System and method for battery health management

Abstract

The application discloses a kind of system and method of battery health management, belong to battery management technical field, system includes implantable sensor network, edge and cloud collaborative computing platform, multi-modal prediction model and dynamic optimization execution unit;Implantable sensor network integrates multiple sensors to collect the internal multiple physical field parameters of battery, is transmitted by energy collection unit power supply by across shielding communication;Edge and cloud collaborative platform processes data, trains model and block chain notarization;Multi-modal prediction model fuses space-time double-flow Transformer and causal reasoning, predicts SOH, RUL and early warning thermal runaway;Dynamic optimization execution unit realizes efficient control by hierarchical balancing and adaptive thermal management;Method contains initialization, running and maintenance stage, forms closed loop.The application breaks through traditional limitation, improves battery safety, reliability and economy, is applicable to electric vehicle, energy storage power station and the like scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery management, in particular to a system and method for battery health management. BACKGROUND

[0002] With the rapid development of new energy industry, lithium ion batteries, as the core components of energy storage and conversion, have been widely used in electric vehicles, energy storage power stations and other fields. The accurate assessment and safety protection of battery state of health (SOH) and remaining useful life (RUL) directly affect the operation reliability, economy and safety of equipment. Therefore, battery health management technology has become the core topic of industry research.

[0003] Currently, although traditional battery management systems play a role in monitoring and protecting basic battery parameters, they still have obvious shortcomings in fine management capability under complex working conditions.

[0004] In terms of state perception, existing systems mainly rely on external voltage, current, surface temperature and other macroscopic parameters to indirectly infer the battery health state, and cannot directly obtain the physical and chemical changes inside the battery cell. For example, the microcrack initiation of electrode materials, the attenuation of electrolyte ion conductivity, and the decomposition of SEI film and other microscopic states are difficult to capture, resulting in health assessment based on empirical models, and the accuracy is limited by the mapping error between external parameters and internal states, which cannot provide essential basis for health management.

[0005] In terms of data processing and collaborative architecture, existing technologies mostly adopt single edge computing or centralized cloud computing mode: edge computing is limited by hardware computing power and is difficult to realize complex model training, cloud computing needs to upload a large amount of raw data, which has the risk of data privacy leakage, and lacks effective distributed data sharing mechanism. At the same time, traditional systems lack dynamic mapping capability of battery full life cycle state, real-time synchronization between digital model and physical battery is poor, and storage of key health data depends on centralized server, which is easy to be tampered or lost, and cannot meet the traceability requirement.

[0006] In terms of health prediction and early warning, existing prediction models are mostly based on single-dimensional data or simple machine learning algorithms, which are difficult to integrate multi-source heterogeneous information. For example, SOH prediction often uses empirical formula, ignoring the coupling effect of dynamic factors such as temperature and charge-discharge rate, and thermal runaway warning only relies on surface temperature threshold, lacking causal analysis of early features such as pressure and gas generation, resulting in insufficient early warning time of less than 10 minutes, which is difficult to meet the safety protection demand.

[0007] At the level of optimization control, the traditional active balancing system adopts fixed path energy transfer, does not consider the dynamic SOC difference between battery groups, and the balancing efficiency is lower than 90%. The thermal management strategy is mostly based on simple PID control, which can only maintain single-point temperature stability and cannot realize precise regulation of the global temperature difference of the battery group. Moreover, the balancing and thermal management strategies are independent of each other and do not form a synergistic optimization, resulting in limited improvement of battery cycle life.

[0008] Therefore, the application proposes a brand-new battery health management system and method. SUMMARY

[0009] One object of the application is to propose a battery health management system and method. The application can break through the limitations of traditional battery health management technology in state perception, data processing, prediction and early warning, and optimization control. Through implantable multi-physical field parameter acquisition, edge and cloud collaborative computing, multi-modal intelligent prediction, and dynamic synergistic optimization, the application realizes precise evaluation of battery health status, early safety warning, and efficient management throughout the life cycle, and significantly improves the safety, reliability, and economy of battery use.

[0010] According to an embodiment of the application, a battery health management system comprises an implantable sensor network, an edge and cloud collaborative computing platform, a multi-modal prediction model, and a dynamic optimization execution unit.

[0011] The implantable sensor network is used to collect multi-physical field parameters inside the battery, including temperature, pressure, electrode deformation, and electrolyte ion conductivity.

[0012] The edge and cloud collaborative computing platform is in communication connection with the implantable sensor network, and is used to process the collected multi-physical field parameters and build a battery health state model.

[0013] The multi-modal prediction model is deployed on the edge and cloud collaborative computing platform, and is used to predict battery health indicators based on the processed multi-physical field parameters. The health indicators include state of health (SOH) and remaining useful life (RUL).

[0014] The dynamic optimization execution unit is in communication connection with the edge and cloud collaborative computing platform, and is used to perform battery balancing control and thermal management adjustment according to the predicted health indicators, forming a health management closed loop.

[0015] Further, the implantable sensor network comprises a miniaturized sensor array, a cross-shield communication module, and an energy harvesting unit.

[0016] The micro-sensor array comprises a temperature and voltage integrated sensor, a distributed fiber Bragg grating strain sensor and a micro-nano flow chip type ion conductivity sensor, the sensor adopts a ceramic and polyimide composite packaging material, forms an electrolyte corrosion resistant coating through an atomic layer deposition process, the coating has a temperature resistance range of -40℃-150℃ and a voltage resistance of ≥500V;

[0017] The cross-shield communication module adopts a 2.4GHz carrier modulation technology and a LoRa protocol to penetrate a battery metal shell to transmit data, and the transmission delay is ≤10ms.

[0018] The energy collection unit generates an induced current to supply power to the sensor through battery charging and discharging voltage fluctuation, and the energy conversion efficiency is ≥65%.

[0019] Further, the central wavelength offset of the distributed fiber Bragg grating strain sensor satisfies the formula:

[0020] ;

[0021] wherein, is the initial wavelength of the grating, is the photoelastic coefficient, is the strain value of the electrode material, is the thermo-optic coefficient, is the temperature change, the strain and temperature cross-sensitivity are separated through a temperature compensation algorithm, and the deformation measurement accuracy is ±2με.

[0022] Further, the impedance spectrum of the micro-nano flow chip type ion conductivity sensor satisfies the formula:

[0023] ;

[0024] wherein, is the impedance of the sensor, is the solution resistance, is the double-layer capacitance, is the charge transfer resistance, is the alternating current frequency, is the imaginary unit, and the ion conductivity is calculated by L is the flow channel length, A is the cross-sectional area, and the measurement error is <3%.

[0025] Further, the edge and cloud collaborative computing platform comprises an edge node and a cloud intelligent hub.

[0026] The edge node is deployed with a MobileNetv3-based anomaly detection model and a federal learning node, the anomaly detection model filters invalid data in real time, the recognition rate is greater than or equal to 99.2%, the federal learning node shares desensitized aging features with adjacent battery groups, differential privacy protection is realized by adding Laplace noise, and the model accuracy loss is less than or equal to 2%;

[0027] The cloud intelligent hub includes a digital twin engine and a blockchain storage module, the blockchain storage module stores health indicators by using a PBFT consensus mechanism, the number of fault-tolerant nodes is , wherein n is the total number of nodes.

[0028] Further, the multi-modal prediction model includes a space-time double-flow Transformer module and a causal reasoning module, the space flow feature of the space-time double-flow Transformer module satisfies:

[0029] ;

[0030] , wherein, is a space flow feature vector of the i th sensor node, is a multilayer perceptron, is a feature splicing operation, is a position encoding of the spatial coordinates of the i th sensor node , is a strain feature of the i th node, is a temperature feature of the i th node, is a pressure feature of the i th node;

[0031] The time flow self-attention calculation satisfies:

[0032] ;

[0033] , wherein, is a self-attention output feature matrix, is a query matrix, is a key matrix, is a value matrix, , , , is a query matrix weight parameter, is a time sequence feature matrix at time t, is a key matrix weight parameter, is a time sequence feature matrix at time t, , is a value matrix weight parameter, is an adjustable time step, is the dimension of and , is a normalization function;

[0034] The causal inference module analyzes the causal relationship of multi-physical field parameters based on the Do-Calculus algorithm, the thermal runaway early warning time is greater than or equal to 30 minutes, and the accuracy is greater than or equal to 98%.

[0035] Further, the state of health (SOH) is calculated by a formula:

[0036] ;

[0037] wherein, , , is an aging coefficient, is a charge and discharge current, is a cell temperature, is a voltage change rate, is an initial capacity, is an integral time infinitesimal;

[0038] The remaining useful life RUL is calculated by a formula , is a retirement threshold.

[0039] Further, the hierarchical active balancing system of the dynamic optimization execution unit includes bottom layer cell balancing and top layer inter-group balancing, the bottom layer uses a Buck-Boost circuit to realize energy transfer, and the balancing efficiency wherein ;

[0040] wherein, is a cell balancing efficiency, is a circuit output voltage, is an output current, is a balancing time, is a circuit input voltage, is an input current, and ;

[0041] The top layer realizes inter-group energy scheduling through a single inductor circuit, optimizes the path based on a target function , and the constraint condition is , and the solution is obtained through an improved genetic algorithm;

[0042] wherein, is an inter-group energy transfer current, is an energy transfer path resistance, is a scheduling time, is a summation operation, is a minimum value operation.

[0043] Further, the adaptive thermal management module of the dynamic optimization execution unit adopts model predictive control, and a temperature prediction model satisfies the formula:

[0044] ;

[0045] Wherein is a state matrix, is a control matrix, is a disturbance matrix, is an actual temperature of the battery at time t, is a liquid cooling flow rate / heating power, is a disturbance term;

[0046] A control target function satisfies , and a constraint condition is , the battery temperature difference can be controlled within ±1.5 DEG C by solving through a sequential quadratic programming algorithm in cooperation with a paraffin and graphene composite phase change material.

[0047] Wherein, is a control target function value, n is a prediction time domain length, is a prediction step, is a temperature deviation weight coefficient, is a control weight coefficient, is a target temperature.

[0048] A battery health management method, comprising the following steps:

[0049] S1, an initialization stage, an initial impedance spectrum of the battery is obtained through pulse test, an individual feature library is constructed, a digital twin virtual battery is generated synchronously, and a blockchain is stored;

[0050] S2, in the running stage, the edge node collects multi-physical field parameters every 100ms, uploads to the cloud after noise reduction processing, the cloud updates the health prediction model based on the SOH formula every hour, and issues the optimization parameters of the target function solution to the edge node;

[0051] S3, in the maintenance stage, when the predicted remaining life RUL is less than or equal to 50 cycles, a personalized life extension scheme is generated, and after the battery is retired, the residual value is evaluated based on the blockchain data to guide the echelon utilization, and the remaining life RUL is calculated by , is a retirement threshold.

[0052] The beneficial effects of the present application are:

[0053] 1、The present application directly collects multiple physical field parameters such as battery internal temperature, pressure, electrode deformation and electrolyte ion conductivity through an implantable sensor network, combines specific formula quantitative sensing principle, realizes accurate capture of battery microstate, and provides more essential basis for health assessment, and the prior art cannot realize such comprehensive and accurate direct collection and quantitative analysis of multiple physical field parameters in the battery.

[0054] 2、The edge and cloud collaborative computing architecture is adopted in the present application, the federated learning mechanism of the edge node realizes aging feature sharing while protecting data privacy, the digital twin engine of the cloud and the blockchain storage are combined, which not only dynamically maps the battery state but also ensures that the data cannot be tampered with, and such collaborative mode and data processing and storage method are not possessed by the prior art, and the data utilization efficiency and security are improved.

[0055] 3、The multi-modal prediction model in the present application fuses space-time double-flow Transformer and causal reasoning algorithm, and constructs health index calculation and prediction logic through specific formula, which can not only accurately predict SOH and RUL, but also can early warn thermal runaway for more than 30 minutes, and the prediction accuracy and warning timeliness are far superior to traditional experience model, and the prior art lacks such multi-dimensional fusion and accurate quantitative prediction mechanism.

[0056] 4、The hierarchical active balancing system of the dynamic optimization execution unit in the present application adopts an improved genetic algorithm to optimize the energy path, combines an adaptive thermal management MPC control strategy, realizes efficient balancing and accurate temperature control through a quantitative formula, and significantly prolongs the battery cycle life by more than 25%, and the prior art cannot realize such efficient and accurate dynamic optimization control. BRIEF DESCRIPTION OF DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0058] Figure 1 A system overall framework structure schematic diagram of a battery health management system and method proposed in the present application;

[0059] Figure 2 A method flowchart of a battery health management system and method proposed in the present application. DETAILED DESCRIPTION

[0060] In order to make the technical means and purposes and effects of the present application easy to understand, the embodiments of the present application are described in detail below in combination with specific drawings.

[0061] Embodiment 1

[0062] As Figure 1As shown, this embodiment discloses a battery health management system that achieves precise health management throughout the entire battery lifecycle through deep integration of embedded sensing, edge and cloud collaborative computing, multimodal prediction, and dynamic optimization execution.

[0063] The system specifically includes an implanted sensor network, an edge and cloud collaborative computing platform, a multimodal prediction model, and a dynamic optimization execution unit. The structure and working principle of each part are as follows:

[0064] Implantable sensor networks are integrated inside battery cells to collect multi-physics parameters during battery operation in real time. Their core components include a miniaturized sensor array, a cross-shielded communication module, and an energy harvesting unit. The implantable sensor network acts as the system's "sensory nerves," and the arrangement of its miniaturized sensor array is as follows:

[0065] The integrated temperature and pressure sensor is surface-mounted near the battery cell tabs, where temperature and pressure changes best reflect the reaction intensity of the active materials in the battery cell. The sensor's sensing element is a silicon-doped piezoresistor. The temperature measurement range is -40℃ to 150℃, and the pressure measurement range is 0 to 500 kPa. The internal ADC module converts the analog signal into a 16-bit digital signal, ensuring a temperature accuracy of ±0.2℃ and a pressure accuracy of ±0.1 kPa.

[0066] The distributed fiber Bragg grating strain sensor is embedded in a serpentine path between the electrode plates and the current collector, using a fiber grating with a grating length of 10 mm and an initial wavelength of... =1550nm, elastic coefficient =0.22, thermo-optical coefficient The wavelength shift is monitored in real time using an optical demodulator. Then according to the formula Calculate electrode deformation To eliminate the interference of temperature on deformation measurement, another strain-free grating on the same optical fiber is used as a temperature reference. Temperature compensation is achieved through differential calculation, ultimately enabling the deformation measurement accuracy to reach ±2. It can capture the minute strain caused by the volume expansion of electrode materials during charge-discharge cycles.

[0067] The micro / nanofluidic chip-type ion conductivity sensor is integrated near the electrolyte filling port of the battery cell. The chip is fabricated using PDMS material through soft photolithography, and the flow channel length is... =5mm, cross-sectional area =0.1mm², platinum electrodes are placed at both ends of the flow channel, and an AC signal of 10Hz-1MHz is applied. The impedance spectrum is measured by an impedance analyzer. The solution resistance is: The ion conductivity can be obtained by the real part intercept of impedance spectrum in high frequency band The measurement error of the sensor is less than 3%, and the change of the migration ability of lithium ions in the electrolyte can be effectively reflected. For example, when the SEI film is broken and the electrolyte is decomposed, the ion conductivity will decrease by more than 5% within 1 hour.

[0068] The cross-shield communication module adopts a system-on-chip design and works in the 2.4GHz ISM frequency band. In order to penetrate the metal shell of the battery, frequency hopping spread spectrum technology is used, the channel switching rate is 100 hops per second, the transmission power is controlled within 10dBm, and the receiving sensitivity is -148dBm. By pasting a flexible FPCB antenna on the inside of the battery shell, wireless transmission of sensor data is realized. Through actual testing, when the thickness of the battery stack reaches 30cm, the transmission delay can still be controlled to be ≤10ms, and the packet loss rate is <0.1%.

[0069] The energy harvesting unit is composed of a micro coil, a rectifier bridge and a super capacitor. The coil is wound around the cell shell. When the battery charges and discharges, the coil induces an alternating voltage. After the rectifier bridge converts it into a direct current voltage, it supplies power to the 3.3V sensor. The energy conversion efficiency is ≥65%. At a 1C charge / discharge rate, it can output a stable current of 5mA. With a 1mF super capacitor, the sensor can still work continuously for more than 4 hours when the battery is at rest.

[0070] The edge and cloud collaborative computing platform undertakes data processing and model building functions, and is divided into edge nodes and cloud intelligent hubs.

[0071] The edge node is deployed based on the MobileNetv3 anomaly detection model. The model input is the time series segment of temperature, pressure, deformation and ion conductivity collected by the sensor. The features are extracted through a depth separable convolution layer, and the abnormal probability is output through a Sigmoid activation function. When the probability is >0.5, it is determined as invalid data. After training on 100,000 labeled data, the invalid data recognition rate of this model reaches 99.2%.

[0072] At the same time, the edge node integrates a federated learning node, establishes encrypted communication with the edge nodes of adjacent battery packs, and shares the desensitized aging features. The workflow of the federated learning node is as follows:

[0073] The edge node standardizes the local aging features, encrypts them through a homomorphic encryption algorithm, and exchanges parameters with the edge nodes of the adjacent 10 battery packs. The aggregation server updates the model parameters using the federated averaging algorithm. To protect data privacy, Laplace noise is added before uploading the features. The noise scale is dynamically adjusted according to the feature sensitivity. Through testing, the model accuracy loss is ≤2%.

[0074] The cloud intelligent hub adopts a GPU cluster architecture, and its digital twin engine builds a three-dimensional electrochemical virtual model of the battery based on COMSOL Multiphysics;

[0075] Specifically, the construction process of the digital twin engine is as follows:

[0076] First, the three-dimensional structure of the battery is obtained through CT scanning, and an electrochemical model containing the positive electrode, negative electrode, separator, and electrolyte is established in COMSOL. The initial values of the key parameters in the model are obtained from literature, and then the multi-physics field parameters uploaded by the edge node every hour are received. The particle swarm optimization algorithm is used to dynamically calibrate the model parameters, so that the error between the simulation values of the virtual battery voltage and temperature and the measured values of the physical battery is less than 2%, and dynamic mapping is achieved.

[0077] The blockchain storage module is built based on the Hyperledger Fabric framework, which includes 3 ordering nodes and 10 peer nodes, adopts the PBFT consensus mechanism, and generates a block every 10 seconds. Each block contains 50 health indicator records, and the data is stored after being hashed by SHA-256, ensuring that it cannot be tampered with. Its fault tolerance capability satisfies When = 10, = 3, it can resist malicious attacks from 3 nodes.

[0078] The multi-modal prediction model is deployed on the edge and cloud collaborative computing platform, which realizes health state prediction by fusing multi-physics field data.

[0079] The model includes a spatio-temporal dual-flow Transformer module and a causal reasoning module:

[0080] The specific structure of the spatio-temporal dual-flow Transformer module is as follows: the spatial flow contains 4 convolutional attention modules. Each module first performs 3x3 convolution on the 32x32 sensor spatial distribution matrix, and then calculates the correlation weight between nodes through the self-attention mechanism, outputting:

[0081] ;

[0082] Where is the position encoding of the sensor coordinates, are the strain, temperature, and pressure features of the i-th node, respectively;

[0083] The time flow adopts a 6-layer Transformer encoder. The input is the time series data of the past 1 hour, which is processed through the self-attention mechanism after position encoding, where , is the feature matrix at time t, The time step is adjustable from 5 to 30 seconds to capture long-term trends of battery aging.

[0084] The causal graph constructed by the causal reasoning module based on the Do-Calculus algorithm contains 12 nodes and 30 causal edges. The key causes are identified through intervention analysis. When the causal chain of "temperature > 60℃ and pressure increase rate > 5kPa / min and ion conductivity decrease > 10%" is detected, a thermal runaway warning is triggered. Through 100 simulated thermal runaway experiments, the warning time is ≥30 minutes and the accuracy is ≥98%.

[0085] In the calculation of the state of health SOH, the aging coefficient is calibrated through accelerated aging experiments: 100 batteries of the same specification are cycled to SOH=0.8 at different temperatures and different rates, and the , , data are recorded. Specifically, the formula is used for calculation, where is the aging coefficient calibrated through experiments, is the charge and discharge current, is the cell temperature, is the voltage change rate, is the initial capacity;

[0086] The prediction process of the remaining useful life RUL is as follows: first, predict the SOH curve of the next 100 cycles through the LSTM network, then calculate , is the retirement threshold, i.e. the cycle number when SOH first falls below the threshold after n cycles, where the retirement threshold can be set to 0.8.

[0087] The dynamic optimization execution unit executes active balancing control and adaptive thermal management adjustment based on the output of the multi-modal prediction model.

[0088] The hierarchical active balancing system is divided into bottom cell balancing and top group balancing:

[0089] The bottom layer uses a Buck-Boost circuit to realize energy transfer between cells, and its balancing efficiency where D is a 0-1 adjustable duty cycle, which is adjusted in real time through adaptive PI control to ensure that the cell voltage difference is ≤5mV;

[0090] The top layer realizes energy scheduling between battery groups through a single inductor circuit, and optimizes the energy transfer path based on the objective function The constraint condition is for any battery group i, j, and an improved genetic algorithm is used to solve it:

[0091] The crossover probability of this algorithm is wherein is the population fitness variance, the mutation probability wherein is the individual fitness, is the population maximum fitness, by mixing the fitness function Simultaneously optimize energy loss and balanced speed, and retain the top 5% of elite individuals in each generation directly into the next generation, the convergence speed is improved by 40% compared with the standard genetic algorithm.

[0092] The adaptive thermal management module adopts model predictive control, and the temperature prediction model satisfies wherein is the state matrix, is the control matrix, is the liquid cooling flow rate / heating power, is the ambient temperature / charge-discharge power disturbance, and the control objective function is wherein, is the weight coefficient, and the constraint condition is The optimal control amount is solved by a sequential quadratic programming algorithm, and the paraffin and graphene composite phase change material filled in the gap between the batteries controls the temperature difference of the battery pack within ±1.5℃.

[0093] Embodiment 2

[0094] As Figure 2 shown, the embodiment discloses a battery health management method based on the system of embodiment 1, including an initialization phase, a running phase and a maintenance phase, and the specific process is as follows:

[0095] S1, in the initialization phase, when the battery is activated for the first time, an impedance response under different frequencies is collected by applying an alternating current pulse signal of 5Hz-1kHz to the battery through a pulse test system, and an individual characteristic library containing initial impedance spectrum, capacity, internal resistance and other parameters is constructed.

[0096] At the same time, the cloud intelligent hub generates a digital twin virtual battery based on the characteristic library, and the geometric parameters of the three-dimensional model, such as electrode thickness and diaphragm aperture, are completely consistent with the physical battery, and the initial parameters, such as production batch and activation date, are uploaded to the block chain storage module, and the first writing of the distributed ledger is completed.

[0097] S2, in the running phase, the implantable sensor network continuously collects multi-physical field parameters such as temperature, pressure, electrode deformation and electrolyte ion conductivity at an interval of 100ms, and transmits them to the edge node through the cross-shield communication module.

[0098] The edge node first filters invalid data through the MobileNetv3 anomaly detection model, denoises the valid data, extracts features, and uploads them to the cloud intelligent hub. The cloud calls the spatiotemporal double-flow Transformer module every hour, combines the real-time state of the digital twin virtual battery, updates the SOH and RUL prediction results, and solves the optimal balance strategy and thermal management parameters through the improved genetic algorithm and MPC algorithm, and issues them to the edge node.

[0099] The edge node drives the dynamic optimization execution unit according to the issued parameters: adjusts the Buck-Boost circuit duty cycle to achieve cell balancing, controls the liquid cooling valve opening degree and PTC heating power to adjust the temperature, and forms a closed-loop control of "perception-analysis-decision-execution".

[0100] S3, maintenance phase, when the system predicts RUL≤50 cycles, automatically generates a personalized life extension scheme, including limiting the charge and discharge depth and reducing the maximum charge and discharge rate, and pushes a maintenance reminder through the user terminal.

[0101] When the battery SOH drops to 0.8 retirement threshold, the system evaluates the remaining value based on the full life cycle data stored in the blockchain, such as cycle number, maximum temperature, and fault record:

[0102] For batteries with SOH between 0.6 and 0.8, recommend for use in energy storage power stations and other step utilization scenarios;

[0103] For batteries with SOH<0.6, guide to professional recycling agencies for material regeneration.

[0104] At the same time, the historical data stored in the blockchain can support traceability analysis and provide a basis for battery design optimization.

[0105] Through the above system and method, the battery health state can be accurately perceived, predicted and optimized, the battery cycle life is extended by more than 25% compared with the traditional BMS system, the thermal runaway early warning accuracy is improved to more than 98%, and the safety and economy of battery use are significantly improved.

[0106] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A system for battery health management, the system comprising: The system comprises an implantable sensor network, an edge and cloud collaborative computing platform, a multi-modal prediction model, and a dynamic optimization execution unit. The implantable sensor network is used to collect multi-physical field parameters inside the battery, including temperature, pressure, electrode deformation, and electrolyte ion conductivity. The edge and cloud collaborative computing platform is in communication connection with the implantable sensor network, used to process the collected multi-physical field parameters and construct a battery health state model. The multi-modal prediction model is deployed on the edge and cloud collaborative computing platform, and is used for predicting battery health indicators based on the processed multi-physical field parameters, the health indicators including a state of health (SOH) and a remaining useful life (RUL), and the state of health (SOH) calculation formula is ; wherein, , , is an aging coefficient, is a charge-discharge current, is a cell temperature, is a voltage change rate, is an initial capacity, is an integration time infinitesimal; The remaining useful life RUL calculation formula is ; wherein, is a retirement threshold; The dynamic optimization execution unit is in communication connection with the edge and cloud collaborative computing platform, used to perform battery equalization control and thermal management adjustment according to the predicted health indicators, forming a health management closed loop.

2. The system for battery health management of claim 1, wherein, The implantable sensor network comprises a miniaturized sensor array, a cross-shield communication module, and an energy harvesting unit. The miniaturized sensor array comprises a temperature and pressure integrated sensor, a distributed fiber Bragg grating strain sensor, and a micro-nano flow chip type ion conductivity sensor. The cross-shield communication module uses 2.4GHz carrier modulation technology and LoRa protocol to penetrate the battery metal shell to transmit data, with a transmission delay of ≤10ms. The energy harvesting unit generates induced current through battery charging and discharging voltage fluctuations to power the sensor, with an energy conversion efficiency of ≥65%.

3. The system for battery health management of claim 2, wherein, The center wavelength shift of the distributed fiber Bragg grating strain sensor satisfies the formula: ； Wherein, is the initial wavelength of the grating, is the photoelastic coefficient, is the strain value of the electrode material, is the thermo-optic coefficient, is the temperature change, the strain and temperature cross-sensitivity are separated by temperature compensation algorithm, and the deformation measurement accuracy reaches .

4. The system for battery health management of claim 2, wherein, The impedance spectrum of the micro-nano flow chip type ion conductivity sensor satisfies the formula: ； wherein, Zs is the sensor impedance, Rs is the solution resistance, Cdl is the double layer capacitance, Rct is the charge transfer resistance, f is the alternating current frequency, j is the imaginary unit, ion conductivity By Calculation, L is the flow path length, A is the cross-sectional area, measurement error <3%.

5. The system for battery health management of claim 1, wherein, The edge and cloud collaborative computing platform comprises an edge node and a cloud intelligent hub. The edge node is deployed with an abnormality detection model based on MobileNetv3 and a federal learning node. The cloud intelligent hub comprises a digital twin engine and a blockchain storage module, the blockchain storage module stores health indicators by adopting a PBFT consensus mechanism, the number of fault-tolerant nodes wherein n is the total number of nodes.

6. The system for battery health management of claim 1, wherein, The abnormality detection model filters invalid data in real time with an identification rate of ≥99.2%, and the federal learning node shares desensitized aging features with adjacent battery groups, realizes differential privacy protection by adding Laplace noise, and the model accuracy loss is ≤2%. ； wherein, is a spatial stream feature vector for the i-th sensor node, is a multilayer perceptron, is a feature concatenation operation, is a spatial coordinate of the i-th sensor node is a position encoding, is a strain feature for the i-th node, is a temperature feature for the i-th node, is a pressure feature for the i-th node; The multi-modal prediction model comprises a space-time double-flow Transformer module and a causal reasoning module. ； wherein, is a self-attention output feature matrix, is a query matrix, is a key matrix, is a value matrix, , , , is a query matrix weight parameter, is a time sequence feature matrix at time t, is a key matrix weight parameter, is a time sequence feature matrix at time t, is a value matrix weight parameter, is an adjustable time step, is a time sequence feature matrix at time t, is a time sequence feature matrix at time t, is a dimension of , is a normalization function; The space flow feature of the space-time double-flow Transformer module satisfies:

7. The system for battery health management of claim 1, wherein, The time flow self-attention calculation satisfies: wherein ; wherein, is the cell equalization efficiency, is the circuit output voltage, is the output current, is the equalization time, is the circuit input voltage, is the input current, and ; The top layer realizes inter-group energy scheduling through a single-inductor circuit, and is based on a target function The optimization path is constrained by , and is solved by an improved genetic algorithm wherein, is an inter-group energy transfer current is an energy transfer path resistance, is a scheduling time, is a summation operation, is a minimum operation.

8. The system for battery health management of claim 7, wherein, The causal reasoning module analyzes the causal relationship of multi-physical field parameters based on the Do-Calculus algorithm, with a thermal runaway early warning time of ≥30 minutes and an accuracy rate of ≥98%. ； wherein is a state matrix, is a control matrix, is a disturbance matrix, is the actual temperature of the battery at time t, is the liquid cooling flow rate / heating power, is a disturbance term; The control target function satisfies , the constraint condition is , and the battery temperature difference can be controlled within ±1.5℃ by using a sequence quadratic programming algorithm and a paraffin and graphene composite phase change material. wherein, is a control target function value, n is a prediction horizon length, is a prediction step, is a temperature deviation weight coefficient, is a control weight coefficient, is a target temperature.

9. A method of battery health management, the method comprising: The hierarchical active equalization system of the dynamic optimization execution unit comprises bottom-layer cell equalization and top-layer inter-group equalization. The adaptive thermal management module of the dynamic optimization execution unit uses model predictive control, and the temperature prediction model satisfies the formula: The battery health management system according to any one of claims 1-8 comprises the following steps: S1, initialization stage, obtain the initial impedance spectrum of the battery through pulse test, construct the individual feature library, synchronously generate the digital twin virtual battery, and complete the blockchain notarization. S2, running stage, edge node collects multi-physical field parameters every 100 ms, uploads to the cloud after noise reduction processing, the cloud updates the health prediction model every hour based on the SOH formula, and issues optimization parameters for solving the objective function to the edge node; S3, a maintenance stage, when the predicted remaining useful life RUL is less than or equal to 50 cycles, a personalized life extension solution is generated, and after the battery is retired, the residual value is evaluated based on the blockchain data to guide the echelon utilization, wherein the remaining useful life RUL is calculated by , a retirement threshold.

Images