Methods and devices for sensing the microstate of power batteries in hybrid-electric aircraft

By constructing a physical information neural network model and combining electrochemical and thermodynamic subnetworks, the problem of sensing the microscopic state of batteries under extreme operating conditions of low-altitude aircraft was solved, enabling accurate sensing of the internal state of batteries and real-time early warning of safety risks.

CN123069852APending Publication Date: 2026-11-10BEIHANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611110754.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-24
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing automotive-grade BMS algorithms are ill-equipped to handle extreme conditions such as the changing flow field heat dissipation boundary, high-frequency ripple interference, and high-rate transient impact superposition in low-altitude aircraft. They cannot accurately describe the microscopic state under extreme conditions, resulting in inaccurate battery thermal management and difficulty in identifying lithium plating risks.

Method used

A physical information neural network model integrating flight fluid dynamics characteristics and electrochemical-thermodynamic mechanisms is constructed. Using the pre-trained electrochemical and thermodynamic subnetworks of the physical information neural network model, the microscopic state vector of the battery is output by real-time acquisition of multidimensional state vectors, including lithium-ion concentration on the negative electrode surface, solid-phase potential and center temperature. Dynamic convection heat transfer boundary residuals and hybrid power ripple thermal correction terms are introduced to improve estimation accuracy and robustness.

Benefits of technology

Under extreme operating conditions of low-altitude aircraft, it achieves precise perception of the battery's micro-state, improves the real-time performance and accuracy of thermal state perception, and can identify internal lithium plating risks and insufficient power output of the battery, providing precise early warning safety boundaries.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069852A_ABST
    Figure CN123069852A_ABST
Patent Text Reader

Abstract

This invention discloses a method and device for sensing the micro-state of a power battery in a hybrid-electric aircraft, belonging to the field of power battery management. The method includes: real-time acquisition of operational data to construct a multi-dimensional state vector; and outputting a battery micro-state vector based on the multi-dimensional state vector using a physical information neural network model. The physical information neural network model is generated using a target loss function, which includes micro-physical constraint terms composed of thermodynamic physical residuals and electrochemical kinetic residuals. The thermodynamic physical residuals are constructed based on the radial heat conduction partial differential equation, which introduces convective heat transfer boundary residuals that dynamically change with flight conditions, as well as hybrid-electric ripple thermal correction terms. This solution can accurately sense the battery micro-state under extreme operating conditions of low-altitude aircraft, including superimposed "variable flow field heat dissipation boundary," "high-frequency ripple interference," and "high-rate transient impact."
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power battery management technology, and in particular to a method and apparatus for sensing the microstate of power batteries for hybrid-electric aircraft. Background Technology

[0002] To balance the demands of long range and high power requirements for vertical takeoff and landing, series hybrid power architecture has become the mainstream solution for addressing range anxiety and payload limitations associated with pure electric vehicles. These aircraft typically employ high-energy-density ternary lithium (NCM) batteries as energy buffers and high-power output units. However, compared to ground-based electric vehicles, the operating environment and conditions of hybrid low-altitude aircraft are extremely complex, posing new challenges to battery status monitoring and safety management.

[0003] First, the thermal management boundary conditions under flight conditions are highly time-varying and nonlinear. While the heat dissipation environment of ground vehicles is relatively stable, low-altitude aircraft experience dramatic changes in atmospheric pressure, composite airspeed, and rotor downwash during hovering, vertical climb, high-altitude cruise, and rapid descent. Existing battery thermal management models typically treat the convective heat transfer coefficient as a constant or a simple linear variable only related to velocity, failing to adapt to the complex fluid environment of aircraft, characterized by "strong wind cooling during vertical takeoff and landing, and low-pressure, thin air during high-altitude cruise." This leads to severe model mismatch and physical measurement lag when the BMS infers internal temperature from surface temperature sensors. This can easily mask heat accumulation at the battery's internal geometric center during high-rate discharge during takeoff and landing, potentially causing thermal runaway due to "internal heat and external cold."

[0004] Secondly, the electromechanical coupling effect of the hybrid power system introduces complex electromagnetic interference and additional heat sources. In range-extending mode, high-frequency, high-amplitude current ripple is superimposed on the DC bus due to rectifier switching and engine speed fluctuations. This ripple not only interferes with the BMS current sampling signal, masking the true electrochemical polarization characteristics of the battery, but also generates additional temperature rise inside the battery through the AC Joule heating effect. Existing BMS algorithms lack decoupling and compensation mechanisms for this hybrid-specific "ripple heat source," leading to an underestimation of the battery's thermal state.

[0005] Secondly, high-dynamic instantaneous power throughput faces the risk of microscopic lithium plating. When an aircraft encounters gusts and needs to adjust its attitude or perform emergency obstacle avoidance maneuvers, the motor current experiences dramatic fluctuations on a millisecond scale. During vertical descent, the transient braking feedback current of the rotor motors may impact the battery charging boundary. Under these high-dynamic conditions, the lithium-ion concentration gradient on the battery's negative electrode surface can establish itself instantaneously, easily causing the local negative electrode solid-phase potential to drop below 0V, inducing irreversible lithium plating side reactions. Traditional equivalent circuit models can only observe macroscopic terminal voltages and cannot reveal the internal microscopic potentials, making it difficult to provide effective power limit boundaries before lithium plating occurs.

[0006] In summary, existing automotive-grade BMS algorithms struggle to handle the extreme conditions of low-altitude aircraft, which involve the superposition of "variable flow field heat dissipation boundary," "high-frequency ripple interference," and "high-rate transient impact." While some fully physical mechanism models can accurately describe the microscopic state, their computational demands are enormous, making them difficult to run in real-time on computationally limited airborne embedded chips. On the other hand, purely data-driven neural network models heavily rely on complete fault datasets and exhibit poor generalization ability.

[0007] Therefore, there is an urgent need for a method and device for sensing the microstate of power batteries in hybrid-powered aircraft. Summary of the Invention

[0008] To address the problem that existing automotive-grade BMS algorithms struggle to handle the extreme conditions of low-altitude aircraft, including the superposition of "variable flow field heat dissipation boundary," "high-frequency ripple interference," and "high-rate transient impact," and thus cannot accurately describe the microscopic state under extreme conditions, this invention provides a method and apparatus for sensing the microscopic state of power batteries for hybrid-electric aircraft.

[0009] On the one hand, a method for sensing the microstate of a power battery in a hybrid-electric aircraft is provided, the method comprising: Real-time operational data from the battery management system, flight control system, and range extender controller are collected to construct a multi-dimensional state vector that includes electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics, and aging characteristics. Using the pre-trained physical information neural network model with electrochemical and thermodynamic subnetworks, based on the multidimensional state vector, a battery microstate vector is output, including the normalized lithium-ion concentration at the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature. The physical information neural network model is generated using a target loss function with physical mechanism constraints. This target loss function includes a macroscopic data loss term and a microscopic physical constraint term composed of thermodynamic physical residuals and electrochemical kinetic residuals. The thermodynamic physical residuals are constructed based on a radial heat conduction partial differential equation, which introduces a convective heat transfer boundary residual that dynamically changes with flight state and a hybrid power ripple thermal correction term. A battery safety risk signal is generated based on the battery microstate vector and broadcast to the downstream controller.

[0010] On the other hand, a microstate sensing device for a power battery of a hybrid-powered aircraft, based on the steps described in any method embodiment of the specification, is provided. The device includes: The acquisition unit is used to acquire real-time operating data from the battery management system, flight control system, and range extender controller, and construct a multi-dimensional state vector that includes electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics, and aging characteristics. The sensing unit is used to utilize the electrochemical and thermodynamic subnets of a pre-trained physical information neural network model to output a battery microstate vector based on the multidimensional state vector. This vector includes the normalized lithium-ion concentration at the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature. The physical information neural network model is generated by training using a target loss function with physical mechanism constraints. The target loss function includes a macroscopic data loss term and a microscopic physical constraint term composed of thermodynamic physical residuals and electrochemical kinetic residuals. The thermodynamic physical residuals are constructed based on the radial heat conduction partial differential equation, which introduces a convective heat transfer boundary residual that dynamically changes with flight state and a hybrid power ripple thermal correction term. The broadcasting unit is used to generate a battery safety risk signal based on the battery microstate vector and broadcast it to the downstream controller.

[0011] On the other hand, a computer device is provided, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for executing the computer program stored in the memory to implement the steps of the method described above.

[0012] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when the computer program is executed by a processor, it implements the steps of the method described above.

[0013] On the other hand, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0014] The technical solution provided by this invention can bring at least the following beneficial effects: When constructing a physical information neural network that integrates flight fluid dynamics characteristics and electrochemical-thermodynamic mechanisms, a convective heat transfer boundary residual that dynamically varies with flight conditions is introduced into the thermodynamic physical residual term of the physical mechanism-constrained objective loss function, instead of the constant convective heat transfer coefficient of traditional techniques. This allows the system to accurately calculate the core temperature change trend inside the battery based on fluid dynamics characteristics, even when the surface temperature sensor exhibits physical lag due to heat conduction, thus improving the real-time performance and accuracy of thermal state perception under high-dynamic flight conditions. Furthermore, the range extender speed hybrid power interference characteristics are introduced into the model input, and a hybrid power ripple thermal correction term is added to the thermodynamic physical residual term, enabling... This paper corrects the impact of high-frequency ripple interference in hybrid power systems on battery state estimation, effectively improving the robustness and accuracy of battery thermal state estimation in hybrid mode. Furthermore, a physical information neural network model is constructed by combining neural networks with physical mechanism models. By constraining the neural network using physical mechanisms, a mapping relationship between macroscopic electrical quantities and microscopic electrochemical states is established. This model can output in real-time, non-measurable physical quantities such as the battery's internal core temperature, negative electrode solid-phase potential, and surface ion concentration at the airborne terminal. This solves the problem that traditional methods relying solely on terminal voltage cannot identify internal transient lithium plating risks and insufficient power output from the battery, providing a basis for setting precise early warning safety boundaries for energy management systems. In summary, this solution can accurately perceive the battery's microscopic state under extreme operating conditions of low-altitude aircraft, including "variable flow field heat dissipation boundaries," "high-frequency ripple interference," and "high-rate transient impacts." Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for sensing the microstate of a power battery in a hybrid-electric aircraft, provided by an embodiment of the present invention. Figure 2 This is a structural diagram of a power battery microstate sensing device for hybrid-powered aircraft, provided by an embodiment of the present invention. Figure 3This is a hardware architecture diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] The following describes the specific implementation of the above concept.

[0019] Please refer to Figure 1 This invention provides a method for sensing the microstate of a power battery in a hybrid-electric aircraft, the method comprising: Step 100: Real-time acquisition of operating data from the battery management system, flight control system, and range extender controller to construct a multi-dimensional state vector containing electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics, and aging characteristics; Step 102: Using the pre-trained electrochemical and thermodynamic subnets of the physical information neural network model, based on the multidimensional state vector, output the battery microstate vector containing the normalized lithium-ion concentration at the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature; wherein, the physical information neural network model is generated by training using a target loss function containing physical mechanism constraints. The target loss function includes macroscopic data loss terms and microscopic physical constraint terms composed of thermodynamic physical residual terms and electrochemical kinetic residual terms. The thermodynamic physical residual terms are constructed based on the radial heat conduction partial differential equation, and the radial heat conduction partial differential equation introduces the convective heat transfer boundary residual that dynamically changes with the flight state and the hybrid power ripple thermal correction term; Step 104: Generate a battery safety risk signal based on the battery microstate vector and broadcast it to the downstream controller.

[0020] In this embodiment of the invention, when constructing a physical information neural network that integrates flight fluid dynamics characteristics and electrochemical-thermodynamic mechanisms, a convective heat transfer boundary residual that dynamically changes with flight state is introduced into the thermodynamic physical residual term of the physical mechanism constraint target loss function, instead of the constant convective heat transfer coefficient of traditional techniques. This allows the system to accurately calculate the core temperature change trend inside the battery based on fluid dynamics characteristics, even when the surface temperature sensor has a physical lag due to heat conduction, thus improving the real-time performance and accuracy of thermal state perception under high-dynamic flight conditions. Furthermore, the range extender speed hybrid power interference characteristics are introduced at the model input, and hybrid power ripple thermal correction is added to the thermodynamic physical residual term. This approach can correct the impact of high-frequency ripple interference in hybrid power systems on battery state estimation, effectively improving the robustness and accuracy of battery thermal state estimation in hybrid mode. Furthermore, by combining neural networks with physical mechanism models, a physical information neural network model is constructed. Using physical mechanism constraints, a mapping relationship between macroscopic electrical quantities and microscopic electrochemical states is established. This model can output in real-time, non-measurable physical quantities such as the battery's internal core temperature, negative electrode solid-phase potential, and surface ion concentration at the airborne terminal. This solves the problem that traditional methods relying solely on terminal voltage cannot identify internal transient lithium plating risks and insufficient power output from the battery, providing a basis for setting precise early warning safety boundaries for energy management systems. In summary, this solution can accurately perceive the battery's microscopic state under extreme conditions of superimposed "variable flow field heat dissipation boundaries," "high-frequency ripple interference," and "high-rate transient impact" in low-altitude aircraft.

[0021] The following description Figure 1 The execution method for each step is shown.

[0022] For step 100: Understandably, the offline-trained model needs to be lightweighted before being deployed on the aircraft. By collecting real-time operational data from the battery management system, flight control system, and range extender controller, a multi-dimensional state vector is constructed, encompassing electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics, and aging characteristics. This multi-dimensional state vector includes at least the following parameters: load current, battery terminal voltage, battery state of charge, battery module surface temperature, ambient atmospheric temperature, aircraft vertical speed, aircraft composite airspeed, ambient atmospheric static pressure, range extender generator speed, and battery health status, as shown below: In the formula, the electrochemical characteristic is the load current. Battery terminal voltage Battery state of charge The thermal characteristic is the surface temperature of the battery module. External ambient temperature The hydrodynamic characteristics of flight are the vertical velocity of the aircraft. Combined airspeed of aircraft External atmospheric static pressure The interference characteristic of hybrid power systems is the range extender generator speed. Used to identify the high-frequency ripple intensity and mechanical vibration level on the DC bus; aging characteristics indicate the battery's health status. It can be used to adaptively correct the solid-phase diffusion coefficient in subsequent physical equations. and dynamic ohmic resistance Parameters that decline with aging.

[0023] Regarding step 102: Next, the training process of the physical information neural network model will be explained.

[0024] In some implementations, the physical information neural network model is trained as shown in steps S1-S7 below: S1. Using multiphysics simulation, multidimensional feature samples are generated under the full flight envelope conditions of the hybrid-electric aircraft, and the microstate truth labels under each multidimensional feature sample are calculated to obtain a virtual dataset. Among them, the multidimensional feature samples include at least the load current, battery terminal voltage, battery state of charge, battery module surface temperature, ambient atmospheric temperature, aircraft vertical speed, aircraft composite airspeed, ambient atmospheric static pressure, range extender generator speed, and battery health status. The microstate truth labels include the true value of the normalized lithium-ion concentration on the negative electrode surface, the true value of the negative electrode solid-phase potential, and the true value of the battery center temperature.

[0025] Since it is impossible to non-destructively implant sensors into actual flight batteries to measure microscopic states, this embodiment uses simulation software to generate ground truth labels for training data. A multiphysics simulation model of the battery pack is established using simulation software. The model is input with the specific operating condition spectrum of the hybrid-electric aircraft (including high-rate pulses during takeoff and landing, and high-frequency ripples during range extension charging), covering the entire flight envelope including vertical takeoff and landing, high-altitude cruise, range extension charging, and emergency obstacle avoidance. Different SOH levels (e.g., 100%, 90%, 80%) are set, and the normalized lithium-ion concentration ground truth value on the negative electrode surface is calculated for each multidimensional feature sample. True value of negative electrode solid phase potential True value of battery center temperature That is, the true value label of the microstate, which is used to obtain the virtual dataset.

[0026] S2, for each multidimensional feature sample, performs the following: input the multidimensional feature sample into the electrochemical subnet and the thermodynamic subnet respectively, use the spatiotemporal distribution function of the absolute lithium-ion concentration of the negative electrode solid phase in the electrochemical subnet to calculate the normalized lithium-ion concentration and the negative electrode solid phase potential, and use the thermodynamic subnet to calculate the battery center temperature.

[0027] In this embodiment, in order to address the timescale mismatch and physical mechanism differences between electrochemical reactions (milliseconds) and heat conduction (seconds), the physical information neural network constructed in this embodiment adopts a dual-current coupling architecture of an electrochemical subnetwork and a thermodynamic subnetwork.

[0028] Both the electrochemical and thermodynamic subnetworks employ fully connected deep neural networks, comprising an input layer, several hidden layers, and an output layer. The hidden layers uniformly use the hyperbolic tangent function as the activation function to ensure that the network output has second-order continuous differentiability within the solution domain of the physical partial differential equations.

[0029] The thermodynamic subnet focuses on solving the radial temperature field of the battery, and its hidden layers are preferably set with 3 to 4 fully connected layers, each containing 32 to 64 neurons.

[0030] The electrochemical subnetwork focuses on solving the internal concentration and potential fields. Its hidden layers preferably consist of 4 to 6 fully connected layers, each containing 64 to 128 neurons, to provide sufficient capacity to fit the highly nonlinear Butler-Volmer polarization curve. The network incorporates explicit physical parameter coupling pathways. For example, the solid-phase diffusion coefficient in the physical layers of the electrochemical subnetwork... Defined as the output battery center temperature of the thermodynamic subnet and input Joint functions: .

[0031] Therefore, the output layer of the physical information neural network model outputs the internal microstate vector of the battery. Defined as a column vector: in, This refers to the normalized lithium-ion concentration at the negative electrode surface, i.e., the absolute lithium-ion concentration at the negative electrode surface. With the maximum theoretical lithium intercalation concentration The ratio; The negative electrode solid-state potential is a direct indicator of lithium plating risk. The battery center temperature is a direct indicator for determining the risk of internal short circuits and thermal runaway.

[0032] S3. Using the negative electrode solid-phase potential output by the electrochemical subnet and the battery center temperature output by the thermodynamic subnet, the predicted terminal voltage is reconstructed. Using the predicted terminal voltage, the battery terminal voltage in the multidimensional feature sample, the predicted surface temperature in the thermodynamic subnet, and the battery module surface temperature in the multidimensional feature sample, the macroscopic data loss term used to constrain the consistency of the neural network's predictions on macroscopic data is calculated.

[0033] In this embodiment of the invention, the target loss function contains physical mechanism constraints. This function is used to guide neural network training. It consists of two parts: macroscopic observation error and microscopic physical constraint terms. The macroscopic observation error is a macroscopic data loss term. The microscopic physical constraint term is derived from the thermodynamic physical residual term. and electrochemical kinetic residuals composition.

[0034] Macroeconomic data loss item This is used to constrain the neural network to maintain consistency with the physical entity in terms of macroscopic data, ensuring the predicted terminal voltage based on the output reconstruction deduction of the electrochemical subnet. extrapolated surface temperature in thermodynamic subnet It can accurately match the input data of the neural network, namely the battery terminal voltage and battery module surface temperature in the multidimensional feature sample.

[0035] In some implementations, the predicted terminal voltage is calculated as follows: In the formula, To predict the terminal voltage, Based on battery state of charge The positive open-circuit potential is obtained by looking up the table. The negative solid-phase potential output by the electrochemical subnet. For load current, Based on battery health status Battery center temperature calculated by thermodynamic subnet The corrected dynamic ohmic resistance is obtained.

[0036] In this embodiment, based on the principle of full-cell voltage balance, the positive electrode open-circuit potential obtained from the macroscopic state of charge (SOC) table of the BMS is introduced, and the terminal voltage is reconstructed and predicted by combining the negative electrode solid-phase potential output by the electrochemical subnetwork. .

[0037] Macroeconomic data loss item Calculated as follows: in, The total number of training sample points refers to the number of time points containing the measured labels used for training. Indicates the sampling time, representing the time corresponding to the i-th data point; For predicting terminal voltage; This refers to the battery terminal voltage in this multidimensional feature sample; This represents the reverse-engineered surface temperature, which is the temperature value output by the thermodynamic subnet at the battery surface boundary (i.e., at r=R). It is the surface temperature of the battery module in this multidimensional feature sample.

[0038] S4 introduces a hybrid power ripple thermal correction term and calculates the unit volume coupling heat generation rate based on the true value of the battery center temperature. It defines the internal residual function based on the radial heat conduction partial differential equation and calculates the convective heat transfer boundary residual using the convective heat transfer coefficient function. Based on the internal residual function and the convective heat transfer boundary residual, it calculates the thermodynamic physical residual term used to constrain the internal temperature field of the battery to satisfy energy conservation.

[0039] In some implementations, the thermodynamic physical residuals are calculated as follows: Substitute the generator speed of the range extender into the mapping function of the effective value of the ripple current to determine the effective value of the ripple current. Based on the effective value of ripple current, the effective geometric volume of a single cell, and the high-frequency AC internal resistance of the battery, the ripple thermal correction term for hybrid power is calculated. Based on the hybrid power ripple thermal correction term and the true value of the battery center temperature, the coupling heat generation rate per unit volume is calculated. The temperature field distribution of the thermodynamic subnet solution is obtained, and the internal residual function is calculated by combining the unit volume coupling heat generation rate with the radial heat conduction partial differential equation. The vertical velocity of the aircraft, the combined airspeed of the aircraft, and the static pressure of the external atmosphere are input into the convective heat transfer coefficient function to calculate the target convective heat transfer coefficient. Based on the extrapolated surface temperature and the ambient atmospheric temperature in the thermodynamic subnet, the convective heat transfer boundary residual is calculated. Several collocation points are randomly selected inside the battery and at the boundary. Based on the internal residual function and the convection heat transfer boundary residual, the thermodynamic physical residual term is calculated.

[0040] In this embodiment, the hybrid power ripple thermal correction term is calculated using the following formula: In the formula, For hybrid power ripple thermal correction term, This is the ripple thermal correction factor. The effective geometric volume of a single battery cell. This is a mapping function for the effective value of ripple current. For the range extender generator speed, This is the high-frequency AC internal resistance of the battery.

[0041] The unit volume coupling heat generation rate is explicitly superimposed from the range extender speed in the input vector. Determined hybrid power ripple thermal correction term, coupled heat generation rate per unit volume Calculated using the following formula: + + The corresponding expansion is as follows: In the formula, Joule heating represents the irreversible heat generated when current flows through the ohmic and polarization resistances inside a battery. Reversible entropic heat characterizes the reversible endothermic or exothermic effect generated by entropy change during an electrochemical reaction. For hybrid power ripple thermal correction term; This is the load current; This refers to the battery terminal voltage. The open-circuit voltage corresponding to the SOC can be obtained by looking up the estimated SOC value from the onboard battery management system. The effective geometric volume of a single battery cell. The entropy heat coefficient is a nonlinear function of SOC, typically obtained through offline calibration and stored in a lookup list in a BMS. This is the true value of the battery center temperature. This is the ripple thermal correction factor. The effective geometric volume of a single battery cell. This is a mapping function for the effective value of ripple current. For the range extender generator speed, This is the high-frequency AC internal resistance of the battery.

[0042] Next, the temperature field distribution calculated from the thermodynamic subnet is obtained, and combined with the unit volume coupling heat generation rate, the internal residual function is calculated using the radial heat conduction partial differential equation. : In the formula, The average density of the battery characterizes the mass distribution characteristics of the battery materials. Specific heat capacity is the capacity of a battery, characterizing the ability of battery materials to store thermal energy. Radial thermal conductivity is the rate at which heat is transferred radially within the battery. Temperature field distribution for thermodynamic subnet solution It's about time. and radius The function; The rate of change of temperature over time; This represents the temperature gradient along the radial direction. The radial coordinate represents the distance from the geometric center of the battery cell; The heat generation rate per unit volume is the coupling heat generation rate.

[0043] Convection heat transfer boundary residual Calculated as follows: In the formula, Radial thermal conductivity, For battery surface The temperature gradient along the radial direction, For the vertical velocity of the aircraft Combined airspeed of aircraft static pressure of the outside atmosphere The convective heat transfer coefficient function, This indicates the projected surface temperature of the battery. The ambient atmospheric temperature.

[0044] In this embodiment, due to the drastic changes in the aircraft's operating conditions, the convective heat transfer coefficient is no longer a constant, but rather a factor of the flight parameters in the input vector. The function can correct the boundary conditions of the thermal management model in real time. This allows the system to accurately calculate the core temperature change trend inside the battery based on hydrodynamic characteristics, even when the surface temperature sensor has a physical lag due to heat conduction. This improves the real-time performance and accuracy of thermal state perception under high-dynamic flight conditions.

[0045] Finally, the thermodynamic physical residual term The final calculation formula is to randomly select from inside the battery. Select a coordinate point on the boundary. Each point is matched, combined with adaptive penalty weights. Calculate the sum of mean square errors: In the formula, For internal residual functions, This represents the boundary residual for convective heat transfer.

[0046] In this embodiment, internal points Alignment with boundary points These are unsupervised spatiotemporal coordinate points, and no corresponding real temperature label is required. During model training, the Latin hypercube sampling (LHS) algorithm is preferred for the battery spatial domain. With the time domain The set of collocation points is generated internally. In addition, in order to capture the drastic physical changes during the high-rate take-off and landing of the aircraft, an adaptive resampling strategy can be introduced to dynamically increase the density of collocation points in the spatiotemporal concentration where the physical residual is high, thereby accelerating the convergence of the physical information neural network and improving the local accuracy.

[0047] S5 defines the internal residual function of active particles based on the spatiotemporal distribution function in the electrochemical subnet, and defines the boundary residual of active particles based on the normalized lithium-ion concentration true value on the negative electrode surface. Based on the internal residual function and boundary residual of active particles, the electrochemical kinetic residual term used to constrain the spatiotemporal distribution of lithium-ion concentration in the electrochemical subnet to conform to the microscopic diffusion kinetics is calculated.

[0048] In some implementations, the electrochemical kinetic residual term is calculated as follows: The solid-phase diffusion coefficient is calculated based on the battery center temperature and battery health status output from the thermodynamic subnet. Based on the spatiotemporal distribution function and solid-phase diffusion coefficient in the electrochemical subnet, the internal residual function of the active particles is calculated. Based on the true value of the normalized lithium-ion concentration on the negative electrode surface, the true value of the radial gradient on the particle surface is determined, and the boundary residual of the active particles is calculated by combining the load current and the solid-phase diffusion coefficient. Several collocation points are randomly selected inside and at the boundary of the active particles. Based on the internal residual function and boundary residual of the active particles, the electrochemical kinetic residual term is calculated.

[0049] In this embodiment, the solid-phase diffusion coefficient is calculated as follows: In the formula, Indicates the nominal reference temperature Initial solid-phase diffusion coefficient at ambient temperature 298.15 K; The aging degradation function is based on the battery health state, characterizing the internal diffusion channel blockage effect caused by the decrease in battery cycle life; The diffusion activation energy is the energy threshold required for lithium ions to overcome lattice resistance. It is the ideal gas constant; This represents the battery center temperature output by the thermodynamic subnet.

[0050] Inside the active particles (spherical coordinate system, radius) Lithium-ion transport follows Fick's second law. The internal residual function of the active particle is: In the formula, Let be the spatiotemporal distribution function of the absolute lithium-ion concentration in the negative electrode solid phase of the electrochemical subnet, representing the concentration at time t and distance from the center of the microparticle. The absolute physical quantity of lithium intercalation at the site; and These represent the rate of change of absolute lithium-ion concentration with respect to time and the distribution gradient in microscopic space, respectively, and the solid-phase diffusion coefficient. It is a core parameter that determines the high-rate performance of a battery. In this invention, it is not a static constant, but is constructed as a dynamically coupled variable.

[0051] On the surface of micro-active particles ( The diffusion flux of lithium ions into and out of particles must be strictly controlled by the external macroscopic load current. The boundary residuals of the active particles are calculated using the following formula: In the formula, The dynamic solid-phase diffusion coefficient; To determine the true value of the radial gradient on the particle surface based on the normalized true value of the lithium-ion concentration on the negative electrode surface; This is the load current; It is Faraday's constant; The effective active specific surface area of ​​the negative electrode per unit volume; This refers to the macroscopic cross-sectional area of ​​the battery cell electrode. The electrode coating thickness.

[0052] It should be noted that the normalized lithium-ion concentration at the negative electrode surface... .

[0053] Several collocation points are randomly selected inside and at the boundary of the active particles. Based on the internal residual function and boundary residual of the active particles, the electrochemical kinetic residual term is calculated: The mean square error was calculated using the collocation method, and samples were taken from inside the active particles. An internal point is selected on the boundary. Each point is matched, combined with adaptive penalty weights. Calculate the sum of mean square errors: Electrochemical kinetic residuals are used to constrain the lithium-ion concentration output by the electrochemical subnet. This conforms to the laws of microscopic diffusion kinetics, thus reconstructing the polarization state inside the battery even without measured concentration data. By minimizing this loss term, The concentration field of the neural network output was ensured. It is continuous in both time and space and conforms to the physical laws of diffusion.

[0054] S6 calculates the target loss function based on the macroscopic data loss term, the thermodynamic physical residual term, and the electrochemical kinetic residual term.

[0055] The target loss function is: in, This is a loss item for macroeconomic data; This is the thermodynamic physical residual term; This is the residual term in electrochemical kinetics; For network weights, These are dynamically adaptive weighting coefficients.

[0056] S7 trains the electrochemical and thermodynamic subnets by minimizing the objective loss function until a physical information neural network model with optimal network parameters is obtained.

[0057] Regarding step 104: The trained physical information neural network model is deployed on the aircraft's onboard terminal to calculate the microstate vector inside the battery in real time. Based on the observation results, a microscopic lithium plating risk status signal, a power output insufficiency signal, and a thermal accumulation status signal are generated and sent to the downstream controller via the airborne bus to execute control actions.

[0058] Lithium plating risk warning: When When the threshold is less than the specified value, the bit is set to indicate that the downstream system needs to limit the charging and discharging power.

[0059] Thermal accumulation warning: When When the rate of increase is greater than the threshold and the heat dissipation conditions are poor, the setting is activated to indicate that the downstream system needs to intervene in thermal management.

[0060] Insufficient power warning: When If the value is less than the threshold, the device is set to request the EMS to limit the discharge power in advance or to start the range extender to replenish power.

[0061] Please refer to Figure 2 This invention provides a microscopic state sensing device for a power battery in a hybrid-electric aircraft, used to implement the steps of any method embodiment described in the specification. The device includes: The acquisition unit 201 is used to acquire real-time operating data of the battery management system, flight control system and range extender controller, and construct a multi-dimensional state vector that includes electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics and aging characteristics. The sensing unit 202 is used to utilize the electrochemical and thermodynamic subnets of the pre-trained physical information neural network model to output a battery microstate vector based on a multidimensional state vector, which includes the normalized lithium-ion concentration at the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature. The physical information neural network model is generated by training using a target loss function containing physical mechanism constraints. The target loss function includes a macroscopic data loss term and a microscopic physical constraint term composed of thermodynamic physical residual term and electrochemical kinetic residual term. The thermodynamic physical residual term is constructed based on the radial heat conduction partial differential equation, and the radial heat conduction partial differential equation introduces a convective heat transfer boundary residual that dynamically changes with the flight state and a hybrid power ripple thermal correction term. Broadcast unit 203 is used to generate a battery safety risk signal based on the battery microstate vector and broadcast it to the downstream controller.

[0062] It should be noted that the above device embodiments and method embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0063] Embodiments of this application also provide a computer device, please refer to... Figure 3 The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to implement the microstate sensing method for power batteries of hybrid-electric aircraft provided in the above-described method embodiments.

[0064] Embodiments of this application also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the micro-state sensing method for power batteries of hybrid-electric aircraft provided in the above-described method embodiments.

[0065] Embodiments of this application also provide a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium and executes the computer program, causing the computer device to perform any of the above embodiments of the power battery microstate sensing method for hybrid-electric aircraft.

[0066] For ease of description, the above devices or apparatuses are described separately according to their functions, divided into various modules or units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0067] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of the embodiments of this application.

[0068] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0069] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for sensing the microscopic state of a power battery in a hybrid-electric aircraft, characterized in that, include: Real-time operational data from the battery management system, flight control system, and range extender controller are collected to construct a multi-dimensional state vector that includes electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics, and aging characteristics. Using the pre-trained physical information neural network model with electrochemical and thermodynamic subnetworks, based on the multidimensional state vector, a battery microstate vector is output, including the normalized lithium-ion concentration at the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature. The physical information neural network model is generated using a target loss function with physical mechanism constraints. This target loss function includes a macroscopic data loss term and a microscopic physical constraint term composed of thermodynamic physical residuals and electrochemical kinetic residuals. The thermodynamic physical residuals are constructed based on a radial heat conduction partial differential equation, which introduces a convective heat transfer boundary residual that dynamically changes with flight state and a hybrid power ripple thermal correction term. A battery safety risk signal is generated based on the battery microstate vector and broadcast to the downstream controller.

2. The method as described in claim 1, characterized in that, The physical information neural network model is trained in the following way: Multiphysics simulation is used to generate multidimensional feature samples under the full flight envelope conditions of a hybrid-electric aircraft, and the microstate truth labels under each multidimensional feature sample are calculated to obtain a virtual dataset. The multidimensional feature samples include at least the load current, battery terminal voltage, battery state of charge, battery module surface temperature, ambient atmospheric temperature, aircraft vertical speed, aircraft composite airspeed, ambient atmospheric static pressure, range extender generator speed, and battery health status. The microstate truth labels include the true values ​​of the normalized lithium-ion concentration on the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature. For each multidimensional feature sample, perform the following: The multidimensional feature samples are input into the electrochemical subnet and the thermodynamic subnet respectively. The spatiotemporal distribution function of the absolute lithium-ion concentration of the negative electrode solid phase in the electrochemical subnet is used to calculate the normalized lithium-ion concentration and the negative electrode solid phase potential. The battery center temperature is calculated using the thermodynamic subnet. Using the negative electrode solid-phase potential output by the electrochemical subnetwork and the battery center temperature output by the thermodynamic subnetwork, the predicted terminal voltage is reconstructed. Using the predicted terminal voltage, the battery terminal voltage in the multidimensional feature sample, the predicted surface temperature in the thermodynamic subnetwork, and the battery module surface temperature in the multidimensional feature sample, a macroscopic data loss term is calculated to constrain the consistency of the neural network's predictions on macroscopic data. A hybrid power ripple thermal correction term is introduced, and the unit volume coupling heat generation rate is calculated based on the true value of the battery center temperature. An internal residual function is defined based on the radial heat conduction partial differential equation. At the same time, the convective heat transfer coefficient function is used to calculate the convective heat transfer boundary residual. The thermodynamic physical residual term used to constrain the internal temperature field of the battery to satisfy energy conservation is calculated based on the internal residual function and the convective heat transfer boundary residual. The internal residual function of the active particles is defined based on the spatiotemporal distribution function in the electrochemical subnet, and the boundary residual of the active particles is defined based on the normalized lithium ion concentration true value on the negative electrode surface. Based on the internal residual function and boundary residual of the active particles, the electrochemical kinetic residual term used to constrain the spatiotemporal distribution of lithium ion concentration in the electrochemical subnet to conform to the microscopic diffusion kinetics is calculated. The target loss function is calculated based on the macroscopic data loss term, the thermodynamic physical residual term, and the electrochemical kinetic residual term. By minimizing the target loss function, the electrochemical subnet and the thermodynamic subnet are trained until a physical information neural network model with optimal network parameters is obtained.

3. The method as described in claim 2, characterized in that, The predicted terminal voltage is calculated as follows: In the formula, To predict the terminal voltage, Based on battery state of charge The positive open-circuit potential is obtained by looking up the table. The negative electrode solid-phase potential output by the electrochemical sub-network, The load current, Based on the battery health status The battery center temperature calculated by the thermodynamic subnet The corrected dynamic ohmic resistance is obtained.

4. The method as described in claim 2, characterized in that, The thermodynamic physical residual term is calculated as follows: Substitute the generator speed of the range extender into the ripple current effective value mapping function to determine the effective value of the ripple current. Based on the RMS value of the ripple current, the effective geometric volume of a single cell, and the high-frequency AC internal resistance of the battery, the hybrid power ripple thermal correction term is calculated. Based on the hybrid power ripple thermal correction term and the true value of the battery center temperature, the unit volume coupling heat generation rate is calculated. The temperature field distribution of the thermodynamic subnet solution is obtained, and the internal residual function is calculated using the radial heat conduction partial differential equation in combination with the unit volume coupling heat generation rate. The vertical velocity of the aircraft, the combined airspeed of the aircraft, and the static pressure of the external atmosphere are input into the convective heat transfer coefficient function to calculate the target convective heat transfer coefficient. Based on the extrapolated surface temperature and the ambient atmospheric temperature in the thermodynamic subnet, the convective heat transfer boundary residual is calculated. Several collocation points are randomly selected inside the battery and at its boundary. Based on the internal residual function and the convection heat transfer boundary residual, the thermodynamic physical residual term is calculated.

5. The method as described in claim 4, characterized in that, The hybrid power ripple thermal correction term is calculated using the following formula: In the formula, For hybrid power ripple thermal correction term, This is the ripple thermal correction factor. The effective geometric volume of a single battery cell. This is a mapping function for the effective value of ripple current. The speed of the range extender generator. This is the high-frequency AC internal resistance of the battery.

6. The method as described in claim 2, characterized in that, The electrochemical kinetic residual term is calculated as follows: The solid-phase diffusion coefficient is calculated based on the battery center temperature and battery health status output by the thermodynamic subnet. Based on the spatiotemporal distribution function and the solid-phase diffusion coefficient in the electrochemical subnet, the internal residual function of the active particles is calculated. Based on the true value of the normalized lithium-ion concentration on the negative electrode surface, the true value of the radial gradient on the particle surface is determined, and the boundary residual of the active particles is calculated by combining the load current and the solid-phase diffusion coefficient. Several collocation points are randomly selected inside and at the boundary of the active particles. Based on the internal residual function and boundary residual of the active particles, the electrochemical kinetic residual term is calculated.

7. A microscopic state sensing device for a power battery in a hybrid-electric aircraft, used to implement the steps of the method described in any one of claims 1-6, characterized in that, include: The acquisition unit is used to acquire real-time operating data from the battery management system, flight control system, and range extender controller, and construct a multi-dimensional state vector that includes electrochemical characteristics, thermal characteristics, flight hydrodynamic characteristics, hybrid power interference characteristics, and aging characteristics. The sensing unit is used to utilize the electrochemical and thermodynamic subnets of a pre-trained physical information neural network model to output a battery microstate vector based on the multidimensional state vector. This vector includes the normalized lithium-ion concentration at the negative electrode surface, the negative electrode solid-phase potential, and the battery center temperature. The physical information neural network model is generated by training using a target loss function with physical mechanism constraints. The target loss function includes a macroscopic data loss term and a microscopic physical constraint term composed of thermodynamic physical residuals and electrochemical kinetic residuals. The thermodynamic physical residuals are constructed based on the radial heat conduction partial differential equation, which introduces a convective heat transfer boundary residual that dynamically changes with flight state and a hybrid power ripple thermal correction term. The broadcasting unit is used to generate a battery safety risk signal based on the battery microstate vector and broadcast it to the downstream controller.

8. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-6.