Battery State of Charge Estimation Method Based on Embedded Fiber Optic Multiphysics Sensing

Abstract

This invention discloses a battery state of charge estimation method based on embedded fiber optic multiphysics sensing. A dual-FBG sensing unit is arranged inside the pouch cell: the first FBG is fitted with a quartz capillary to isolate the strain output temperature reference, and the second FBG outputs a thermomechanical coupling wavelength along the interlayer microchannel flexible contact with the negative electrode / separator interface. Voltage, current, and wavelength signals are simultaneously acquired. Decoupling is achieved by dynamically matching pre-calibrated mechanical sensitivity parameters according to the coarse SOC range and charge / discharge direction, yielding internal temperature and strain. A multiphysics sequence of voltage, current, temperature, and strain is constructed and input into an electrochemically constrained state estimation model. A learnable physics prior module outputs time-varying polarization parameters and recursively calculates the SOC. The lightweight model is deployed on an edge computing unit for real-time output. This invention achieves low-intrusion sensing, state-related decoupling, and physically consistent estimation with high accuracy and robustness, making it suitable for deployment on resource-constrained platforms.

Description

Technical Field

[0001] This invention relates to the fields of battery management and fiber optic sensing technology, specifically to a method for estimating the state of charge of a battery based on embedded fiber optic multiphysics sensing. Background Technology

[0002] With the rapid development of new energy vehicles, energy storage systems, electric forklifts, and portable electrical devices, the safety, reliability, and energy utilization efficiency of power batteries and energy storage batteries under complex operating conditions are receiving increasing attention. State of Charge (SOC) is a crucial parameter characterizing the remaining usable capacity of a battery and is a vital basis for battery management systems to predict range, control charge and discharge, allocate energy, provide safety protection, and manage battery life. Accurate, real-time, and stable estimation of SOC is of great significance for improving the operational safety and management efficiency of battery systems.

[0003] Currently, battery management systems (BMS) typically estimate State of Charge (SOC) based on externally measurable signals such as battery terminal voltage, current, and surface temperature. Common methods include coulomb counting, open-circuit voltage, equivalent circuit modeling, Kalman filtering, and data-driven modeling. While these methods can achieve SOC estimation to some extent, they primarily rely on external electrical or surface temperature signals and struggle to directly reflect internal multi-physics information such as temperature changes, electrode expansion and contraction, interlayer pressure changes, and polarization state evolution. Under dynamic loads, high and low temperature environments, charge-discharge switching, and long-term operation, a significant nonlinearity and hysteresis relationship exists between the battery's external port response and its actual internal state, easily leading to increased SOC estimation errors and decreased cross-condition generalization ability.

[0004] To improve the observability of the internal state of batteries, research has begun to incorporate fiber Bragg grating (FBG) sensors into battery state monitoring and SOC estimation. FBG sensors are characterized by their small size, strong resistance to electromagnetic interference, high sensitivity, and good embeddability, enabling them to acquire signals such as temperature, strain, and pressure from inside or on the surface of the battery. However, some existing solutions primarily focus on acquiring temperature or strain signals from the fiber optic sensors, neglecting the low-intrusion placement of the fiber optic sensing unit within the pouch cell, mechanical coupling control, packaging security, and the reliability of the lead-out path. Inappropriate placement or packaging of the fiber optic sensor may affect the integrity of the electrode structure, the safety of the separator, and the cycle stability of the battery.

[0005] Furthermore, the wavelength response of fiber optic sensors is typically affected by both temperature and mechanical strain, thus requiring temperature compensation or thermo-mechanical decoupling before use for SOC estimation. Most existing methods employ fixed temperature or mechanical sensitivity parameters, assuming the battery's internal mechanical response remains consistent across different SOC stages and charge / discharge directions. However, in reality, the degree of electrode expansion, interlayer pressure state, and polarization behavior differ across low, medium, and high SOC ranges. Furthermore, lithium insertion expansion during charging and lithium desorption contraction during discharging may exhibit asymmetry and path dependence. Therefore, using a fixed mechanical sensitivity for thermo-mechanical decoupling can easily introduce systematic decoupling errors across different SOC ranges or during charge / discharge transitions, reducing the reliability of the input features for subsequent SOC estimation models.

[0006] While some existing fiber-assisted SOC estimation methods incorporate decoupled internal temperature, strain, or pressure information, they typically treat these signals as ordinary input features and directly feed them into neural networks or other data-driven models for SOC regression. This approach fails to adequately consider battery polarization kinetics, the direction of SOC recursion, the continuity of the time constant, and the influence of temperature and mechanical response on polarization parameters. Under complex dynamic conditions or across temperature ranges, purely data-driven models may exhibit problems such as inconsistencies between the direction of SOC change and the current direction, unstable polarization state characterization, and local fluctuations in prediction results. Furthermore, many existing methods remain at the offline experimental analysis stage, failing to fully consider the requirements of actual battery management system operation regarding model size, computational complexity, recursion efficiency, and edge deployment capabilities.

[0007] Therefore, a new battery SOC estimation method is urgently needed, which can achieve in-situ sensing of the battery's internal temperature and mechanical response while minimizing the impact on the internal structure and safety of the pouch battery; can perform state-related decoupling of the fiber thermo-mechanical coupling signal according to the battery's current SOC range and charging / discharging direction, reducing the decoupling error caused by the fixed sensitivity assumption; and can combine the decoupled internal multi-physics information with the electrochemically constrained state estimation model, thereby improving the accuracy, robustness, and physical consistency of battery SOC estimation under dynamic operating conditions, temperature changes, and online deployment scenarios. Summary of the Invention

[0008] This invention addresses the needs of existing technologies by providing a battery state of charge estimation method based on embedded fiber optic multiphysics sensing. This method solves the problems of high invasiveness of fiber optic embedding in pouch cells, fixed thermo-mechanical decoupling parameters, lack of physical consistency in SOC estimation models, and difficulty in edge deployment.

[0009] A battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing includes the following steps: Step 1: Arrange dual FBG sensing units inside the pouch battery, which are used to output a temperature reference wavelength signal and a thermal signal simultaneously affected by temperature and mechanical strain. Mechanically coupled wavelength signals; Step 2: Synchronously acquire the temperature reference wavelength signal and thermal... Mechanically coupled wavelength signals, as well as voltage and current signals from the pouch battery, and using the coulomb counting method based on the current signal to calculate the coarse SOC value at the current sampling moment; Step 3: Based on the pre-calibrated set of mechanical sensitivity parameters corresponding to different SOC ranges and different charge / discharge directions, select the mechanical sensitivity parameter that matches the SOC range and charge / discharge direction of the current coarse SOC value, and utilize the temperature reference wavelength signal and thermal... Mechanically coupled wavelength signal thermal Mechanical decoupling yields the internal temperature signal and mechanical strain change signal at the current sampling moment; Step 4: Perform time-series alignment and normalization on the voltage signal, current signal, internal temperature signal, and mechanical strain change signal to construct a multiphysics sequence; Step 5: Establish an electrochemically constrained multiphysics state estimation model, using the multiphysics sequence, the SOC estimate at the previous sampling time, and the hidden state at the previous sampling time as inputs, to perform recursive calculations and obtain the final SOC estimate at the current sampling time. Step 6: Deploy the trained multiphysics state estimation model to the edge computing unit and output the SOC estimation results in real time during actual operation.

[0010] Further, the SOC interval mentioned in step 3 is divided into: low SOC interval (0%–30%), medium SOC interval (30%–70%), and high SOC interval (70%–100%); the current at the current sampling moment is denoted as... The charging / discharging direction is determined by setting a current threshold. ,when The charging direction is determined at that time. The direction of discharge is determined by the time. The direction state continues from the previous sampling time.

[0011] Furthermore, the mechanical sensitivity parameters mentioned in step 3 are pre-calibrated in the following way: within each SOC interval, multiple sets of charging pulses and discharging pulses are applied respectively, the temperature component in the thermo-mechanical coupling wavelength signal is compensated by the temperature reference wavelength signal, and the equivalent reference strain change is obtained by an external reference strain measuring device set on the outer surface of the pouch battery, and the mechanical sensitivity parameters in the charging direction and discharging direction are calculated respectively.

[0012] Furthermore, the multiphysics state estimation model in step 5 includes a learnable physical prior module. The inputs of this learnable physical prior module include the current signal, voltage signal, internal temperature signal, mechanical strain change signal at the current sampling time, and the SOC estimate at the previous sampling time. The outputs are polarization resistance, polarization capacitance, and polarization time constant. The outputs of the polarization resistance and polarization capacitance use the Softplus activation function to ensure that their values ​​are positive.

[0013] Furthermore, the recursive calculation in step 5 includes: A physical constraint state transition matrix is ​​constructed based on the polarization resistance, polarization capacitance, and polarization time constant, and then superimposed with the data-driven state transition matrix to obtain the state transition matrix at the current moment, thereby calculating the original hidden state at the current moment. The SOC change rate is calculated based on the original hidden state, and the original SOC estimate is obtained by integrating the SOC change rate. Based on the uncertainty of the current estimation result estimated from the original hidden state, a correction amount negatively correlated with the uncertainty is generated, and this correction amount is superimposed on the original SOC estimate to obtain the final SOC estimate at the current sampling time.

[0014] Furthermore, before integrating the SOC change rate, the amplitude of the SOC change rate is constrained to ensure that it meets the physical achievable range under actual battery charging and discharging conditions.

[0015] Furthermore, before step 6, a training step for the multiphysics state estimation model is included: using battery operating data under multiple temperature conditions, dividing the training set, validation set, and test set in a battery-level isolation manner, using a joint loss function to jointly constrain the SOC estimation error, polarization dynamic consistency, time constant smoothness, and consistency between the SOC change direction and the current direction, using the Adam optimizer to update parameters, and setting an early stopping strategy.

[0016] Furthermore, the edge computing unit mentioned in step 6 is an embedded platform; the deployed model size is no greater than 0.1MB, the number of parameters is no greater than 0.03M, and the single-batch inference latency is no greater than 3ms.

[0017] Further, the dual FBG sensing unit in step 1 includes a first FBG sensing unit and a second FBG sensing unit. A quartz capillary tube is sleeved on the outer side of the gate region of the first FBG sensing unit to isolate mechanical strain and is used to output the temperature reference wavelength signal. The first FBG sensing unit is disposed at the adjacent interface between the separator and the positive electrode along the reserved first interlayer microchannel. The gate region of the second FBG sensing unit is not sleeved with a capillary tube and is disposed at the adjacent interface between the negative electrode and the separator along the reserved second interlayer microchannel. Flexible positioning points are provided on both sides of its gate region to keep the middle part of the gate region in flexible contact with the negative electrode and the separator and is used to output the thermo-mechanical coupling wavelength signal.

[0018] Furthermore, the flexible positioning points on both sides of the grid region of the second FBG sensing unit are locally fixed with an electrolyte-resistant adhesive, and the middle of the grid region maintains flexible contact with the negative electrode and the separator without rigid constraints; the ordinary optical fiber segments of the first FBG sensing unit and the second FBG sensing unit are integrally and continuously led out from the inside of the soft pack battery to the outside, and the lead-out position is provided with a sealing structure.

[0019] The beneficial effects of this invention are: By employing a low-invasive dual-FBG differential arrangement (the first FBG is fully isolated by a quartz capillary tube as a temperature reference, and the second FBG flexibly contacts the negative electrode / separator interface along the interlayer microchannel), in-situ sensing of the battery's internal temperature and mechanical strain is achieved without damaging the electrodes and separator. By pre-calibrating mechanical sensitivity parameters for different SOC ranges and charge / discharge directions and dynamically matching them during online operation, the systematic error of fixed-sensitivity decoupling is overcome, improving the accuracy and continuity of the strain signal. The decoupled multi-physics field signal is input into an electrochemically constrained state estimation model, which outputs time-varying polarization parameters through a learnable physical prior module, and constructs a physical... The mixed state transition matrix enhances the physical consistency and cross-condition generalization ability of SOC estimation; the lightweight model design (volume ≤ 0.1MB, number of parameters ≤ 0.03M, inference latency ≤ 3ms) is suitable for real-time deployment on resource-constrained BMS edge computing units. Attached Figure Description

[0020] Figure 1 The flowchart of this invention Figure 2 This is a diagram showing the arrangement of the dual FBG sensing units within the inner layers of the pouch battery in this invention. Figure 3 This is an internal cross-sectional view of the soft-pack battery in the thickness direction of the present invention; Figure 4 This is a partially enlarged view of the FBG sensing unit extending from the edge of the pouch battery in this invention; Figure 5The SOC estimation results for the B107 battery under different input signals. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings. Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The directional terms such as left, center, right, top, and bottom in the embodiments of the present invention are only relative concepts or referenced to the normal use state of the product, and should not be considered restrictive.

[0022] A battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing includes the following steps: Step 1: Low-intrusion dual FBG sensing unit arrangement inside the pouch battery like Figures 2 to 4 As shown, a dual FBG sensing unit is arranged inside the soft-pack battery 1, which is used to output a temperature reference wavelength signal and a thermo-mechanical coupling wavelength signal that is simultaneously affected by temperature and mechanical strain. A quartz capillary tube 51 is sleeved on the outer side of the grid area of ​​the first FBG sensing unit to completely isolate mechanical strain and is used to output the temperature reference wavelength signal. Specifically, the first FBG sensing unit 5 is disposed in the recessed area 4 of the non-tab side edge of the electrode group 2 of the soft-pack battery 1 (with the side where the tab 3 is located as the near tab side and the non-tab side as the side away from the tab), 3mm to 8mm inward from the edge of the electrode group. In the thickness direction, the first FBG sensing unit 5 is disposed at the adjacent interface between the separator 7 and the positive electrode 9 along the reserved first interlayer microchannel 52. The first interlayer microchannel 52 is located in the corresponding separator 7 and is disposed along the length direction of the battery. The inner diameter of the quartz capillary tube 51 is larger than the outer diameter of the first FBG sensing unit, and the outer diameter is smaller than the thickness that can be accommodated between adjacent layers to avoid local pressure damage to the separator and the electrode. The grid area is arranged along the length direction of the battery, with the center located in the middle area of ​​the length direction, avoiding the tab welding area, the heat-sealed edge 12 and the electrode cutting edge burr area. The second FBG sensing unit 6 does not have a capillary tube fitted over its gate region, and is positioned near the interface between the negative electrode 8 and the diaphragm 7 along a reserved second interlayer microchannel. Flexible positioning points B are provided on both sides of its gate region to maintain flexible contact between the center of the gate region and the negative electrode 8 and the diaphragm 7, for outputting the thermo-mechanical coupling wavelength signal. Specifically, the second FBG sensing unit and the first FBG sensing unit are located on the same side, with a center-to-center distance A between their gate regions of 2mm to 10mm. Before the electrode assembly is stacked, a second [unclear - possibly a reference to a specific electrode assembly] is reserved in the recessed area of ​​the non-electrode side edge. Interlayer microchannels, the second interlayer microchannel 62 is located in the corresponding separator 7 and is arranged along the length of the battery, with a width of 1.2 to 2.0 times the outer diameter of the second FBG sensing unit; the second FBG sensing unit is laid along the second interlayer microchannel 62 and covered with a thin layer of polyimide insulating layer 61; its grid area is not rigidly bonded in the whole section, but only flexible positioning points B are set on the ordinary optical fiber segments 11 on both sides of the grid area, and the electrolyte-resistant adhesive 15 is used for local fixation, so that the middle of the grid area maintains flexible contact with the negative electrode 8 and the separator 7 without rigid constraints; Sealed lead-out structure: The ordinary optical fiber segments (i.e., the non-gate area optical fiber segments that continuously extend outward from the FBG gate area) of the first FBG sensing unit and the second FBG sensing unit are integrally and continuously led out from the inside of the pouch battery to the outside, and a sealing structure is provided at the lead-out position; specifically, at the position where the ordinary optical fiber segment passes through the pouch sealing edge, an electrolyte-resistant insulating transition layer 14 is pre-coated around its outer periphery, and a heat-sealing adhesive layer or an electrolyte-resistant sealing adhesive layer formed by an electrolyte-resistant adhesive 15 is provided on the outside of the insulating transition layer 14; the pouch aluminum-plastic film is heat-sealed, so that the heat-sealing layer, the heat-sealing adhesive layer (or the electrolyte-resistant sealing adhesive layer), the insulating transition layer and the outer wall of the optical fiber form a continuous bonding interface, constituting an initial liquid-sealed structure; wherein, the ordinary optical fiber segment is arranged in a gently bending path inside the electrode assembly, and the bending radius is preferably not less than 5mm, so as to reduce the risk of optical fiber bending damage, packaging stress concentration and fatigue fracture caused by the expansion and contraction of the electrode assembly during charge and discharge cycles. After heat sealing, the pouch cell is injected with electrolyte, allowed to stand and wet, and then formed. After formation, it is encapsulated again to enhance the reliability of the liquid seal. A flexible reinforcing adhesive layer 16 or a sheath structure is set on the outside of the external optical fiber outlet 13 to disperse the tensile, vibration, and bending stresses on the external optical fiber and prevent the stress from being directly transmitted to the liquid seal interface at the optical fiber outlet 13. As a result, ordinary optical fiber segments can be continuously led out from the inside of the battery without damaging the integrity of the pouch cell encapsulation, and electrolyte leakage prevention and optical fiber mechanical protection are achieved. Step 2: Synchronous acquisition of multi-source signals and coarse SOC calculation The temperature reference wavelength signal was acquired synchronously. (i.e., the wavelength signal of the first FBG sensing unit), heat Mechanically coupled wavelength signal (i.e., the wavelength signal of the second FBG sensing unit) and the voltage signal of the pouch battery. and current signal ; In this embodiment, an FBG demodulation system is used to read the wavelength, while a battery testing system or battery management system simultaneously acquires voltage and current. The FBG demodulation system includes a light source, an optical fiber coupler, a spectral demodulation module, and a data acquisition module. Light emitted from the light source enters the first and second FBG sensing units, which then reflect light signals of corresponding wavelengths. The spectral demodulation module identifies the peak position of the reflected spectrum and outputs the corresponding wavelength data. In this embodiment, a sampling frequency of 1Hz can meet the requirements for conventional charge / discharge SOC estimation. Battery voltage and current signals are synchronously acquired through a battery testing system or battery management system to obtain the voltage and current signals at the current sampling moment. In this embodiment, the charging current direction is defined as positive, and the discharging direction as negative. All signals are synchronized with a unified timestamp. When the sampling frequencies of the FBG demodulation system and the battery testing system are inconsistent, all signals are resampled to the same time interval. For missing sampling points, linear interpolation is used to fill in the gaps. For obviously abnormal isolated jump points, the median or mean value within adjacent time windows is used as a substitute. After time synchronization, the original observation vector at each sampling moment is obtained. ; Before the formal data acquisition, the pouch cell was adjusted to a preset initial state: 25℃, 50% SOC, and open-circuit rest for 30-60 minutes; under this initial state, the initial center wavelength of the first FBG sensing unit was recorded. The initial center wavelength of the second FBG sensing unit Initial internal temperature The initial state of charge (SOC0) can be obtained through capacity calibration, open-circuit voltage calibration, or coulomb counting initialization; the initial center wavelength... and initial center wavelength Used for subsequent calculation of wavelength change: The coarse SOC value at the current sampling moment is calculated using the coulomb counting method based on the current signal. : in, This is the coarse SOC value from the previous sampling time. The sampling period is (s). Battery capacity refers to the rated or calibrated capacity of the battery. Let Ah represent the sampling period. When represented by s, if If A·s units are used, then 3600 in the above formula can be omitted; the coarse SOC value is only used to determine the current SOC interval and is not used as the final SOC estimation result. Step 3: State-dependent thermo-mechanical decoupling Based on a pre-calibrated set of mechanical sensitivity parameters corresponding to different SOC ranges and charging / discharging directions, a mechanical sensitivity parameter matching the SOC range and charging / discharging direction of the current coarse SOC value is selected, and the temperature reference wavelength signal is used. and heat Mechanically coupled wavelength signal Perform heat Mechanical decoupling yields the internal temperature signal at the current sampling moment. and mechanical strain change signal ; Step 3.1: Temperature sensitivity calibration The packaged pouch cells were placed in a constant temperature chamber and subjected to temperature step calibration (15℃, 25℃, 35℃, 45℃) under open-circuit static conditions, with each temperature point held for 30 to 60 minutes; the wavelength change of the first FBG sensing unit at different temperature points was recorded. and temperature change The temperature sensitivity of the first FBG sensing unit was obtained by linear fitting of the relationship. ( Similarly, under conditions of no significant mechanical disturbance, the wavelength changes of the second FBG sensing unit at different temperature points were recorded. and temperature change The temperature sensitivity of the second FBG sensing unit was obtained by linear fitting of the relationship. ( ); 3.2: SOC Interval Division The battery SOC range is divided into low SOC, medium SOC and high SOC ranges; the low SOC range is 0% to 30%, the medium SOC range is 30% to 70%, and the high SOC range is 70% to 100%; the coarse SOC value at the current sampling time is used to determine the current SOC range. 3.3: Directional Mechanical Sensitivity Calibration Within each SOC range, the mechanical sensitivity parameters for the charging and discharging directions are calibrated separately; taking the first... jTaking a SOC range as an example: Adjust the soft pack battery to the SOC anchor point in the middle of the SOC range, let it stand for 10 minutes, and then apply multiple sets of charging pulses and discharging pulses respectively (the charging pulses and discharging pulses can use the same or different current amplitudes, the current amplitude is 0.2C to 1C, the pulse duration is 30s to 120s, and the resting time is 120s to 600s). During the charging pulse, the wavelength change of the first FBG sensing unit is read. Wavelength change of the second FBG sensing unit Calculate the temperature change in the charging direction. : And the mechanically correlated wavelength change under the charging direction is obtained. : During the discharge pulse, the wavelength change of the first FBG sensing unit is read. Wavelength change of the second FBG sensing unit Calculate the temperature change in the direction of discharge. : And the mechanically correlated wavelength change under the discharge direction is obtained. : During the calibration phase, an external reference strain measurement device (such as a resistance strain gauge) is set on the projection area corresponding to the second FBG sensing unit on the outer surface of the pouch battery. The equivalent reference strain change obtained by the external reference strain measurement device (i.e., a standard strain measurement device, which is used to obtain the external equivalent reference strain change in this area during the charging and discharging pulses, and is not embedded inside the pouch battery) under charging and discharging pulses is acquired. and Then the first j The mechanical sensitivity parameters for charging direction and discharging direction for each SOC range are as follows: 3.4: Determining the direction of charging and discharging Let the current at the current sampling time be... Set current threshold The charging and discharging direction is determined in the following way: when The charging direction is determined at that time; when The direction of discharge is determined at that time; when The direction state continues from the previous sampling time; That is, direction variable dir t satisfy: in, Indicates charging direction ， Indicates the direction of discharge. This indicates the directional state that continues from the previous sampling time under low current or near-static conditions; 3.5: Calculation of Internal Temperature Signal The internal temperature change at the current sampling moment is calculated based on the temperature reference wavelength signal output by the first FBG sensing unit. Further, the internal temperature signal was obtained: in, This represents the local internal temperature signal characterized by the location of the first FBG sensing unit.

[0023] 3.6: Calculation of Mechanical Strain Change Signal Based on the current crude SOC value z t Determine the current SOC range based on the current current. I t Determine the current direction variable dir t And select the corresponding mechanical sensitivity parameters: Subsequently, the temperature component from the second FBG sensing unit is subtracted to obtain the mechanical strain change signal at the current sampling time. : In this embodiment, the corresponding mechanical sensitivity parameters are selected based on the coarse SOC range and current direction, and a smooth band is set at the boundary of the SOC range. This reduces the systematic decoupling error caused by the fixed sensitivity and avoids sudden changes in strain signal when the range is switched. The resulting internal mechanical strain signal is more continuous and better matches the actual thermo-mechanical response mechanism of the battery, and can also provide more reliable physical characteristics for the subsequent SOC estimation model. The mechanical sensitivity parameters can be determined by fitting multiple sets of charge and discharge pulse data, thereby improving the decoupling consistency and parameter stability under bidirectional charging and discharging conditions. Step 4: Construction of Multiphysics Sequence The voltage signal Current signal Internal temperature signal and mechanical strain change signal Temporal alignment (interpolation, elimination, or smoothing, such as moving average filtering) is performed to obtain a temporally consistent multiphysics sequence. To reduce the impact of differences in the dimensions of different physical quantities on model training and inference, min-max normalization is used to map each feature to the interval [0,1]. The normalization parameters are determined only based on the statistical values ​​of the training set. During the model training phase, a sliding window is used to construct continuous time series samples: using data from the past L=120 consecutive time steps as an input sample, the input data X∈R is constructed. L×F Where L represents the time step and F represents the feature dimension, which includes at least current, voltage, internal temperature, and strain, i.e.: The input features can also include the SOC estimate from the previous sampling time, with a time step L of 120. During the online inference phase, the edge computing unit performs recursive calculations based on the input data at the current sampling time, the SOC estimate from the previous sampling time, and the hidden state from the previous sampling time. Step 5: Electrochemically Constrained Multiphysics State Estimation Model A multi-physics state estimation model for electrochemical confinement is established. Using the multi-physics sequence, the SOC estimate from the previous sampling time, and the hidden state from the previous sampling time as inputs, a recursive calculation is performed to obtain the final SOC estimate for the current sampling time. Specifically: 5.1: Input Mapping The input data constructed in step 4 is input into the input mapping module (fully connected layer, input dimension F, output dimension D=32). To enhance the local temporal pattern extraction capability, a one-dimensional convolutional embedding layer is set, and the hidden state representation is obtained after mapping. In the model training stage, the input data is the continuous time series sample; in the online inference stage, the input data is the voltage signal, current signal, internal temperature signal, and mechanical strain change signal at the current sampling time. 5.2: Learnable Physics Prior Modules The learnable physical prior module is a trainable submodule in the multiphysics state estimation model. It dynamically generates physical prior parameters related to the battery polarization state based on the multiphysics observation information at the current sampling time, providing physical constraints for the subsequent state transition process. Specifically, the learnable physical prior module includes an input feature concatenation layer, a feature mapping layer, a hidden feature extraction layer, and a parameter output layer connected in sequence. The input feature concatenation layer receives the current signal at the current sampling time. Voltage signal Internal temperature signal Mechanical strain change signal and the SOC estimate at the previous sampling time. The above input features are then concatenated to form an input vector; The feature mapping layer consists of a fully connected layer and is used to perform feature mapping on the input vector, converting the original observation information into a feature representation suitable for characterizing the internal state changes of the battery; the hidden feature extraction layer includes a fully connected hidden layer and a corresponding nonlinear activation function, which is used to further extract hidden feature vectors that reflect the battery polarization behavior, temperature change characteristics and strain response characteristics. The parameter output layer constructs a polarization parameter output branch based on the hidden feature vector, which is used to output the polarization resistance at the current sampling time. and polarization capacitor To ensure that the output parameters meet the physical constraints of the battery equivalent circuit model, the polarization resistor... and polarization capacitor The outputs are all constrained with positive values ​​using the Softplus activation function, and the polarization time constant at the current sampling time is further calculated. ; 5.3: State Space Recursion Before performing state-space recursion, the initial variables involved in the recursive calculation are initialized. The initial value of the SOC estimate at the initial sampling time is determined by at least one of the experimental calibration value, the static open-circuit voltage calibration value, the historical stored value, or the preset initial value; the initial hidden state vector is set to a zero vector, or obtained from the input data at the initial sampling time through the input mapping module; the initial polarization state is set to zero, or determined based on the voltage deviation in the initial static state; the initial charging and discharging direction variable is determined based on the current signal at the initial sampling time. When the absolute value of the initial current signal is not greater than the preset current threshold, the initial charging and discharging direction variable is set to the previous effective direction state or the preset direction state. A physical constraint state transition matrix is ​​constructed based on the polarization resistance, polarization capacitance, and polarization time constant. This matrix is ​​then linearly superimposed with the data-driven state transition matrix to obtain the current state transition matrix. Then calculate the original hidden state at the current time. : in, This indicates the hidden state at the previous sampling time. This represents the input mapping matrix at the current sampling time. This represents the input data at the current sampling time; wherein, the state transition matrix... It includes physical constraint information related to polarization dynamics, enabling the model to have both temporal modeling capability and physical consistency during the recursive process; the first dimension or preset dimension in the hidden state vector is defined as the polarization state, which is constrained by discretized RC polarization dynamics; the remaining hidden states represent higher-order dynamics and residual nonlinear behavior. 5.4: Rate of Change and Integral of SOC The SOC change rate is calculated based on the original hidden state. An amplitude constraint is applied to this SOC change rate (ensuring it meets the physical reach of the battery under actual charge and discharge conditions). Then, the SOC change rate is integrated to obtain the original SOC estimate. : in, This represents the SOC estimate at the previous sampling time. This represents the rate of change of SOC at the current sampling time; 5.5: Adaptive Uncertainty Correction Estimate the uncertainty of the SOC estimation result based on the original hidden state at the current sampling time, and adjust the original SOC estimate based on the uncertainty. Adaptive correction is performed. Specifically, the original hidden state is input into the uncertainty output branch and the error output branch respectively to obtain the uncertainty factor at the current sampling time. and potential error signals : in, A linear mapping function representing an uncertain output branch. The linear mapping function represents the error output branch; the Sigmoid function is used to account for uncertainty factors. The constraint is in the range of 0 to 1. A larger value indicates higher uncertainty in the current SOC estimation result; the Tanh function is used to limit the potential error signal. The amplitude range; Based on uncertainty factors Constructing adaptive correction gain The adaptive correction gain With uncertainty factors Negative correlation: in, These are preset correction coefficients used to limit the correction magnitude; when the uncertainty factor... When the value is large, the adaptive correction gain Reduce to suppress overcorrection under unreliable conditions; when the uncertainty factor When the value is small, the adaptive correction gain Increase the size to improve the ability to correct for potential estimation errors; Based on adaptive correction gain and potential error signals Generate the SOC correction value at the current sampling time. : SOC correction amount Superimposed on the original SOC estimate The final SOC estimate at the current sampling time is obtained. : in, This represents the original SOC estimate at the current sampling time; This represents the final SOC estimate after adaptive uncertainty correction; 5.6: Model Training Battery operating data collected under multiple temperature conditions (15℃, 25℃, 35℃, 45℃) were used to divide the training, validation, and test sets using battery-level isolation; a joint loss function was employed. Among them, the supervision error term Polarization consistency loss term is used to constrain the error between the predicted SOC and the calibrated SOC. Time constant smoothing loss term used to constrain polarization dynamic consistency The sign consistency loss term is used to constrain the smoothness of the polarization time constant variation. Used to constrain the direction of SOC change to be consistent with the direction of current; =0.1, =0.01, =0.1; The Adam optimizer is used to update the model parameters with a learning rate of 0.001, a batch size of 64, and 200 training rounds. An early stopping strategy is set (the model is terminated early if the validation set loss does not decrease for 20 consecutive rounds). After training, the optimal model parameters, normalized parameters, and mechanical sensitivity parameter sets for each SOC interval and direction are written into the storage unit for subsequent online deployment. Step 6: Deploy the trained multiphysics state estimation model to an edge computing unit (embedded platform, such as Jetson Orin Nano). The deployed model size should not exceed 0.1MB, the number of parameters should not exceed 0.03M, and the single-batch inference latency should not exceed 3ms. In actual operation, the edge computing unit receives voltage signals, current signals, temperature reference wavelength signals from the first FBG sensing unit, and thermo-mechanical coupling wavelength signals from the second FBG sensing unit in real time according to the sampling frequency. It then sequentially executes step 2, which involves synchronous acquisition of multi-source signals and coarse SOC calculation; step 3, which involves state-related thermo-mechanical decoupling; and step 4, which involves constructing a multi-physics sequence. Based on the multi-physics state estimation model in step 5, it performs recursive calculations and outputs the SOC estimation result at the current sampling time in real time.

[0024] Experimental verification Verification experiments were conducted on the method of this invention, and the results are shown in Table 1. B107, B207, B307, and B407 represent the test battery numbers corresponding to different temperature groups. The estimation error indices for multiple experimental samples are listed, including MAE, RMSE, and MAXE. In the multiphysics input gain verification, only voltage was used. V and current I The input scheme is used as the baseline scheme, and the internal temperature signal will be added. T Strain signals and the original wavelength signal W 1. W The input scheme of option 2 was used as a comparison scheme. Experimental results show that, after fusing internal multiphysics information, compared with the baseline scheme using only voltage and current, the RMSE is reduced by approximately 46%–62%, the MAE is reduced by approximately 25%–66%, and the maximum error remains within the range of approximately 2.5%; combined with Figure 5 The SOC estimation results of B107 show that by introducing decoupled internal temperature and strain signals, the observability of the battery's internal state can be significantly improved, thereby enhancing the accuracy and robustness of SOC estimation.

[0025] Table 1: Evaluation Indicators for SOC Estimation under Different Input Signals After training the model on 25℃ data, it was directly tested at 15℃, 35℃, and 45℃ to verify the model's generalization ability to temperature changes. Cross-temperature generalization experiments (Table 2) show that the proposed scheme maintains RMSE between 0.56% and 0.73%, MAXE between 1.7% and 2.9%, and MAE between 0.44% and 0.59% across temperature scenarios. This indicates that the proposed scheme can maintain high SOC estimation accuracy even when polarization parameters drift due to temperature changes.

[0026] Table 2: Evaluation Indicators for SOC Estimation under Different Input Signals In edge deployment validation, the trained model was deployed to an embedded platform. Experimental results show that the model size is approximately 0.09 MB, the number of parameters is approximately 0.020 M, the FLOPs are approximately 0.02 M, and the single-batch inference latency is approximately 2.93 ms. These results demonstrate that the proposed solution maintains high SOC estimation accuracy while featuring a small model size and low computational overhead, making it suitable for real-time deployment on resource-constrained BMS platforms.

[0027] 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 illustrative of the principles of 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 present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing, characterized in that, Includes the following steps: Step 1: Arrange dual FBG sensing units inside the pouch battery, which are used to output a temperature reference wavelength signal and a thermal signal simultaneously affected by temperature and mechanical strain. Mechanically coupled wavelength signals; Step 2: Synchronously acquire the temperature reference wavelength signal and thermal... Mechanically coupled wavelength signals, as well as voltage and current signals from the pouch battery, and using the coulomb counting method based on the current signal to calculate the coarse SOC value at the current sampling moment; Step 3: Based on the pre-calibrated set of mechanical sensitivity parameters corresponding to different SOC ranges and different charge / discharge directions, select the mechanical sensitivity parameter that matches the SOC range and charge / discharge direction of the current coarse SOC value, and utilize the temperature reference wavelength signal and thermal... Mechanically coupled wavelength signal thermal Mechanical decoupling yields the internal temperature signal and mechanical strain change signal at the current sampling moment; Step 4: Perform time-series alignment and normalization on the voltage signal, current signal, internal temperature signal, and mechanical strain change signal to construct a multiphysics sequence; Step 5: Establish an electrochemically constrained multiphysics state estimation model, using the multiphysics sequence, the SOC estimate at the previous sampling time, and the hidden state at the previous sampling time as inputs, to perform recursive calculations and obtain the final SOC estimate at the current sampling time. Step 6: Deploy the trained multiphysics state estimation model to the edge computing unit and output the SOC estimation results in real time during actual operation.

2. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 1, characterized in that: The SOC intervals mentioned in step 3 are divided into: low SOC interval (0%–30%), medium SOC interval (30%–70%), and high SOC interval (70%–100%); the current at the current sampling time is denoted as... The charging / discharging direction is determined by setting a current threshold. ,when The charging direction is determined at that time. The direction of discharge is determined at that time. The direction state continues from the previous sampling time.

3. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 1 or 2, characterized in that: The mechanical sensitivity parameters mentioned in step 3 are pre-calibrated in the following way: In each SOC interval, multiple sets of charging pulses and discharging pulses are applied respectively, the temperature component in the thermo-mechanical coupling wavelength signal is compensated by the temperature reference wavelength signal, and the equivalent reference strain change is obtained by the external reference strain measurement device set on the outer surface of the pouch battery, and the mechanical sensitivity parameters in the charging direction and the discharging direction are calculated respectively.

4. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 1, characterized in that: The multiphysics state estimation model in step 5 includes a learnable physical prior module. The inputs of the learnable physical prior module include the current signal, voltage signal, internal temperature signal, mechanical strain change signal at the current sampling time, and the SOC estimate at the previous sampling time. The outputs are polarization resistance, polarization capacitance, and polarization time constant. The outputs of the polarization resistance and polarization capacitance use the Softplus activation function to ensure that their values ​​are positive.

5. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 4, characterized in that: The recursive calculation in step 5 includes: A physical constraint state transition matrix is ​​constructed based on the polarization resistance, polarization capacitance, and polarization time constant, and then superimposed with the data-driven state transition matrix to obtain the state transition matrix at the current moment, thereby calculating the original hidden state at the current moment. The SOC change rate is calculated based on the original hidden state, and the original SOC estimate is obtained by integrating the SOC change rate. Based on the uncertainty of the current estimation result estimated from the original hidden state, a correction amount negatively correlated with the uncertainty is generated, and this correction amount is superimposed on the original SOC estimate to obtain the final SOC estimate at the current sampling time.

6. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 5, characterized in that: Before integrating the SOC change rate, the amplitude of the SOC change rate is first constrained to ensure that it meets the physical achievable range under actual battery charging and discharging conditions.

7. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 1, characterized in that: Before step 6, a training step for the multiphysics state estimation model is also included: using battery operating data under multiple temperature conditions, dividing the training set, validation set and test set in a battery-level isolation manner, using a joint loss function to jointly constrain the SOC estimation error, polarization dynamic consistency, time constant smoothness and consistency between the SOC change direction and the current direction, using the Adam optimizer to update parameters, and setting an early stop strategy.

8. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 1, characterized in that: The edge computing unit mentioned in step 6 is an embedded platform; the deployed model size is no more than 0.1MB, the number of parameters is no more than 0.03M, and the single-batch inference latency is no more than 3ms.

9. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 1, characterized in that: The dual FBG sensing unit in step 1 includes a first FBG sensing unit and a second FBG sensing unit. A quartz capillary tube is sleeved on the outer side of the gate region of the first FBG sensing unit to isolate mechanical strain and is used to output the temperature reference wavelength signal. The first FBG sensing unit is disposed at the interface between the separator and the positive electrode along the reserved first interlayer microchannel. The gate region of the second FBG sensing unit is not sleeved with a capillary tube and is disposed at the interface between the negative electrode and the separator along the reserved second interlayer microchannel. Flexible positioning points are provided on both sides of its gate region to keep the middle of the gate region in flexible contact with the negative electrode and the separator and is used to output the thermo-mechanical coupling wavelength signal.

10. The battery state-of-charge estimation method based on embedded fiber optic multiphysics sensing according to claim 9, characterized in that: The flexible positioning points on both sides of the grid area of ​​the second FBG sensing unit are locally fixed with an electrolyte-resistant adhesive, and the middle of the grid area maintains flexible contact with the negative electrode and the separator without rigid constraints; the ordinary optical fiber segments of the first FBG sensing unit and the second FBG sensing unit are integrally and continuously led out from the inside of the soft pack battery to the outside, and the lead-out position is provided with a sealing structure.

Images