60 seconds is set to avoid frequent start-stop cycles. In some alternative implementations, environmental factors can be introduced to modify the mapping function. For example, in high-temperature environments (T... ambient At temperatures above 35℃, all thresholds and upper limits are multiplied by a correction factor of 0.8. In low-temperature environments (T... ambient At temperatures below 5℃, considering the increased internal resistance of the battery, the upper limit of the current is further reduced by 20%. The parameters of the mapping function can be optimized using machine learning methods. Equalization effect data under different MRI scans are collected, and the segmentation points and slopes are adjusted using least squares or Bayesian optimization to achieve the optimal long-term equalization effect.

[0121] In another embodiment of this application, the nonlinear temperature rise model is linearized to make it applicable to quadratic programming solutions. The temperature rise generated by battery equalization originally follows a nonlinear relationship: ΔT core =f(I bal T slot T ambient (, SOC), where I bal To balance the current, T slot To balance the duration, T ambient Let f() be the ambient temperature, and f be a nonlinear temperature rise prediction function. Directly handling this nonlinearity would drastically increase the complexity of the optimization problem. Therefore, a Taylor expansion is used to linearize the function around the operating point. The operating point (I0, T0) is chosen, typically as the midpoint of the equilibrium parameter or expected value from the previous cycle. The first-order Taylor expansion is: ΔT core ≈ΔT0+(Ψf / ΨI)|0×(I bal -I0)+(Ψf / ΨT)|0×(T slot-T0); where Ψ is the partial derivative and ΔT0 is the operating temperature rise. The partial derivatives are calculated online using the finite difference method. Current partial derivative: (Ψf / ΨI)|0≈(f(I0+ΔI,T0)-f(I0-ΔI,T0)) / (2×ΔI); where the current perturbation step size ΔI is 0.1A. The duration partial derivative is calculated similarly, with ΔT taken as 10 seconds. Considering that the temperature rise is mainly generated by Joule heating for equilibrium, a simplified physical model can be used: ΔT core =R th ×(I bal 2 ×R internal ×T slot / 3600). Here R th For thermal resistance, a typical value is 1.5 K / W; R internal This represents the battery's internal resistance, typically 20-50 mΩ. After linearization: ΔT core ≈α i ×I bal +β i ×T slot ;where α i =2×R th ×R internal ×I0×T0 / 3600;β i =R th ×I0 2 ×R internal / 3600. To ensure linearization accuracy, set the effective range: |I bal -I0| < 0.3 × I0 and |T slot -T0| < 0.3 × T0. If the value is outside the range, reselect the operating point and update the linearization coefficients. Optionally, piecewise linearization can be used to improve accuracy. The decision space is divided into multiple regions, each using independent linearization coefficients. During optimization, piecewise constraints are handled using mixed integer programming (MIP).

[0122] In another optional embodiment, the battery differential balancing control method for a compact power system may further be:

[0123] The original electrical and environmental quantities of the battery are obtained, and based on the original electrical and environmental quantities, the core temperature estimate, the filtered environmental quantities, and the derived environmental features are derived.

[0124] Based on the core temperature estimation, a thermal steady-state label characterizing the thermal stability of the battery is generated;

[0125] Based on the filtered environmental quantities and derived environmental characteristics, a dew point crossing indicator is generated to characterize the risk of battery condensation.

[0126] The equilibrium permitting status is determined by combining the thermal steady-state indicator and the dew point crossing indicator. When the permit is granted, a micro-environmental risk index characterizing the current environmental safety level is constructed based on the core temperature estimate, filtered environmental quantities and derived environmental characteristics.

[0127] When the equilibrium permission state is true, the final equilibrium parameters are adaptively determined by solving the constrained optimization problem that takes into account the micro-environment risk index. The final equilibrium parameters include the voltage difference trigger threshold, the upper limit of the equilibrium current, and the duration of a single equilibrium.

[0128] Optionally, within a preset scheduling period, thermal hysteresis steady-state gating is performed on individual battery cells to generate a thermal steady-state flag; dew point crossing event gating is performed on individual battery cells to generate a dew point crossing flag; if and only if the thermal steady-state flag is true and the dew point crossing flag is false, risk budget-based adaptive allocation is performed based on the micro-environmental risk and electrical state of the individual battery cells to determine differentiated equilibrium parameters. Further, the steps for performing thermal hysteresis steady-state gating include: estimating the core temperature online using measured surface temperature and battery pack current based on a first-order or second-order thermal resistance-thermal capacity model; constructing a state machine containing three states: thermally unstable, thermally stable, and equilibrium-ready; and using the rate of change of the core temperature estimate, the temperature difference between the core temperature estimate and the surface temperature, and the amplitude of the battery pack current as state transition criteria; and setting the thermal steady-state flag to true when the state machine is in an equilibrium-ready state, by combining the state machine with hysteresis and a timer mechanism. Furthermore, the steps for implementing dew point crossing event gating judgment include: calculating the dew point temperature and the difference between the surface temperature and the dew point temperature Gap based on the measured surface temperature and relative humidity; generating a condensation risk warning based on the current value and one-step predicted value of the difference Gap, combined with the changing trends of temperature and humidity; setting the dew point crossing flag to true when the real-time value of the difference Gap is lower than the first threshold or the condensation risk warning is true; and setting the dew point crossing flag to false after the relative humidity drops and the difference Gap recovers to above the second threshold and remains so for a preset time. Furthermore, the steps for implementing risk budget-based adaptive allocation include: constructing an objective function that uses the expected imbalance after equilibrium as a benefit term and the product of the microenvironment risk index and the equilibrium dose as a risk penalty term; under the conditions of satisfying the hard constraint of condensation and the thermal safety constraint, solving the objective function with the voltage difference trigger threshold, the upper limit of the equilibrium current, and the duration of a single equilibrium cycle for each individual unit as decision variables to obtain differentiated equilibrium parameters. For example, the microenvironment risk index is obtained by weighted synthesis based on the surface temperature of the battery cell, relative humidity, the gap between surface temperature and dew point temperature, and core temperature estimation.

[0129] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for differentiated balancing control of batteries in a compact power system, characterized in that, include: Obtain the original electrical and environmental quantities of the battery, and determine the battery state and environmental characteristic quantities that characterize the battery's operating state and microenvironmental properties accordingly. Based on battery state and environmental characteristics, generate thermal steady-state indicators and dew point crossing indicators; The combined thermal steady-state indicator and dew point crossing indicator are used to determine the equilibrium permissible state. When the equilibrium permitting state is true, the final equilibrium parameters are determined by solving a constrained optimization problem that takes into account micro-environmental risks, based on the battery state and environmental characteristics.

2. The method according to claim 1, characterized in that, The final equilibrium parameters are determined, including: A microenvironment risk index is constructed based on battery state and environmental characteristics. An optimization objective function is established, which has the joint objective of minimizing the proxy quantity of the imbalance after equilibrium and minimizing the risk penalty term composed of the product of the microenvironment risk index and the equilibrium dose. Apply constraints, which include at least: a hard constraint that forces the equalization current to zero based on the state of the dew point crossing mark, and a thermal safety constraint that ensures the upper limit of the temperature rise caused by the equalization dose does not exceed a preset value. Solve for the optimal solution to the objective function under constraints to obtain the final equilibrium parameters.

3. The method according to claim 2, characterized in that, Establish the optimization objective function, including: Quantify the expected equalization benefit brought about by the equalization dose, wherein the expected equalization benefit is used to characterize the expected reduction in battery imbalance after equalization. Assess the risk cost arising from a balanced dose in a predetermined microenvironment, wherein the risk cost is determined by multiplying the microenvironment risk index, the balanced dose, and a preset risk weighting coefficient; Combine expected equilibrium returns with risk costs to construct an optimal objective function.

4. The method according to claim 2, characterized in that, Apply constraints, including: Establish a hard constraint on condensation, which stipulates that when the dew point crossing flag of any cell is true, the equalization current decision variable allocated to that cell is always zero. A thermal safety constraint is established, which maps the equalization dose to the upper limit of the temperature rise of the battery core through a pre-configured online thermal model, and requires that the upper limit of the temperature rise must not exceed the preset thermal safety threshold.

5. The method according to claim 1, characterized in that, Generate dew point crossing markers, including: The difference between surface temperature and dew point temperature (Gap) and the trend of environmental change are determined from battery state and environmental characteristics. Based on the gap difference and environmental change trends, the risk of condensation is predicted and a condensation risk warning is generated. When the condensation risk warning is true, or the real-time value of the gap is lower than the preset trigger threshold, the dew point crossing flag is set and the drying timer is started. Once the gap value returns to the preset release threshold and the relative humidity drops and continues until the drying timer expires, the dew point crossing flag is cleared.

6. The method according to claim 5, characterized in that, Generate condensation risk warnings, including: By applying a pre-configured short-term forecasting model and estimating the predicted value of the gap at the next time step based on the historical sequence of the gap, the model is used. The trends in temperature and humidity were analyzed from the trends in environmental change. When the predicted difference Gap value is lower than the preset condensation warning threshold, and the temperature change trend is downward while the humidity change trend is upward, a condensation risk warning is generated.

7. The method according to claim 1, characterized in that, Generate thermal steady-state indicators, including: Construct a finite state machine that includes at least three states: thermal instability, thermal stabilization, and equilibrium, to model the thermal stabilization process of the battery. Based on battery state and environmental characteristics, determine whether the thermal stability trend condition is met; Based on whether the thermal stability tendency condition is satisfied and its duration, transitions are performed between states of the finite state machine; The thermal steady-state flag is set if and only if the finite state machine is in an equilibrium state; otherwise, the thermal steady-state flag is cleared.

8. The method according to claim 7, characterized in that, Determining whether the thermal stability tendency condition is met includes: Within a preset time window, the core temperature estimate, the filtered surface temperature, and the aligned battery pack current are extracted from the battery status and environmental characteristics. The joint review will consider at least three criteria: the rate of change of the core temperature estimate, the temperature difference between the core temperature estimate and the filtered surface temperature, and the amplitude of the aligned battery pack current. The thermal stability tendency condition is considered satisfied if and only if the criteria simultaneously satisfy their respective stability thresholds within the time window.

9. The method according to claim 7, characterized in that, Performing transitions between states in a finite state machine includes: Configure a hysteresis threshold for state transitions so that the thermal stability tendency condition required to enter a more stable state is stricter than the condition for exiting that stable state. A minimum dwell time is set for state transitions, stipulating that a transition from the current state to the next state is only permitted if the thermal stability trend condition is met for an extended period of time.

10. The method according to claim 2, characterized in that, Construct a microenvironment risk index, including: The filtered surface temperature, filtered relative humidity, the difference between surface temperature and dew point Gap, and core temperature estimation are extracted from battery state and environmental characteristics. The resolved surface temperature, relative humidity, gap difference, and core temperature estimate are weighted and synthesized to generate a microenvironment risk index.

Images