Intelligent management system for lithium battery pack based on multi-modal sensing

Abstract

The present application relates to the technical field of battery management, in particular to a lithium battery pack intelligent management system based on multi-modal sensing. It comprises: a multi-modal sensing and fusion unit that collects battery pack operating environment and internal state data in real time, and fuses the operating environment and internal state data through weighted average to generate fused multi-modal data; an intelligent management master unit that predicts battery health state calculation based on cumulative gas production life mapping model, and predicts thermal runaway risk through a long short-term memory network model to generate control instructions, state information and alarm signals; a power execution unit that performs overcharge and overdischarge protection operations on the battery pack based on the control instructions. The present application collects multi-dimensional data such as voltage, current, temperature and gas concentration through a multi-modal sensing module, and extracts key features such as voltage consistency, temperature distribution and current fluctuation by combining statistical analysis and time-frequency transformation methods, thereby achieving comprehensive perception of the battery pack operating state.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and more specifically, to an intelligent management system for lithium battery packs based on multimodal sensing. Background Technology

[0002] With the widespread application of lithium batteries in electric vehicles, energy storage power stations, and smart grids, the safety, reliability, and lifespan management of battery packs have increasingly become critical bottlenecks in system operation. Traditional battery management systems (BMS) mainly rely on basic electrical parameters such as voltage, current, and temperature for state monitoring. This single-dimensional sensing approach fails to comprehensively reflect the electrochemical aging process and potential fault characteristics within the battery, particularly lacking effective monitoring methods for irreversible side reactions such as electrolyte decomposition, gas generation, and micro-internal short circuits, resulting in insufficient accuracy in state assessment. Regarding state of health (SOH) and remaining usable life (RUL) prediction, existing methods are mostly based on empirical models or purely data-driven algorithms, lacking physical mechanism support, exhibiting poor generalization ability, and failing to adapt to complex and variable operating conditions and individual differences, leading to low reliability of prediction results. Simultaneously, safety warning mechanisms generally employ static threshold criteria, resulting in delayed responses and an inability to achieve early identification of serious faults such as thermal runaway, leading to high false alarm rates and significant missed alarm risks. Furthermore, traditional control strategies are mostly triggered by fixed rules, lacking the ability to dynamically adjust based on battery health status and risk level, making it difficult to reconcile the contradictions between safety protection, lifespan extension, and operational efficiency. At the system integration level, existing battery management systems lack communication compatibility with field monitoring equipment (such as DTUs and SCADA systems), have weak alarm linkage capabilities, and struggle to meet the needs of intelligent and remote operation and maintenance. Therefore, a lithium battery pack intelligent management system based on multimodal sensing is designed. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent management system for lithium battery packs based on multimodal sensing, so as to solve the problems of traditional battery management systems mentioned in the background art, such as single sensing dimension, low life prediction accuracy, delayed safety warning, rigid control strategy and insufficient system integration.

[0004] To achieve the above objectives, the present invention aims to provide an intelligent management system for lithium battery packs based on multimodal sensing, comprising:

[0005] A multimodal sensing and fusion unit collects real-time data on the battery pack's operating environment and internal state, and fuses the operating environment and internal state data through a weighted average to generate fused multimodal data.

[0006] The intelligent management control unit predicts the battery health status based on the cumulative gas production life mapping model, and predicts the thermal runaway risk through the long short-term memory network model, generating control commands, status information and alarm signals.

[0007] The power execution unit performs overcharge and over-discharge protection operations on the battery pack based on control commands, and performs energy transfer between battery cells according to an intelligent balancing strategy.

[0008] The communication and interaction unit uploads status information and alarm signals to the field DTU equipment, and displays and interacts with the lithium battery pack data in real time through the remote monitoring platform.

[0009] As a further improvement to this technical solution, the multimodal sensing and fusion unit includes a multimodal sensing module, a multimodal fusion module, and a multimodal feature extraction module;

[0010] The multimodal sensing module collects operating environment and internal state data, and preprocesses the collected environmental and internal state data. The operating environment and internal state data include total voltage, single-core voltage, charging and discharging current, temperature, gas concentration, and historical key data.

[0011] The multimodal fusion module performs time-series alignment and normalization on the preprocessed environmental and internal state data, and then fuses them by weighted averaging to generate fused multimodal data.

[0012] The multimodal feature extraction module extracts multimodal features from the fused multimodal data through statistical analysis and time-frequency transformation methods.

[0013] As a further improvement to this technical solution, the intelligent management main control unit includes a state calculation and life prediction module, a risk prediction and early warning module, and a control strategy and decision output module.

[0014] The state calculation and lifetime prediction module calculates the state of charge, health status and capacity decay trend of the lithium battery pack in real time based on the fused multimodal data and the equivalent circuit model, and predicts the remaining usable lifetime and remaining cycle number of the battery based on historical key data.

[0015] The risk prediction and early warning module monitors potential dangerous states through a long short-term memory network model and triggers an early warning mechanism to generate an alarm signal when the risk approaches a preset safety boundary.

[0016] The control strategy and decision output module dynamically generates the battery pack's operation control strategy based on the lifetime prediction and risk assessment results, and receives feedback signals from the power execution unit. It integrates the lithium battery pack's state of charge, health status, capacity decay trend, battery cycle life and remaining usable life, and feedback signals into status information.

[0017] As a further improvement to this technical solution, the state calculation and lifetime prediction module performs real-time calculations on the state of charge, state of health, and capacity decay trend of the lithium battery pack, and predicts the remaining usable lifetime and remaining cycle count of the battery, including the following steps:

[0018] S1.1 Preprocess the fused multimodal data;

[0019] S1.2 Based on the equivalent circuit model, the state of charge of the battery is calculated in real time; combined with temperature and current information, the capacity utilization rate and voltage plateau characteristics are corrected to obtain the preliminary health status.

[0020] S1.3. The battery state of charge, health status and multimodal features are fused to form a time-series input feature sequence, which is then input into a long short-term memory network model to perform nonlinear correction on the battery state of charge, health status results and multimodal features, and generate the corrected battery state of charge and health status.

[0021] S1.4 Extract key degradation features based on historical key data to determine the current battery capacity decay rate;

[0022] S1.5 Based on the capacity decay rate, combined with historical operating condition data and key degradation characteristics, the cumulative gas production life mapping model is used to predict the cycle life and remaining usable life of the battery pack.

[0023] As a further improvement to this technical solution, in step S1.5, the cycle life and remaining usable life of the battery pack are predicted using a cumulative gas generation life mapping model, including the following steps:

[0024] S1.51. Calculate the gas generation rate based on real-time gas concentration data from a multimodal sensing unit. ;

[0025] S1.52 Calculate the effective cumulative gas production based on the effective cumulative gas production. ;

[0026] S1.53. Based on the effective cumulative gas production, a preliminary reference mapping relationship between the effective cumulative gas production and the battery capacity retention rate is established. At the same time, an adaptively adjustable parameter set is constructed. The cumulative gas production lifetime mapping model is used to calculate the health status.

[0027] S1.54. Compare the health status with the health status corrected in step S1.3 to construct a dual-redundant health assessment system;

[0028] S1.55. Set the capacity retention rate threshold corresponding to the end of battery life, which is used to determine the threshold point when the battery enters its battery life.

[0029] S1.56, Based on effective cumulative gas production Historical data predicts remaining usable lifespan ;

[0030] S1.57, Prediction results based on remaining usable lifetime Analyze key historical data and calculate the average number of equivalent full cycles completed. And predict the remaining number of cycles.

[0031] As a further improvement to this technical solution, the risk prediction and early warning module monitors potential dangerous conditions and triggers an early warning mechanism when the risk approaches a preset safety boundary, including the following steps:

[0032] S2.1 Compare the multimodal raw data with the preset threshold. If any multimodal raw data exceeds the upper limit threshold b, it is marked as a potential risk event.

[0033] S2.2 Output a high-frequency micro-current excitation signal to the battery pack, monitor the voltage response of the battery pack, calculate the high-frequency impedance spectrum of the battery pack, and fuse the deviation characteristics and multi-mode characteristics in the high-frequency impedance spectrum to generate enhanced multi-mode characteristics.

[0034] S2.3. Input the enhanced multimodal features into the long short-term memory network model to predict the trends of thermal runaway, overcharging, and over-discharging risks, and assess the risk probability within a future period T.

[0035] S2.4. Based on the risk prediction results and risk probability, classify the risk levels and dynamically adjust the early warning trigger conditions;

[0036] S2.5 When the risk probability reaches the preset threshold v, the early warning mechanism is triggered, a corresponding alarm signal is generated, and the alarm signal is transmitted to the communication and interaction unit.

[0037] As a further improvement to this technical solution, in step S2.2, the specific steps involved in generating enhanced multimodal features are as follows:

[0038] During the resting period of the battery pack's charging and discharging, a set of micro-amplitude AC current excitation signals with a specific frequency range are applied to the battery through the equalization circuit of the power execution unit, and the voltage response signals of each battery cell are collected simultaneously. Based on the micro-amplitude AC current excitation signals and voltage response signals, the AC impedance of each cell at multiple characteristic frequency points is calculated, and a high-frequency impedance spectrum of each cell is constructed based on the AC impedance. The deviation characteristics between the high-frequency impedance spectrum and the reference impedance spectrum of healthy cells are extracted, and the deviation characteristics are fused with multi-modal characteristics to form enhanced multi-modal characteristics.

[0039] As a further improvement to this technical solution, the control strategy and decision output module dynamically generates the battery pack's operation control strategy based on lifespan prediction and risk assessment results, including the following steps:

[0040] S3.1 Based on the battery health status and risk level, call the preset control strategy library and generate a dynamic control strategy in combination with the actual working conditions;

[0041] S3.2. The generated dynamic control strategy is converted into control commands and sent to the power execution unit (3).

[0042] S3.3 Receive the execution status and running results returned by the power supply execution unit, compare them with the original control commands, and use them to update the dynamic control strategy.

[0043] As a further improvement to this technical solution, the power supply execution unit includes a protection control module and a balancing control module;

[0044] The protection control module receives and parses the control commands output by the control strategy and decision output module, performs overcharge and over-discharge protection operations on the battery pack, and restricts the charging and discharging circuit under abnormal conditions.

[0045] The equalization control module compares the cell voltages of the battery pack according to the intelligent equalization strategy, identifies high and low voltage cells, and balances the energy between cells through resistance dissipation.

[0046] As a further improvement to this technical solution, the communication and interaction unit uploads the state of charge, health status, capacity decay trend, remaining usable life and remaining cycle count of the lithium battery pack to the field DTU device through the communication interface. At the same time, it uploads the alarm signal to the alarm signal port of the original charging module of the field DTU through the dry contact, and the DTU forwards it to the power distribution master station backend for real-time operation monitoring and alarm linkage.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0048] 1. The intelligent management system for lithium battery packs based on multimodal sensing, as disclosed in this invention, collects multidimensional data such as voltage, current, temperature, and gas concentration through a multimodal sensing module. It then extracts key features such as voltage consistency, temperature distribution, and current fluctuations using statistical analysis and time-frequency transformation methods, achieving comprehensive perception of the battery pack's operating status. Furthermore, it introduces a dual-redundancy health assessment mechanism combining a cumulative gas production lifetime mapping model and a Long Short-Term Memory (LSTM) network. Irreversible gas production is used as the core characterization of battery chemical aging, constructing a physically interpretable lifetime decay model. The LSTM is then used to dynamically learn and correct the nonlinear degradation trend. Simultaneously, adaptive filtering technology is employed to optimize the model parameters online, ensuring that the prediction of State of Health (SOH) and Remaining Usable Life (RUL) not only has mechanistic support but also data-driven flexibility, significantly improving the accuracy of lifetime prediction under complex operating conditions and the long-term reliability of the system.

[0049] 2. The intelligent management system for lithium battery packs based on multimodal sensing involved in this invention actively detects changes in the electrochemical impedance spectrum inside the battery cells during periods of rest or intermittent operation, extracts deviation features such as amplitude shift and phase drift, and fuses multimodal data to form an enhanced feature input LSTM model, enabling trend prediction of potential risks such as thermal runaway and overcharging. Combined with a dynamic risk grading mechanism and an adjustable early warning threshold strategy, the alarm sensitivity is adaptively adjusted according to the rate of change of risk probability, effectively avoiding false alarms and missed alarms. Based on this, the control strategy module generates dynamic control commands including current limiting, heat dissipation adjustment, and equalization management according to the risk level and lifespan status, forming a closed-loop control system of "sensing—evaluation—decision-execution—feedback," achieving an upgrade from passive protection to proactive prevention and control in safety management, significantly improving the operational safety of lithium battery packs under high load and extreme environments. Attached Figure Description

[0050] Figure 1 This is an overall flowchart of the present invention;

[0051] Figure 2 This is a schematic diagram of the battery pack state at the start of the high-temperature float charge test in this embodiment;

[0052] Figure 3 This is a schematic diagram of the battery pack status after 72 hours of high-temperature float charging test in this embodiment;

[0053] Figure 4 This is a schematic diagram of the battery pack status during the high-temperature discharge test in this embodiment. Figure 1 ;

[0054] Figure 5 This is a schematic diagram of the battery pack status during the high-temperature discharge test in this embodiment. Figure 2 ;

[0055] Figure 6 This is a schematic diagram of the battery pack status during the high-temperature discharge test in this embodiment. Figure 3 ;

[0056] Figure 7 This is a schematic diagram of the battery pack status during the high-temperature discharge test in this embodiment. Figure 4 ;

[0057] Figure 8 This is a schematic diagram of the battery pack status during the low-temperature environment startup test in this embodiment;

[0058] Figure 9 This is a schematic diagram of the battery pack status during the low-temperature environment operation test in this embodiment;

[0059] The meanings of the labels in the diagram are as follows:

[0060] 1. Multimodal sensing and fusion unit; 2. Intelligent management main control unit; 21. State calculation and life prediction module; 22. Risk prediction and early warning module; 23. Control strategy and decision output module; 3. Power execution unit; 4. Communication and interaction unit. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example: Please refer to Figure 1-9 As shown, a lithium battery pack intelligent management system based on multimodal sensing is provided, including:

[0063] The multimodal sensing and fusion unit 1 collects real-time data on the battery pack's operating environment and internal state, and fuses the operating environment and internal state data through weighted averaging to generate fused multimodal data.

[0064] In this embodiment, the multimodal sensing and fusion unit 1 includes a multimodal sensing module, a multimodal fusion module, and a multimodal feature extraction module;

[0065] The multimodal sensing module collects operating environment and internal state data, and preprocesses the collected environmental and internal state data. The operating environment and internal state data include total voltage, single cell voltage, charging and discharging current, temperature, gas concentration, and historical key data (including total battery voltage, battery charging and discharging current, and temperature).

[0066] The multimodal fusion module performs time-series alignment and normalization on the preprocessed environmental and internal state data, and then fuses them by weighted averaging to generate fused multimodal data.

[0067] The multimodal feature extraction module extracts multimodal features from the fused multimodal data through statistical analysis and time-frequency transformation methods. These features include voltage consistency indices (single-cell voltage variance, extreme value difference), temperature distribution characteristics (maximum temperature difference, temperature gradient), charge-discharge cycle count, and current fluctuation characteristics. Specifically, firstly, statistical analysis is performed on multi-dimensional time-series signals such as voltage, current, temperature, and gas concentration, including calculating the mean, variance, extreme value difference, and correlation coefficient, to quantify the consistency between cells and the overall operational fluctuation characteristics. Subsequently, time-frequency transformation methods (such as Short-Time Fourier Transform (STFT) or Wavelet Transform) are used to decompose the non-stationary signal into the time-frequency domain, extracting dynamic features such as energy distribution, dominant frequency components, and instantaneous frequency drift at different frequency bands. By combining the above statistical features with time-frequency features, the voltage consistency indices (single-cell voltage variance, extreme value difference), temperature distribution characteristics (maximum temperature difference, temperature gradient), charge-discharge cycle count, and current fluctuation characteristics are further obtained, thus forming multimodal features that can characterize the multimodal operating state of the battery pack.

[0068] The intelligent management main control unit 2 predicts the battery health status based on the cumulative gas production life mapping model, and predicts the thermal runaway risk through the long short-term memory network model, generating control commands, status information and alarm signals.

[0069] In this embodiment, the intelligent management main control unit 2 includes a state calculation and life prediction module 21, a risk prediction and early warning module 22, and a control strategy and decision output module 23.

[0070] The state calculation and lifetime prediction module 21 calculates the state of charge, health status and capacity decay trend of the lithium battery pack in real time based on the fused multimodal data and the equivalent circuit model, and predicts the remaining usable lifetime and remaining cycle number of the battery based on historical key data.

[0071] Furthermore, the state calculation and lifetime prediction module 21 performs real-time calculations on the state of charge, state of health, and capacity decay trend of the lithium battery pack, and predicts the remaining usable lifetime and remaining cycle count of the battery, including the following steps:

[0072] S1.1 Preprocess the fused multimodal data;

[0073] S1.2. Based on the equivalent circuit model, the battery state of charge (SOC) is calculated in real time. Combining temperature and current information, the capacity utilization rate and voltage plateau characteristics are corrected to obtain a preliminary health state. Specifically, the fused voltage, current, and temperature data are first input into the equivalent circuit model (Thevenin model). The battery SOC is estimated in real time through model parameter identification and current integration. On this basis, the model results are corrected by combining temperature and current information: according to the battery operating temperature and the experimentally calibrated capacity-temperature characteristic curve, the rated capacity is weighted and corrected by a temperature factor to reflect the effects of accelerated decay of active materials at high temperatures and limited ion diffusion at low temperatures leading to a decrease in effective capacity. At the same time, the real-time collected charge and discharge current and current-voltage polarization model are used to correct the terminal voltage output of the model to compensate for the voltage polarization and hysteresis effects caused by high current charge and discharge, making the corrected voltage plateau closer to the actual open-circuit voltage level. Finally, the corrected capacity utilization rate and voltage characteristics are fused on the basis of SOC to obtain a preliminary health state of the battery pack that is closer to the actual operating state.

[0074] S1.3. The battery state of charge (SOC), state of health (SOH), and multimodal features are fused to form a time-series input feature sequence, which is then input into a Long Short-Term Memory (LSTM) network model. This model performs nonlinear correction on the battery SOC, SOH, and multimodal features to improve the accuracy of state calculation and generate corrected battery SOC and SOH. The LSM model includes an input layer, multiple hidden layers, and an output layer. The input layer receives the time-series feature vector and performs dimension mapping. The hidden layers consist of several memory units with forget gates, input gates, and output gates to capture the long-term dependencies and short-term dynamic changes of the battery's operating state. After nonlinear processing by the hidden layers, the output layer generates the correction results, specifically including the corrected outputs on the nonlinear coupling relationships between the battery SOC, SOH, and multimodal features, thereby improving the accuracy and robustness of state estimation.

[0075] S1.4. Based on historical key data (long-term battery pack operation data), extract key degradation features (including internal resistance growth rate, voltage hysteresis characteristics, and charge / discharge efficiency decay) to determine the current battery capacity decay rate. Specifically: First, clean and time-align the historical key data (including voltage, current, temperature, charge / discharge capacity, etc.) recorded during the long-term operation of the battery pack to ensure data integrity and comparability; then, calculate the internal resistance growth rate based on the curve of internal resistance change with cycle number or time to reflect the degradation trend of polarization and conductivity; next, extract the voltage hysteresis characteristic index by comparing the actual voltage curve and the reference open-circuit voltage curve during charge and discharge to quantify the degree of degradation of the electrochemical kinetic process; at the same time, statistically analyze the change in the ratio of charging capacity to discharging capacity during long-term operation to calculate the charge / discharge efficiency decay rate, reflecting the cumulative effect of side reactions and energy loss; finally, comprehensively analyze the three types of degradation features—internal resistance growth rate, voltage hysteresis characteristics, and charge / discharge efficiency decay—to obtain the current battery capacity decay rate.

[0076] S1.5 Based on the capacity decay rate, combined with historical operating condition data (including electrical data, thermal characteristic data, environmental data, charge and discharge condition data, and life-related data) and key degradation characteristics, the cumulative gas generation life mapping model is used to predict the cycle life and remaining usable life of the battery pack.

[0077] This cumulative gas production lifetime mapping model addresses the problem in traditional lithium battery state of health (SOH) prediction methods that rely on external parameters such as voltage and current while neglecting the direct characterization of internal electrochemical aging. It optimizes this by real-time monitoring of electrolyte decomposition gas production—a byproduct directly related to capacity decay—establishing a quantitative mapping relationship between cumulative gas production and capacity decay rate. This enables direct, online, and non-destructive assessment of battery aging status. Unlike existing technologies, this model dynamically calculates the gas production rate as a multivariable function of temperature, voltage, and current, and introduces an adaptively correctable parameter set. Combined with a Long Short-Term Memory (LSTM) network for double redundancy verification, it significantly improves the accuracy, early warning, and robustness of lifetime prediction. Especially in complex application scenarios with varying operating conditions and multiple stress couplings, it can provide earlier and more accurate warnings of aging trends, offering a more reliable basis for battery safety management and remaining lifetime assessment.

[0078] Predicting the cycle life and remaining usable life of a battery pack using a cumulative gas generation lifetime mapping model includes the following steps:

[0079] S1.51. Calculate the gas generation rate based on the real-time gas concentration data from the multimodal sensing unit 1. (Unit: mol / s), this rate is a function of temperature, voltage (potential), and current. High temperature, high voltage (overcharging tendency), and high current will significantly accelerate gas production. In the formula, Battery operating temperature, This refers to the battery terminal voltage or electrode potential. For battery charging and discharging current, This is a functional relationship representing the variation of gas generation rate with temperature, voltage, and current. Based on electrochemical principles and battery side reaction mechanisms, gas generation is primarily influenced by temperature, terminal voltage, and charge / discharge current. Gas generation rates under different operating conditions are observed using experimental data, and sensitivity coefficients for each factor are extracted using fitting or regression methods. Mathematically, the effects of temperature and voltage are expressed exponentially to reflect thermal acceleration and overvoltage acceleration effects, while the effect of current is represented by a power function to indicate a nonlinear relationship. Finally, all influencing factors are combined into a product form, forming a function that can be directly used for real-time calculations. And the parameters and sensitivity coefficients were calibrated through experiments;

[0080] S1.52 Calculate the effective cumulative gas production based on the effective cumulative gas production. (Considering that not all generated gases will persist for a long time (some can be reabsorbed by the electrolyte or electrode materials or participate in reversible reactions)): In the formula, This is an efficiency factor function used to adjust based on real-time operating conditions. Determine the proportion of irreversible gas production in the current process (e.g., low-rate gas production within a normal voltage window may be partially reversible, while gas production under a high-voltage window is essentially irreversible). , The reversible reaction efficiency represents the percentage of gas that reacts at the current temperature. and voltage The substance will be consumed or absorbed by a reversible reaction, with a value ranging from 0 to 1. For absorption or dissolution efficiency, the reabsorption of gas in the electrolyte or electrode material is considered, and the value is taken from 0 to 1, varying with temperature and pressure. change, This is a correction factor related to current or state of charge, representing the difference in gas production efficiency under high current and large cycle amplitude. , The current sensitivity coefficient (characterizing the increase in the rate of side reactions under high current) The empirical value is between 0.1 and 0.5. This is the current nonlinearity influence factor (empirical value greater than 1, used to describe the nonlinear enhancement of polarization effect with current). This is the current charging / discharging current. The reference current is taken as the rated current or 1C current (based on empirical values). For time, (The unit is mol).

[0081] S1.53. Based on extensive benchtop experiments and electrochemical model analysis in the early stages, and considering the effective cumulative gas production... Preliminary establishment of effective cumulative gas production Reference mapping relationship with battery capacity retention rate (Reference mapping relationship) (Preliminary fitting using linear or exponential functions) and construction of an adaptively adjustable parameter set. Cumulative gas production lifetime mapping model (including additional parameters such as temperature sensitivity and nonlinearity factors): (In the formula, The calculated battery health status (i.e., capacity retention rate). Let be the mapping function, representing the relationship between effective cumulative gas production and health status, and the parameter set . It includes key physicochemical parameters describing decay rate, nonlinearity, and temperature sensitivity. The decay rate coefficient is a key adaptive parameter (belonging to...). (Collection). It comprehensively reflects the sensitivity of battery chemistry systems (such as NMC or LFP) to gas production. The larger the value, the stronger the destructive effect of gas production on capacity and the faster the decay. An empirical value is [value missing]. ; To decay the nonlinear exponent, this is another key adaptive parameter (belonging to...). The set (which describes the behavior of the decay rate as a function of gas production) has an empirical value of 0.5–2.0 and is used to calculate the healthy state. In the initial state, ;

[0082] Among them, the effective cumulative gas production has been initially established. Reference mapping relationship with battery capacity retention rate And construct an adaptively adjustable parameter set. The connection between the cumulative gas production lifetime mapping model and the previous extensive bench experiments and electrochemical model analysis is as follows: based on the effective cumulative gas production... Based on the corresponding data of battery capacity retention rate, a cumulative gas production lifetime mapping model is fitted to determine the model. A set of initial parameters In the formula, The initial decay rate coefficient, The initial decay nonlinear exponent;

[0083] This initial parameter set The defined mapping relationship is the reference mapping relationship. Its functional form is the same as Same, represented as:

[0084] .

[0085] During system initialization, Load the model to enable the adaptive mapping model It possesses initial computing capabilities; subsequently, during battery operation, the parameter set... Based on the individual characteristics of the battery and real-time operating data, the dual-redundancy health assessment system will be used for online dynamic adjustment and optimization, thereby realizing the evolution from a universal reference mapping to personalized and accurate prediction.

[0086] During each charge-discharge cycle, especially under high stress (high temperature, high voltage, high current), trace amounts of electrolyte decompose and produce trace amounts of gas (such as CO2, CH4, C2H4). These side reactions consume active lithium and electrolyte, directly leading to battery capacity decay and internal resistance increase. Therefore, gas production is one of the most direct and online-monitorable physical characteristics of battery chemical aging. The larger the value, the more irreversible side reactions occur, and the more severe the battery aging. The cumulative gas production life mapping model reflects the healthy state of battery capacity decay dominated by gas-producing side reactions such as electrolyte decomposition, with the cumulative gas production... For every additional unit, the state of battery capacity retention (SOH) decreases. The quantity defined by the function is the quantification of battery health from the direct products of the aging reaction;

[0087] S1.54. Compare the health status with the corrected health status from step S1.3 to construct a dual-redundant health assessment system: If there is a discrepancy, the mapping function is dynamically updated and optimized using adaptive filtering (such as extended Kalman filtering or particle filtering). parameter set Dynamic correction is performed to ensure that the gas model can continuously learn and approximate the actual life decay law under different operating conditions. Through the complementary correction of the gas model and the long short-term memory network model, the accuracy and robustness of the long short-term memory network model calculation are improved.

[0088] S1.55. Set the capacity retention rate threshold corresponding to the end of battery life (i.e., the effective cumulative gas production when the battery capacity retention rate drops to 80%) to provide a criterion and reference benchmark for life prediction, which is used to determine the threshold point when the battery enters its battery life.

[0089] S1.56, Based on effective cumulative gas production Historical data predicts remaining usable lifespan Based on effective cumulative gas production Historical data, fitting gas production rate With time or Its own trend model (For example, exponential growth type), combining the current battery state (including SOH, internal resistance) and recent average operating conditions (including temperature, current), predict the future gas production rate. (In the formula, The parameter set for the trend model is used to fit the relationship between gas production and gas production rate (e.g., growth coefficient, time constant). The remaining usable lifetime (RUL) in time form is calculated by solving the following equation:

[0090] ;

[0091] In the formula, This refers to the effective cumulative gas production at the end of battery life, i.e., the cumulative gas production when the capacity retention drops to a critical value (e.g., 80%). This represents the effective gas production accumulated at the current moment. This is a predicted value for future gas generation rates. The current moment;

[0092] S1.57, Prediction results based on remaining usable lifetime Analyze key historical data to calculate the average number of equivalent full cycles completed per day / week. And predict the remaining number of cycles. : .

[0093] The risk prediction and early warning module 22 monitors potential dangerous states through a long short-term memory network model and triggers an early warning mechanism when the risk approaches the preset safety boundary to generate an alarm signal, thereby improving the system's safety response speed.

[0094] The risk prediction and early warning module 22 monitors potential hazardous conditions and triggers an early warning mechanism when the risk approaches a preset safety boundary, including the following steps:

[0095] S2.1 Compare the multimodal raw data (including single-core voltage, total voltage, charging and discharging current, temperature, deformation and gas concentration) with preset thresholds (including maximum allowable temperature, maximum charging and discharging current, single-core voltage range, consistency tolerance, and upper limit of gas concentration). If any multimodal raw data is close to or exceeds the upper limit threshold b, it is marked as a potential risk event.

[0096] S2.2 Output a specific high-frequency micro-current excitation signal to the battery pack, monitor the voltage response of the battery pack, calculate the high-frequency impedance spectrum of the battery pack, and fuse the deviation characteristics and multi-mode characteristics in the high-frequency impedance spectrum to generate enhanced multi-mode characteristics.

[0097] Furthermore, step S2.2 addresses the challenge of traditional battery management systems (BMS) failing to detect early and accurate micro-failures within the battery (such as micro-short circuits, SEI film thickening, and active material degradation). It actively applies high-frequency micro-current excitation to acquire impedance spectra characterizing the electrochemical interface. Unlike existing technologies that rely solely on external operating parameters (voltage, temperature) for passive monitoring, its advantage lies in its ability to penetrate external phenomena and directly and non-destructively capture early electrochemical characteristics of latent faults such as soft short circuits and increased polarization within the battery. By fusing this high-dimensional impedance deviation information with multimodal data, it significantly improves the early detection, sensitivity, and reliability of thermal runaway risk prediction, achieving a leap from monitoring external phenomena to gaining insight into the internal state.

[0098] The specific steps involved in generating enhanced multimodal features are as follows:

[0099] During the battery pack's charging / discharging intervals or rest periods (when potential risk events are marked, waiting for the battery to enter the charging / discharging intervals or rest periods), the equalization circuit of power execution unit 3 (referring to the circuit module used to balance the differences in voltage or state of charge (SOC) of each cell) applies a set of micro-amplitude AC current excitation signals (amplitude less than 0.1C, to ensure safety) within a specific frequency range (e.g., 1Hz-1000Hz) to the battery, and simultaneously acquires high-precision voltage response signals from each battery cell; based on the micro-amplitude AC current excitation signals and voltage response signals, the AC impedance of each cell at multiple characteristic frequency points is quickly calculated (AC impedance includes amplitude). and phase angle The amplitude and phase difference of the voltage and current are obtained by Fourier transform or phase-locked loop detection methods, respectively, and the AC impedance is calculated from this. The amplitude... Reflects the impedance and phase angle of the battery cell at that frequency. Reflecting the time delay between current and voltage, this process yields the high-frequency impedance spectrum of each cell at various characteristic frequencies, used to characterize the cell's health and internal differences (for each characteristic frequency). Furthermore, a high-frequency impedance spectrum for each cell is constructed based on its AC impedance (amplitude values ​​are considered). and phase angle Arranged in frequency order, a frequency-amplitude-phase correspondence curve is formed, which yields the high-frequency impedance spectrum of the cell. The deviation characteristics between the high-frequency impedance spectrum and the reference impedance spectrum of a healthy cell are extracted (the high-frequency impedance spectrum of each cell is compared with the reference impedance spectrum of a healthy cell, and the difference in amplitude and phase is calculated at the same characteristic frequency point to obtain deviation characteristics such as amplitude shift, phase angle shift, growth rate of the real part of impedance in the mid-frequency region, and drift at the characteristic frequency point. The reference impedance spectrum of a healthy cell is obtained by calculating the AC impedance of a cell in the early stage of its lifespan or unused and fully tested, after applying a small AC excitation under the same conditions and measuring the voltage response). The deviation characteristics are fused with multi-mode characteristics to form an enhanced multi-mode characteristic.

[0100] S2.3. Input the enhanced multimodal features into the long short-term memory network model to predict the trends of thermal runaway, overcharging, and over-discharging risks, and assess the risk probability within a future period T.

[0101] S2.4. Based on risk prediction results and risk probabilities, risk levels (including low, medium, and high) are classified, and the warning triggering conditions are dynamically adjusted. The potential risk probability predicted by the Long Short-Term Memory Network Model over a future period is compared with a preset threshold. Based on the magnitude of the risk probability, the battery status is divided into three levels: low, medium, and high. A risk probability below the risk threshold p is considered low risk; a risk probability above the risk threshold p but below the lower limit threshold d is considered medium risk; and a risk probability above the lower limit threshold d is considered high risk. Simultaneously, the system combines real-time operating condition information such as the rate of change of gas concentration and temperature, SOC, and current fluctuations to dynamically adjust the warning triggering conditions. For example, when the rate of change of the predicted risk probability per unit time reaches a preset rate threshold g, the system automatically lowers the warning triggering threshold to improve warning sensitivity; when the risk probability is below the preset lower limit threshold q and the rate of change is below the set rate threshold g, the system appropriately raises the warning triggering threshold to reduce false alarms, thereby achieving adaptive matching between the risk level and the warning triggering mechanism.

[0102] S2.5 When the risk probability reaches or exceeds the preset threshold v, the early warning mechanism is triggered (the early warning mechanism is to issue an alarm signal in advance to indicate the existence of danger), generating a corresponding alarm signal and transmitting the alarm signal to the communication and interaction unit 4.

[0103] The control strategy and decision output module 23 dynamically generates the battery pack's operation control strategy based on the life prediction and risk assessment results, including overcharge / over-discharge protection, current limiting, heat dissipation management and equalization charging strategy, and receives feedback signals from the power execution unit 3. It integrates the lithium battery pack's state of charge, health status, capacity decay trend, battery cycle life and remaining usable life and feedback signals into status information.

[0104] The control strategy and decision output module 23 dynamically generates the battery pack's operation control strategy based on lifespan prediction and risk assessment results, including the following steps:

[0105] S3.1 Based on the battery health status and risk level, the system calls the preset control strategy library (which includes specific operation schemes such as overcharge protection, over-discharge protection, equalization charging strategy, current limiting, and heat dissipation adjustment; each strategy corresponds to specific triggering conditions, execution parameters, and priorities), and generates a dynamic control strategy in combination with the actual operating conditions. When multiple strategies conflict, the execution priority is determined according to the principle of safety over lifespan and lifespan over efficiency. For example, when the battery pack health status is assessed as a moderate degradation level and the predicted risk probability is close to the lower limit threshold d, a dynamic control strategy is generated: under the premise of ensuring safety, the charging current is limited to 80% of the rated current, while the equalization charging module is activated to adjust the energy of high and low voltage cells, and the power of the heat dissipation device is automatically adjusted according to the temperature sensor feedback to keep the voltage of each cell within the allowable range (e.g., 3.0V to 4.2V), while extending the cycle life of the battery pack.

[0106] S3.2. The generated dynamic control strategy is converted into executable control commands and sent to the power execution unit 3 to drive the battery pack to perform specific actions such as overcharge protection, over-discharge protection, inter-cell energy balancing, or heat dissipation management.

[0107] S3.3 Receive the execution status and running results returned by the power supply execution unit 3, compare them with the original control instructions, and use them to update the dynamic control strategy; if the execution effect deviates from the expectation, the control parameters are adjusted through the dynamic correction mechanism, and the strategy library is updated in real time to achieve closed-loop optimization.

[0108] The power execution unit 3 performs overcharge and over-discharge protection operations on the battery pack based on control commands, and transfers energy between the cells according to the intelligent balancing strategy;

[0109] In this embodiment, the power supply execution unit 3 includes a protection control module and a balance control module;

[0110] The protection control module receives and parses the control commands output by the control strategy and decision output module 23, performs overcharge and over-discharge protection operations on the battery pack, and cuts off or limits the charging and discharging circuit in abnormal situations.

[0111] The equalization control module compares the cell voltages of the battery pack according to an intelligent equalization strategy, identifies high- and low-voltage cells, and balances the energy between cells through resistive dissipation or inductor / capacitor energy transfer. (The intelligent equalization strategy specifically involves: sorting and analyzing voltage differences (arranging the real-time voltages of each cell in the battery pack from high to low or low to high, and then calculating the difference between each cell's voltage and the average voltage or the voltage of adjacent cells), identifying high- and low-voltage cells. Subsequently, an energy regulation method is selected according to a preset equalization method: in passive equalization mode, excess energy from high-voltage cells is converted into heat through resistive dissipation; in active equalization mode, the excess energy from high-voltage cells is converted into heat through resistive dissipation; in active equalization mode, the excess energy from high-voltage cells is converted into heat through resistive dissipation.) Excess energy is transferred from high-voltage cells to low-voltage cells via inductors, capacitors, or DC-DC converters. During the equalization process, the module continuously monitors the voltage, current, and temperature of each cell and dynamically adjusts the equalization amplitude and duration based on real-time feedback until the voltage difference between cells is within a safe range (e.g., the voltage difference between any two cells in the same string does not exceed 20–50mV (the specific value is determined according to the battery type and manufacturer's recommendations)). During execution, the module monitors operating parameters such as voltage, current, and temperature in real time, generates protection action status and equalization process data, and feeds it back to the communication and interaction unit 4 to achieve closed-loop control of safety protection and performance optimization.

[0112] The communication and interaction unit 4 uploads status information and alarm signals to the field DTU equipment, and displays and interacts with the lithium battery pack data in real time through the remote monitoring platform;

[0113] In this embodiment, the communication and interaction unit 4 uploads the state of charge, health status, capacity decay trend, remaining usable life, and remaining cycle count of the lithium battery pack to the field DTU device via a communication interface (CAN, RS485, or Ethernet). At the same time, it uploads alarm signals to the alarm signal port (VL / COM) of the original charging module of the field DTU (equipment used for remote data acquisition and transmission) via dry contacts, and the DTU forwards them to the power distribution master station backend for real-time operation monitoring and alarm linkage. It also supports local activation operation and alarm simulation triggering, enabling maintenance personnel to intuitively verify the system's operational stability and safety under conditions such as high temperature, low temperature, and high current.

[0114] In this embodiment, the intelligent lithium power supply was tested as follows:

[0115] High-temperature operation test (ambient temperature 70+℃): Test 1 (float charge): The battery pack was kept in float charge state for 72 hours in an environment with a temperature above 70℃. The system continuously monitored and maintained operation, and the state was stable (initially as follows). Figure 2 As shown, it takes 72 hours to reach... Figure 3As shown); Test 2 (Discharge): Continue to maintain an environment of 70+℃, discharge at a current of 2.2A for 5 hours. The battery appearance shows no abnormalities such as bulging or leakage, and the management system protection functions are normal (e.g. Figure 4 , Figure 5 , Figure 6 , Figure 7 (as shown)

[0116] Low-temperature environment operation test (-20℃): The battery pack started and operated in a low-temperature environment of -20℃. The management system successfully woke up the power supply. Performance parameters were accurately monitored under low-temperature conditions, and there were no abnormalities in operation (e.g., Figure 8 , Figure 9 (As shown).

[0117] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A lithium battery pack intelligent management system based on multimodal sensing, characterized in that, include: Multimodal sensing and fusion unit (1), the multimodal sensing and fusion unit (1) collects the battery pack operating environment and internal state data in real time, and fuses the operating environment and internal state data by weighted average to generate fused multimodal data; The intelligent management main control unit (2) predicts thermal runaway risk through a long short-term memory network model and generates control commands, status information and alarm signals. The power execution unit (3) performs overcharge and over-discharge protection operations on the battery pack based on control commands, and performs energy transfer between the cells according to the intelligent balancing strategy. The communication and interaction unit (4) uploads status information and alarm signals to the field DTU equipment and displays and interacts with the lithium battery pack data in real time through the remote monitoring platform. The intelligent management main control unit (2) includes a state calculation and life prediction module (21), a risk prediction and early warning module (22), and a control strategy and decision output module (23). The state calculation and lifetime prediction module (21) calculates the state of charge, health status, and capacity decay trend of the lithium battery pack in real time based on the fused multimodal data and the equivalent circuit model. It also constructs a cumulative gas production lifetime mapping model to characterize the relationship between effective cumulative gas production and health status, which is used to calculate the health status. Through the health status threshold, it determines the threshold point at which the battery enters its battery life. Based on the historical data of effective cumulative gas production, it predicts the remaining usable life of the battery. Based on remaining available lifetime And the average number of equivalent full cycles completed per day or per week. Predict the remaining number of iterations.

2. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 1, characterized in that: The multimodal sensing and fusion unit (1) includes a multimodal sensing module, a multimodal fusion module, and a multimodal feature extraction module; The multimodal sensing module collects operating environment and internal state data, and preprocesses the collected environmental and internal state data. The operating environment and internal state data include total voltage, single-core voltage, charging and discharging current, temperature, gas concentration, and historical key data. The multimodal fusion module performs time-series alignment and normalization on the preprocessed environmental and internal state data, and then fuses them by weighted averaging to generate fused multimodal data. The multimodal feature extraction module extracts multimodal features from the fused multimodal data through statistical analysis and time-frequency transformation methods.

3. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 1, characterized in that: The risk prediction and early warning module (22) monitors potential dangerous states through a long short-term memory network model and triggers an early warning mechanism to generate an alarm signal when the risk approaches the preset safety boundary. The control strategy and decision output module (23) dynamically generates the battery pack operation control strategy based on the life prediction and risk assessment results, and receives feedback signals from the power execution unit (3), integrating the state of charge, health status, capacity decay trend, cycle life and remaining usable life of the lithium battery pack and feedback signals into status information.

4. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 1, characterized in that: The state calculation and life prediction module (21) performs real-time calculations on the state of charge, state of health, and capacity decay trend of the lithium battery pack, and predicts the remaining usable life and remaining cycle count of the battery, including the following steps: S1.1 Preprocess the fused multimodal data; S1.2 Based on the equivalent circuit model, the state of charge of the battery is calculated in real time; combined with temperature and current information, the capacity utilization rate and voltage plateau characteristics are corrected to obtain the preliminary health status. S1.

3. The battery state of charge, health status and multimodal features are fused to form a time-series input feature sequence, which is then input into a long short-term memory network model to perform nonlinear correction on the battery state of charge, health status results and multimodal features, and generate the corrected battery state of charge and health status. S1.4 Extract key degradation features based on historical key data to determine the current battery capacity decay rate; S1.5 Based on the capacity decay rate, combined with historical operating data and key degradation characteristics, predict the cycle life and remaining usable life of the battery pack. S1.5 includes the following steps: S1.

51. Calculate the gas generation rate based on real-time gas concentration data from the multimodal sensing and fusion unit (1). ; S1.52, Gas-based generation rate Calculate the effective cumulative gas production ; S1.

53. Based on the effective cumulative gas production, a preliminary reference mapping relationship between the effective cumulative gas production and the health status is established. At the same time, an adaptively adjustable parameter set is constructed. The cumulative gas production lifetime mapping model characterizes the relationship between effective cumulative gas production and health status, and is used to calculate the health status. (Parameter set...) It includes descriptions of decay rate and decay nonlinearity exponent; S1.

54. Compare the health status calculated in S1.53 with the corrected health status in step S1.3 to construct a dual-redundant health assessment system. If there is a discrepancy between the two, adaptive filtering is used to adjust the parameter set. Perform dynamic correction; S1.

55. Set the health state threshold corresponding to the end of battery life, which is used to determine the threshold point when the battery enters the battery life prediction. S1.56, Based on effective cumulative gas production Historical data predicts remaining usable lifespan ; S1.57, Prediction results based on remaining usable lifetime Analyze key historical data to calculate the average number of equivalent full cycles completed per day or week. And predict the remaining number of cycles.

5. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 3, characterized in that: The risk prediction and early warning module (22) monitors potential dangerous conditions and triggers an early warning mechanism when the risk approaches a preset safety boundary, including the following steps: S2.1 Compare the multimodal raw data with the preset threshold. If any multimodal raw data exceeds the upper limit threshold b, it is marked as a potential risk event. S2.2 Output a high-frequency micro-current excitation signal to the battery pack, monitor the voltage response of the battery pack, calculate the high-frequency impedance spectrum of the battery pack, and fuse the deviation characteristics and multi-mode characteristics in the high-frequency impedance spectrum to generate enhanced multi-mode characteristics. S2.

3. Input the enhanced multimodal features into the long short-term memory network model to predict the trends of thermal runaway, overcharging and over-discharging risks, and assess the risk probability within a future period T. S2.

4. Based on the risk prediction results and risk probability, classify the risk levels and dynamically adjust the early warning trigger conditions; S2.5 When the risk probability reaches the preset threshold v, the early warning mechanism is triggered, the corresponding alarm signal is generated, and the alarm signal is transmitted to the communication and interaction unit (4).

6. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 5, characterized in that: In step S2.2, the specific steps involved in generating enhanced multimodal features are as follows: During the resting period of battery pack charging and discharging, a set of micro-amplitude AC current excitation signals with a specific frequency range are applied to the battery through the equalization circuit of the power execution unit (3), and the voltage response signals of each battery cell are collected simultaneously. Based on the micro-amplitude AC current excitation signals and voltage response signals, the AC impedance of each cell at multiple characteristic frequency points is calculated, and the high-frequency impedance spectrum of each cell is constructed based on the AC impedance. The deviation characteristics between the high-frequency impedance spectrum and the reference impedance spectrum of the healthy cell are extracted, and the deviation characteristics are fused with the multi-mode characteristics to form an enhanced multi-mode characteristic.

7. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 3, characterized in that: The control strategy and decision output module (23) dynamically generates the battery pack's operation control strategy based on the life prediction and risk assessment results, including the following steps: S3.1 Based on the battery health status and risk level, call the preset control strategy library and generate a dynamic control strategy in combination with the actual working conditions; S3.

2. The generated dynamic control strategy is converted into control commands and sent to the power execution unit (3). S3.3 Receive the execution status and running results returned by the power supply execution unit (3), compare them with the original control instructions, and use them to update the dynamic control strategy.

8. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 1, characterized in that: The power supply execution unit (3) includes a protection control module and a balance control module; The protection control module receives and parses the control instructions output by the control strategy and decision output module (23), performs overcharge and over-discharge protection operations on the battery pack, and restricts the charging and discharging circuit under abnormal conditions. The equalization control module compares the cell voltages of the battery pack according to the intelligent equalization strategy, identifies high and low voltage cells, and balances the energy between cells through resistance dissipation.

9. The intelligent management system for lithium battery packs based on multimodal sensing according to claim 1, characterized in that: The communication and interaction unit (4) uploads the state of charge, health status, capacity decay trend, remaining usable life and remaining cycle count of the lithium battery pack to the field DTU device through the communication interface. At the same time, it uploads the alarm signal to the alarm signal port of the original charging module of the field DTU through the dry contact, and the DTU forwards it to the power distribution master station backend for real-time operation monitoring and alarm linkage.

Images