Lithium battery unsupervised health assessment and residual life prediction method based on physical knowledge

Abstract

The invention discloses a lithium battery unsupervised health assessment and residual life prediction method based on physical knowledge, and the method comprises the steps: collecting the data of a lithium battery charging and discharging experiment sensor, and carrying out the data preprocessing; guiding the construction of a feature extraction network according to the lithium battery mechanism model, extracting unobservable signals, and enhancing the input space of a health assessment and residual life prediction algorithm; based on the lithium battery mechanism model and the degradation model, constructing a health assessment and residual life prediction algorithm based on PINN; inputting the enhanced feature data into a health assessment and residual life prediction algorithm based on PINN, and optimizing model parameters constructed by health indications; health indications fused with physical knowledge are constructed, and residual life estimation based on the health indications is realized. The method can be used for health assessment and residual life prediction when a data set is small in scale and a single sensor signal can be obtained, and a relatively accurate health assessment result is obtained on the premise of ensuring the interpretability of the algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of health status estimation, and in particular to a method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge. Background Art

[0002] In recent years, lithium batteries have become a key component of energy storage systems in fields including electric vehicles (EV) and aerospace equipment (AE) due to their huge advantages in weight energy density, volume energy density, and service life. According to relevant data from the International Energy Agency, global electric vehicle sales in 2024 will increase by about 25% over previous years, reaching about 17 million vehicles by the end of the year. According to data from the China Association of Automobile Manufacturers, the market penetration rate of new energy vehicles in China has shown a trend of sustained and stable growth in recent years, rising to 39.5% in May 2024. Most new energy vehicles use lithium batteries as the main driving force, among which lithium iron phosphate (LFP) batteries account for nearly 70% of the installed capacity, while ternary lithium batteries account for a slightly lower proportion as power batteries for high-end models. Since the British STRV-1d satellite first used lithium batteries as a satellite power supply system, lithium batteries have become the main energy storage equipment for power systems in aerospace equipment including satellites, space shuttles, and various deep space probes, and have gradually become the third-generation power batteries for aerospace applications.

[0003] Under the influence of environmental interference, material aging, long-term cycle cumulative loss and other factors, lithium battery systems will inevitably degrade, which will lead to a decrease in the overall stability, reliability and economy of the system, and even safety accidents. Therefore, in order to ensure the long-term reliable operation of lithium batteries and carry out timely maintenance and component replacement, it is necessary to implement prognostics and health management (PHM) for lithium batteries, which specifically includes two tasks: (i) estimation of lithium battery state of health (SoH); (ii) remaining useful life (RUL) prediction.

[0004] General health assessment algorithms based on physical models or data-driven are often constrained by factors such as the complexity of degradation model construction and the small size of data sets, making it difficult to give full play to the advantages of the algorithms. Therefore, researchers have turned their attention to the field of hybrid models, aiming to use physical models to guide the construction of data-driven models, and then develop machine learning models that integrate physical laws (Physics-informed machine learning, PIML). PIML can be defined as a model that uses prior knowledge of the physical and mathematical understanding of the research object to improve the performance of data-driven algorithms. The basic applications of different types of PIML are reviewed in "Physics-informed machine learning: A comprehensive review on applications in anomaly detection and condition monitoring". Among such models, the Physics-Informed Neural Network (PINN) has received widespread attention. PINN is an algorithmic framework that uses differential operators to simulate the dynamic equations of the research object, aiming to use prior knowledge to guide and constrain the learning direction of the algorithm. "Deep hidden physics models: Deep learning of nonlinear partialdifferential equations" elaborates on the basic structure and mathematical principles of PINN, and proposes a deep hidden physics model (Deep Hidden Physics Model, DeepHPM) based on the PINN framework. This method can still provide prediction results that are relatively consistent with physical principles when the data set is not ideal, and at the same time demonstrates good generalization ability. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a lithium battery unsupervised health assessment and remaining life prediction method based on physical knowledge, which can be used for health assessment and remaining life prediction when the data set is small and a single sensor signal can be obtained, and relatively accurate health assessment results can be obtained while ensuring the interpretability of the algorithm.

[0006] The present invention solves the technical problem by adopting the following technical solution:

[0007] A method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge includes the following steps:

[0008] Step S1, data preprocessing:

[0009] Collect relevant sensor data in lithium battery charging and discharging experiments and perform data preprocessing;

[0010] Step S2, feature extraction network based on physical model:

[0011] The mathematical expression of the lithium battery mechanism model is used to guide the construction of the feature extraction network, extract unobservable signals containing the system degradation law, and enhance the input space of the health assessment and remaining life prediction algorithm;

[0012] Step S3, constructing a network based on the health indicators of PINN:

[0013] Based on the dynamic process of the lithium battery mechanism model and the dynamic expression of the degradation model, a PINN-based health assessment and remaining life prediction algorithm is constructed; the enhanced feature data is input into the PINN-based health assessment and remaining life prediction algorithm to optimize the model parameters for health indicator construction;

[0014] Step S4, health assessment and remaining life expectancy prediction:

[0015] Characteristic signals with physical significance are extracted based on the test data, and on this basis, health indicators integrating physical knowledge are constructed, ultimately achieving remaining life estimation based on health indicators.

[0016] Furthermore, in step S1, the data is preprocessed using the 2-sigma principle, including:

[0017] Step S10, dividing the input sensor data into different groups of data according to a sliding window with a window length of len and a window interval of inter;

[0018] Step S11, calculating the mean μ and standard deviation σ of each group of data;

[0019] Step S12, according to the 2-sigma principle, data points falling in the interval [μ-2σ, μ+2σ] are normal data and are retained; data points falling outside [μ-2σ, μ+2σ] are outliers and are eliminated.

[0020] Furthermore, step S1 further includes: first performing Min-Max normalization processing on normal data, and then performing data division processing.

[0021] Furthermore, in step S2, the specific method of the feature extraction network based on the physical model is:

[0022] Step S21, lithium battery equivalent circuit model and mathematical expression

[0023] Construct a second-order equivalent circuit model of a lithium battery. In the second-order equivalent circuit of a lithium battery, resistance R1-voltage C1 simulates nonlinear voltage drop; resistance R2-voltage C2 simulates ohmic drop; resistance R0 simulates the internal resistance of the battery;

[0024] According to Kirchhoff's law, the mathematical expression of the second-order equivalent circuit model of lithium battery is as follows:

[0025]

[0026] U T =U oc (SOC)-U1-U2-R0i(T)

[0027]

[0028] Among them, T is the sampling time, T0 is the reference time, U1 and U2 are the terminal voltages of R1 and R2, and U oc is the circuit open circuit voltage, i is the loop current, Q N is the rated capacity, U oc There is a functional relationship with the remaining capacity SOC of the battery. The basic formula is as follows:

[0029]

[0030] Where, OCV is the open circuit voltage, exp(·) is the exponential function, E is the activation energy, R is the gas constant, and Temp is the ref is the reference temperature, Temp is the operating temperature, ω is the order of the approximate polynomial, a k is the coefficient of each order before sorting, a p is the coefficient of each order after arrangement, c0 is the corresponding polynomial coefficient in a single charge and discharge cycle;

[0031] U1, U2 and SOC are selected as unobservable virtual sensor signals representing the health status of lithium batteries, i.e., target parameter signals of the feature extraction network;

[0032] The discrete mathematical expressions of the three signals U1, U2 and SOC are as follows:

[0033]

[0034] Among them, k represents the sampling time, SOC k is the remaining power at the sampling time, is the voltage of R1 at the sampling moment, is the voltage of R2 at the sampling time, SOC k-1 is the remaining power at the last sampling moment, is the voltage of R1 at the last sampling moment, is the voltage of R2 at the last sampling moment;

[0035] Step S22: Feature extraction network model-RNN based on recurrent neural network

[0036] The main body of the feature extraction network model-RNN network is the basic RNN network, which is mathematically expressed as:

[0037] h k =g(W hh h k-1 +W hx x k +b h )

[0038] y k =W yh h k +W yx x k

[0039] Among them, h k is the hidden state at that moment, h k-1 is the hidden state at the previous moment, x k Enter y for this moment. k Output at this moment, W hh , W hx , W yh , W yx 、b h are model training parameters, g(·) is the network model;

[0040] The internal structure of the RNN network is constrained according to the discrete equation of the equivalent circuit, and the model-RNN is used to extract unobservable feature signals;

[0041] According to the discrete mathematical expression of the lithium battery equivalent circuit model, the hidden state of the feature extraction network model-RNN is defined as:

[0042]

[0043] According to the discrete equation, W hh Defined as:

[0044]

[0045] Select the diagonal matrix constraint to embed the RNN network, the simplified W hh Defined as:

[0046]

[0047] Among them, ψ1, ψ2, and ψ3 are the matrix weights to be trained;

[0048] According to U oc The functional relationship between U and SOC uses the network model model-Uoc to model the third-order approximate function.

[0049] Furthermore, model-Uoc consists of a power exponential calculation layer, two fully connected layers, a random dropout layer, and an activation layer;

[0050] For the input sample in is a set of real numbers, batch_size is the number of training samples; after passing through the power exponential calculation layer, its output is:

[0051]

[0052] The input dimension of the first fully connected layer is 9 and the output dimension is 3. The specific formula is:

[0053]

[0054] Among them, W1 is the weight of the fully connected layer, b1 is the bias of the fully connected layer, and the output

[0055] Then the data is sent to the random deactivation function, the specific formula is:

[0056] y2=Dropout(y1)

[0057] The data is then fed into the second fully connected layer, whose input dimension is 3 and output dimension is 1, and the output is Finally, the data is activated. The specific formula is:

[0058] y4=ReLU(y3)

[0059] According to Kirchhoff's second law, a fully connected layer is used to calculate the output y of the RNN network unit u , specifically defined as:

[0060] y u =W u x u +b u

[0061] in, W u , b u are the weights and biases of the fully connected layer, and i is the input current.

[0062] Furthermore, in step S3, the method of constructing a network based on the PINN and the health indicator of the degradation model includes:

[0063] Step S31, establishment of degradation dynamic model

[0064] Use multivariate functions to model general degradation processes:

[0065] u=f(t,x)

[0066] Where t is the number of discharge cycles, x is the vector constituting HI, and u is the constructed HI;

[0067] The degradation dynamics equation of the system is expressed as:

[0068]

[0069] Where g(·) is the explicit partial differential equation of the degradation dynamics parameterized by θ, which is a nonlinear function of t, x, and u. Based on the explicit partial differential equation, a generalized function approximator G(·) is defined to characterize the nonlinear degradation dynamics equation. The specific equation is as follows:

[0070] u t ≈G(t,x,u,u t ,u x ,u xx ,…;Θ)

[0071] in, Θ is the network parameter of G(·);

[0072] Based on the DeepHPM algorithm model, the lithium battery kinetic equation is modeled and a generalized function approximator is defined to characterize the nonlinear degradation kinetic equation. The specific equation is as follows:

[0073] u t -DeepHPM(t,x,u,u t ,u x ,u xx ,…;Θ)=0

[0074] Where Θ represents the trainable model parameters of the DeepHPM network model;

[0075] Step S32: constructing a network based on the health indicators of the PINN and degradation model

[0076] The health indicator construction network based on PINN and degradation model mainly includes three modules: health indicator HI construction network N(·), degradation dynamics simulation network G(·), and degradation model embedding network E(·);

[0077] The input dimension of network N(·) is 5, the output dimension is 1, and it consists of two fully connected layers and one sinusoidal nonlinear layer. The input dimension of network G(·) is 11, the output dimension is 1, and it consists of two fully connected layers and one sinusoidal nonlinear layer.

[0078] According to the empirical law of lithium battery degradation process, an exponential degradation model is used to constrain the construction trend of HI. The exponential degradation model is defined as:

[0079]

[0080] Among them, z is the degradation index, φ, θ, α, β are degradation-related parameters, and ε is random noise;

[0081] The corresponding exponential degradation kinetic equation is characterized as:

[0082]

[0083] The network E(·) is designed according to the exponential degradation dynamics equation, which uses three training parameters to constrain the dynamic changes of HI and continuously adjusts the direction of exponential fitting during the training process.

[0084] Furthermore, step S3 also includes training a network model constructed based on the health indicators of the PINN and degradation model, the method being as follows:

[0085] Step S331: Loss function definition

[0086] According to step S2, the feature extraction network based on physical knowledge is essentially an RNN model, which simulates the output voltage of the circuit through the current signal and the network structure embedded in the equivalent model, and is trained using a loss function based on mean square error. The specific equation of the loss function is as follows:

[0087]

[0088] in, is the output voltage regression loss of the lithium battery equivalent circuit, U k is the actual output voltage, is the output voltage predicted by the algorithm, Φ is the network parameter of the model, N is the total number of training samples, is a set of non-negative integers;

[0089] According to step S3, the health indicator-based network constructed based on PINN and the degradation model includes three sub-networks, and the corresponding loss function is constructed based on MSE, which is specifically expressed as follows:

[0090]

[0091] in, is the first-order partial differential regression loss function of the health indicator degradation dynamics, is the regression loss function of the exponential degradation model of health indicator degradation dynamics, Θ is the network parameter of the G(·) model, and Ψ is the network parameter of the E(·) model.

[0092] Step S332: Model training plan

[0093] The training steps of the unsupervised health assessment and remaining life prediction algorithm of lithium batteries based on physical knowledge include two parts: pre-training of feature extraction network based on physical knowledge, and joint training of feature extraction network and health indicator construction algorithm;

[0094] In the pre-training of the feature extraction network, the optimization goal is:

[0095]

[0096] In the joint training of the feature extraction network and the health indicator construction algorithm, the optimization goal is:

[0097]

[0098] Among them, λ1, λ2, λ3, λ4 are loss functions The corresponding weight coefficient is, is a monotonicity constraint; the lithium battery health assessment model is trained according to the optimization objective, and the influence of different loss functions on HI construction is balanced by adjusting the weight coefficient, and finally a stable health indicator that can characterize the degradation characteristics of lithium batteries is constructed.

[0099] Furthermore, in step S4, the health assessment and remaining life prediction specifically include:

[0100] Step S41: Health status estimation and assessment

[0101] Input the corresponding sensor signal of the test data into the feature extraction network based on physical knowledge involved in step S2 to obtain an unobservable sensor degradation signal based on the lithium battery equivalent circuit model, and use the health indicator construction network based on PINN and the degradation model involved in step S3 to complete the construction of the lithium battery health indicator and perform the lithium battery health status assessment; the indicators of the health indicator include monotonicity, trend, correlation, robustness, threshold consistency, and remaining life prediction accuracy;

[0102] Step S42: Remaining life prediction and evaluation

[0103] Based on the historical data of the full life cycle of multiple lithium batteries, corresponding health indicators are constructed, and appropriate failure thresholds are set based on the test time and trend characteristics of multiple health indicators combined with empirical rules; for the test set health indicators involved in step S41, the corresponding remaining life of the test unit is calculated according to the selected failure threshold, and the accuracy of the remaining life prediction is used to further illustrate the effectiveness of the algorithm.

[0104] The present invention discloses a method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge, which has the following beneficial effects:

[0105] The present invention takes into account the two major advantages of physical interpretability and model versatility. It realizes the construction of health indicators with physical significance through a data-driven model embedded in physical units, and achieves good prediction results on small-scale data sets. The algorithm space of the HI construction network is enhanced by using a feature extraction algorithm based on physical knowledge, and the purposeful extraction of degradation feature signals is achieved by integrating the main physical characteristics of the lithium battery performance model; the health status of lithium batteries is evaluated using a health indicator construction algorithm based on the PINN algorithm framework. In the absence of degradation signal label values, the extracted degradation feature signals are combined with the DeepHPM model to obtain health indicators with system failure dynamics, which reduces the dependence on data sets while improving the interpretability and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 is a flow chart of the method of the present invention;

[0107] Figure 2 A schematic diagram of a physical model-based feature extraction network and a PINN-based health indicator construction network of the present invention;

[0108] Figure 3 It is a schematic diagram of the second-order equivalent circuit of the present invention. DETAILED DESCRIPTION

[0109] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0110] The algorithm proposed in the present invention guides the construction of a feature extraction network according to the lithium battery performance model, and purposefully extracts characteristic signals with degradation mechanisms; a health indicator construction network integrating physical knowledge is constructed according to the PINN algorithm framework, and health indicators with system failure dynamics are obtained by integrating the exponential degradation model and the DeepHPM model, thereby reducing the dependence on the size of the data set while improving the interpretability and generalization ability of the model.

[0111] refer to Figure 1 The present invention discloses a method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge, comprising the following steps:

[0112] Step S1, data preprocessing:

[0113] Collect relevant sensor data in lithium battery charging and discharging experiments and perform data preprocessing;

[0114] Step S2, feature extraction network based on physical model:

[0115] The mathematical expression of the lithium battery mechanism model is used to guide the construction of the feature extraction network, extract unobservable signals containing the system degradation law, and enhance the input space of the health assessment and remaining life prediction algorithm;

[0116] Step S3, constructing a network based on the health indicators of PINN:

[0117] Based on the dynamic process of the lithium battery mechanism model and the dynamic expression of the degradation model, a PINN-based health assessment and remaining life prediction algorithm is constructed; the enhanced feature data is input into the PINN-based health assessment and remaining life prediction algorithm to optimize the model parameters for health indicator construction;

[0118] Step S4, health assessment and remaining life expectancy prediction:

[0119] Characteristic signals with physical significance are extracted based on the test data, and on this basis, health indicators integrating physical knowledge are constructed, ultimately achieving remaining life estimation based on health indicators.

[0120] To further optimize the technical solution, in step S1, the 2-sigma principle is used to preprocess the data, including:

[0121] Step S10, dividing the input sensor data into different groups of data according to a sliding window with a window length of len and a window interval of inter;

[0122] Step S11, calculating the mean μ and standard deviation σ of each group of data;

[0123] Step S12, according to the 2-sigma principle, the probability that the data is distributed in [μ-2σ, μ+2σ] is 95.45%. Data points falling in the interval [μ-2σ, μ+2σ] are normal data and are retained; data points falling outside [μ-2σ, μ+2σ] are outliers and are eliminated.

[0124] To further optimize the technical solution, step S1 further includes: first performing Min-Max normalization processing on normal data, and then performing data division processing.

[0125] The specific framework of the algorithm of the present invention is as follows Figure 2 As shown. The main body of the algorithm consists of two parts, including a feature extraction network based on the physical model and a health indicator construction network based on PINN and degradation model. 1) Feature extraction network based on physical model: The discrete mathematical model of lithium battery is derived according to the second-order equivalent circuit model of lithium-ion battery, and the mathematical model is used to guide the node architecture and network weight design in the recurrent neural network (RNN), extracting unobservable virtual sensor signals with lithium battery health status representation to enhance the input space of the health indicator construction network and improve the physical interpretability of the health indicators constructed subsequently. 2) Health indicator construction network based on PINN and degradation model: According to the basic structural composition of PINN, a health indicator construction algorithm integrating degradation model is constructed. Specifically, the relevant dynamic expressions in the lithium battery equivalent model are used to guide the construction of differential operators, the DeepHPM network is used to simulate the degradation dynamics of the lithium battery equivalent model, and the degradation model is used to constrain the health indicator construction method.

[0126] To further optimize the technical solution, in step S2, the specific method of extracting the feature network based on the physical model is:

[0127] Step S21, lithium battery equivalent circuit model and mathematical expression

[0128] The second-order equivalent circuit model of lithium-ion batteries is mainly constructed based on empirical knowledge, and its basic components include variable voltage sources, resistors, and capacitors. Figure 3 As shown, the resistor R1-voltage C1 simulates nonlinear voltage drop; the resistor R2-voltage C2 simulates ohmic drop; the resistor R0 simulates the internal resistance of the battery;

[0129] According to Kirchhoff's law, the mathematical expression of the second-order equivalent circuit model of lithium battery is as follows:

[0130]

[0131] U T =U oc (SOC)-U1-U2-R0i(T)

[0132]

[0133] Among them, T is the sampling time, T0 is the reference time, U1 and U2 are the terminal voltages of R1 and R2, and U oc is the circuit open circuit voltage, i is the loop current, Q N is the rated capacity, U oc There is a functional relationship with the remaining capacity SOC (state of charge) of the battery. The basic formula is as follows:

[0134]

[0135] Where, OCV is the open circuit voltage, exp(·) is the exponential function, E is the activation energy, R is the gas constant, and Temp is the ref is the reference temperature, Temp is the operating temperature, ω is the order of the approximate polynomial, a p is the coefficient of each order, a k is the coefficient of each order before sorting, a p is the coefficient of each order after sorting, and c0 is the corresponding polynomial coefficient in a single charge and discharge cycle. In actual operation, the present invention selects a fourth-order polynomial to approximate the open circuit voltage. ;

[0136] According to the second-order equivalent circuit model of the lithium battery, the three signals U1, U2 and SOC are closely related to the overall operation status of the circuit (SOC is related to the open circuit voltage). Therefore, the present invention selects U1, U2 and SOC as unobservable virtual sensor signals with lithium battery health status representation, that is, the target parameter signals of the feature extraction network;

[0137] Due to the discrete nature of the collected data, the formula needs to be discretized when analyzing the mathematical model. The discrete mathematical expressions of the three signals U1, U2 and SOC are as follows:

[0138]

[0139] Among them, k represents the sampling time, SOC k is the remaining power at the sampling time, is the voltage of R1 at the sampling moment, is the voltage of R2 at the sampling time, SOC k-1 is the remaining power at the last sampling moment, is the voltage of R1 at the last sampling moment, is the voltage of R2 at the last sampling moment;

[0140] Step S22: Feature extraction network model-RNN based on recurrent neural network

[0141] The main body of the feature extraction network model-RNN network is the basic RNN network, which is mathematically expressed as:

[0142] h k =g(W hh h k-1 +W hx x k +b h )

[0143] y k =W yh h k +W yx x k

[0144] Among them, h k is the hidden state at that moment, h k-1 is the hidden state at the previous moment, x k Enter y for this moment. k Output at this moment, W hh , W hx , W yh , W yx 、b h are model training parameters, g(·) is the network model;

[0145] According to the discrete equation of the second-order equivalent circuit of the lithium battery, its composition structure is basically consistent with the mathematical expression of RNN. Based on the above characteristics, the present invention constrains the internal structure of the RNN network according to the discrete equation of the equivalent circuit, and uses model-RNN to extract unobservable feature signals;

[0146] According to the discrete mathematical expression of the lithium battery equivalent circuit model, the hidden state of the feature extraction network model-RNN is defined as:

[0147]

[0148] According to the discrete equation, W hh Defined as:

[0149]

[0150] In order to reduce the dependence of network training on model parameter identification and improve the versatility of feature extraction networks, it is necessary to simplify the constraints. hh The matrix contains multiple important physical features, among which the diagonal matrix is ​​its inherent feature. Therefore, the diagonal matrix constraint is selected to embed the RNN network. The simplified W hh Defined as:

[0151]

[0152] Among them, ψ1, ψ2, and ψ3 are the matrix weights to be trained;

[0153] According to U oc The functional relationship between U and SOC uses the network model model-Uoc to model the third-order approximate function.

[0154] To further optimize the technical solution, model-Uoc consists of a power exponential calculation layer, two fully connected layers (Linear), a random loss layer (Dropout), and an activation layer (Relu);

[0155] For the input sample in is a set of real numbers, batch_size is the number of training samples; after passing through the power exponential calculation layer, its output is:

[0156]

[0157] The input dimension of the first fully connected layer is 9 and the output dimension is 3. The specific formula is:

[0158]

[0159] Among them, W1 is the weight of the fully connected layer, b1 is the bias of the fully connected layer, and the output

[0160] Then the data is sent to the random deactivation function, the specific formula is:

[0161] y2=Dropout(y1)

[0162] The data is then fed into the second fully connected layer, whose input dimension is 3 and output dimension is 1, and the output is Finally, the data is activated. The specific formula is:

[0163] y4=ReLU(y3)

[0164] According to Kirchhoff's second law, a fully connected layer is used to calculate the output y of the RNN network unit u , specifically defined as:

[0165] y u =W u x u +b u

[0166] in, W u , b uare the weights and biases of the fully connected layer, and i is the input current.

[0167] To further optimize the technical solution, in step S3, the method for constructing a network based on the health indicators of the PINN and the degradation model includes:

[0168] Step S31, establishment of degradation dynamic model

[0169] In lithium battery systems, the failure trajectory is affected by many factors including charging rate, discharging rate, temperature, time, etc. Therefore, a multivariate function can be used to model the general degradation process:

[0170] u=f(t,x)

[0171] Where t is the number of discharge cycles, x is the vector constituting HI, and u is the constructed HI;

[0172] In order to characterize the common characteristics of the degradation process in different test units of the same system and avoid the influence of different initial states on the modeling of the degradation process, this paper focuses on the degradation dynamics of the system. In general, the degradation dynamics equation of the system is expressed as:

[0173]

[0174] Among them, g(·) is the explicit partial differential equation of the degradation dynamics parameterized by θ, which is a nonlinear function of t, x, and u. Furthermore, a generalized function approximator G(·) can be defined based on the explicit partial differential equation to characterize the nonlinear degradation dynamics equation. The specific equation is as follows:

[0175] u t ≈G(t,x,u,u t ,u x ,u xx ,…;Θ)

[0176] in, Θ is the network parameter of G(·). For the order of the input partial differential variable in the approximate equation G(·), the first-order differential operator is used for calculation according to the accuracy and computational complexity of the lithium battery degradation model.

[0177] Generally speaking, the explicit equation of G(·) in the function approximator is usually unknown and often difficult to obtain through mathematical calculations. In order to approximate G(·), the present invention models the lithium battery kinetic equation based on the DeepHPM algorithm model and defines a generalized function approximator to characterize the nonlinear degradation kinetic equation. The specific equation is as follows:

[0178] u t -DeepHPM(t,x,u,u t ,ux ,u xx ,…;Θ)=0

[0179] Where Θ represents the trainable model parameters of the DeepHPM network model;

[0180] Step S32: constructing a network based on the health indicators of the PINN and degradation model

[0181] The health indicator construction network based on PINN and degradation model mainly includes three modules: health indicator HI construction network N(·), degradation dynamics simulation network G(·), and degradation model embedding network E(·);

[0182] The input dimension of network N(·) is 5, the output dimension is 1, and it consists of two fully connected layers and one sinusoidal nonlinear layer. The input dimension of network G(·) is 11, the output dimension is 1, and it consists of two fully connected layers and one sinusoidal nonlinear layer.

[0183] According to the empirical law of lithium battery degradation process, the present invention uses an exponential degradation model to constrain the construction trend of HI. The exponential degradation model is defined as:

[0184]

[0185] Among them, z is the degradation index, φ, θ, α, β are degradation-related parameters, and ε is random noise;

[0186] The corresponding exponential degradation kinetic equation is characterized as:

[0187]

[0188] The network E(·) is designed according to the exponential degradation dynamics equation, which uses three training parameters to constrain the dynamic changes of HI and continuously adjusts the direction of exponential fitting during the training process.

[0189] Furthermore, step S3 also includes training a network model constructed based on the health indicators of the PINN and degradation model, the method being as follows:

[0190] Step S331: Loss function definition

[0191] The unsupervised health assessment and remaining life prediction algorithm of lithium batteries based on physical knowledge includes two algorithm modules, one is a feature extraction network based on physical knowledge, and the other is a health indicator construction network based on PINN and degradation model.

[0192] According to step S2, the feature extraction network based on physical knowledge is essentially an RNN model, which simulates the output voltage of the circuit through the current signal and the network structure embedded in the equivalent model, and is trained using a loss function based on mean square error (MSE). The specific equation of the loss function is as follows:

[0193]

[0194] in, is the output voltage regression loss of the lithium battery equivalent circuit, U k is the actual output voltage, is the output voltage predicted by the algorithm, Φ is the network parameter of the model, N is the total number of training samples, is a set of non-negative integers;

[0195] According to step S3, the health indicator construction network based on PINN and degradation model includes three sub-networks, among which G(·) is used to constrain the degradation dynamics equation in the lithium battery equivalent circuit model, and E(·) is used to constrain the construction law of health indicator HI. The corresponding loss function is constructed based on MSE, which is specifically expressed as follows:

[0196]

[0197] in, is the first-order partial differential regression loss function of the health indicator degradation dynamics, is the regression loss function of the exponential degradation model of health indicator degradation dynamics, Θ is the network parameter of the G(·) model, and Ψ is the network parameter of the E(·) model.

[0198] Step S332: Model training plan

[0199] The training steps of the unsupervised health assessment and remaining life prediction algorithm of lithium batteries based on physical knowledge include two parts: pre-training of feature extraction network based on physical knowledge, and joint training of feature extraction network and health indicator construction algorithm;

[0200] In the pre-training of the feature extraction network, the optimization goal is:

[0201]

[0202] In the joint training of the feature extraction network and the health indicator construction algorithm, the optimization goal is:

[0203]

[0204] Among them, λ1, λ2, λ3, λ4 are loss functions The corresponding weight coefficient is, is a monotonicity constraint to ensure the monotonicity of HI. The lithium battery health assessment model is trained according to the optimization objective, and the influence of different loss functions on HI construction is balanced by adjusting the weight coefficient, and finally a stable health indicator that can characterize the degradation characteristics of lithium batteries is constructed.

[0205] Furthermore, in step S4, the health assessment and remaining life prediction specifically include:

[0206] Step S41: Health status estimation and assessment

[0207] Input the corresponding sensor signal of the test data into the feature extraction network based on physical knowledge involved in step S2 to obtain an unobservable sensor degradation signal based on the lithium battery equivalent circuit model, and use the health indicator construction network based on PINN and the degradation model involved in step S3 to complete the construction of the lithium battery health indicator and perform the lithium battery health status assessment; the indicators of the health indicator include monotonicity, trend, correlation, robustness, threshold consistency, and remaining life prediction accuracy;

[0208] In order to illustrate the superiority of health indicators HI, the present invention designs a health indicator evaluation scheme based on the basic characteristics of HI, and uses multiple indicators including monotonicity, trend, correlation, robustness, threshold consistency, and remaining life prediction accuracy to comprehensively evaluate the rationality of health indicators.

[0209] Step S42: Remaining life prediction and evaluation

[0210] Based on the historical data of the full life cycle of multiple lithium batteries, corresponding health indicators are constructed, and appropriate failure thresholds are set based on the test time and trend characteristics of multiple health indicators combined with empirical rules; for the test set health indicators involved in step S41, the corresponding remaining life of the test unit is calculated according to the selected failure threshold, and the accuracy of the remaining life prediction is used to further illustrate the effectiveness of the algorithm.

[0211] The present invention takes into account the two major advantages of physical interpretability and model versatility. It realizes the construction of health indicators with physical significance through a data-driven model embedded in physical units, and achieves good prediction results on small-scale data sets. The algorithm space of the HI construction network is enhanced by using a feature extraction algorithm based on physical knowledge, and the purposeful extraction of degradation feature signals is achieved by integrating the main physical characteristics of the lithium battery performance model; the health status of lithium batteries is evaluated using a health indicator construction algorithm based on the PINN algorithm framework. In the absence of degradation signal label values, the extracted degradation feature signals are combined with the DeepHPM model to obtain health indicators with system failure dynamics, which reduces the dependence on data sets while improving the interpretability and generalization ability of the model.

[0212] Example

[0213] The specific effect of this embodiment is demonstrated and verified using an open source dataset as a case.

[0214] 1. Open Source Datasets

[0215] The present invention uses the CALCE lithium-ion battery data set released by the University of Maryland as a verification data set for the algorithm. The CALCE data set is a data set established by the University of Maryland to evaluate the health status of lithium batteries and conduct SOC prediction, SOH evaluation, and remaining life prediction and other related algorithm research. The CALCE data is generated based on the real constant current-constant voltage (CC-CV) charge and discharge cycle experiment of the CS2 battery. All experiments were carried out using the same standard, that is, using a constant current of 0.5C to charge to 4.2V, and then using a constant voltage of 4.2V to charge until the charging current drops to 0.05A; after charging, discharge with a constant current of 1C until the terminal voltage drops to 2.7V.

[0216] 2. Dataset processing

[0217] In order to maintain the consistency of training data, only discharge cycle data is used for training and result verification in the specific implementation plan. The CS2-33, CS2-34, and CS2-35 battery data sets are used as training sets, denoted as D = {T, I, U, t, OCV, Q}, where T represents the sampling time of the data, I represents the loop current, U represents the port voltage, t represents the discharge cycle, OCV represents the open circuit voltage, and Q represents the remaining power.

[0218] In order to reduce the interference of noise data and improve the quality of the data set, it is necessary to clean the original data set of outliers. The algorithm uses the 2-sigma principle to process the data. The 2-sigma principle is a mathematical concept developed based on the normal distribution. It assumes that a set of data to be tested contains only random errors, and determines an interval through the normal distribution. Data exceeding this interval is considered to be an outlier and should be removed. The specific steps for outlier cleaning based on the 2-sigma principle are as follows:

[0219] ① Divide the input data into different groups of data according to the sliding window with window length len and window interval inter.

[0220] ② Calculate the mean μ and standard deviation σ of each group of data;

[0221] ③According to the 2-sigma principle, the probability that the data is distributed in [μ-2σ,μ+2σ] is 95.45%. The data points falling in the interval [μ-2σ,μ+2σ] are normal data and are retained; the data points falling outside [μ-2σ,μ+2σ] are outliers and are eliminated.

[0222] The feature data collected by different sensors have different dimensions and dimensional units, and their physical meanings vary greatly. If they are directly used in the subsequent feature extraction algorithm, deviations will occur. In order to eliminate the influence of different dimensions, it is necessary to scale the data so that each indicator is at the same order of magnitude, which is convenient for comprehensive comparative calculations. The data normalization method used in this section is Min-Max normalization (Min-Max, Normalization), also known as deviation normalization, which can map data samples to between [0,1]. The calculation formula of Min-Max normalization is as follows:

[0223]

[0224] Assume that the sensor signal to be processed is X, X max , X min Indicates the maximum and minimum values ​​of the data in X, X std Represents normalized data.

[0225] The training set after outlier cleaning and data normalization is D train ={T train ,I train ,U train ,t train ,OCV train ,Q train}.

[0226] According to the algorithm requirements, D train Divide into X train ={T train ,I train},Y train = {U train}, P train ={t train ,OCV train ,Q train}, where X train Input data to the network, Y train is the label data of the feature extraction network, P train It is the operation status data.

[0227] The CS2-36, CS2-37, and CS2-38 battery data sets are used as test sets. After outlier cleaning, data normalization, and data partitioning, we get X test ={T test ,I test},Y test = {U test}, P test ={t test ,OCV test ,Q test}.

[0228] 3. Model building and model training

[0229] According to steps 1 and 2, build an unsupervised health assessment and remaining life prediction algorithm for lithium batteries based on physical knowledge.

[0230] The input signal of the physical model-based feature extraction network model-RNN is X train With the hidden state h0, according to the second-order equivalent circuit model of lithium battery, the hidden dimension of RNN is set to 3. The output signal of the network is hidden, By minimizing With Y train The network is trained using the MSE loss function.

[0231] The input of the network constructed based on the health indicators of PINN and degradation model includes X train , hidden state h0 and P train , where X train and h0 are used to calculate hidden. train t in train and X train I train The 5-dimensional input of N(·) is composed of the health indicator HI constructed by the algorithm, denoted as u train Calculate u train The partial derivative of the input signal with respect to N(·) and the input signal and u train The 11-dimensional data are combined into the network G(·) to simulate the dynamic characteristics of lithium battery degradation. The output of the network G(·) is By minimizing and The MSE loss function constrains the network. The input signal of the network E(·) includes t train with u train , whose output signal is E, by minimizing E and The MSE loss function implements the exponential trend constraint on health indicators.

[0232] According to the above steps, the model is constructed and the model training is performed. First, the model-RNN is pre-trained. In the pre-training, only CS2-33 is used as the training set. The initial weight of the model-RNN is obtained through pre-training to accelerate the training process and improve the training efficiency. Based on the pre-training, the model-RNN and model-PINN are jointly trained, and the CS2-33, CS2-34, and CS2-35 battery data sets are used as training sets to obtain the health assessment and remaining life prediction algorithm model.

[0233] 4. Health status estimation

[0234] According to the training data set D train The obtained u train Calculate the corresponding failure threshold. Here, the 650th cycle of the lithium battery charge and discharge cycle is selected as the failure threshold. Input test data set D test , calculate the corresponding u test .

[0235] 5. Performance evaluation indicators

[0236] In order to illustrate the superiority of the health indicator constructed by the algorithm proposed in the present invention, the present invention designs corresponding performance evaluation indicators according to the basic attributes of the health indicator HI.

[0237] The basic attributes of health indicators HI include:

[0238] (1) After the degradation process begins, the trend of HI is monotonically decreasing;

[0239] (2) After the degradation process begins, the trend of HI is closely related to the running time;

[0240] (3) Under the same environmental influence and the same degradation mode, the variance of the failure thresholds between different units calculated according to HI is the smallest;

[0241] (4) In order to reduce the data fluctuation and uncertainty caused by the slight differences between different units of the same model in HI construction, HI needs to have good versatility;

[0242] (5) The remaining life expectancy predicted based on health indicators should be as close to the actual value as possible.

[0243] According to the above four basic properties, corresponding performance evaluation indicators are proposed, namely: monotonicity, trend, failure threshold variance, robustness, and accuracy of RUL prediction.

[0244] Health indicators for builds Where m is the total number of discharge cycles contained in a test unit, and the specific calculation formula for monotonicity is as follows:

[0245]

[0246] The specific calculation formula of trend is as follows:

[0247]

[0248] The specific calculation formula of the failure threshold variance is as follows:

[0249]

[0250] Where H=[H1,H2,…,H N ] is the set of health indicator vectors in the training set, N is the total number of training set units, represents the failure threshold of the qth training unit.

[0251] The specific calculation formula of robustness is as follows:

[0252]

[0253] in, It is the residual between H and its mean trend, which is generally obtained by smoothing fitting and other methods.

[0254] The specific calculation formula for RUL prediction accuracy is as follows:

[0255]

[0256] Among them, RUL 1:M =[l1,l2,…,l M ] is the remaining life prediction vector set of each test set unit, is the predicted life of the rth unit, l r is the true life of the rth unit, M is the total number of test set units,

[0257] In order to illustrate the superiority of the proposed unsupervised health indicator construction algorithm based on physical knowledge, based on the unsupervised and physically meaningful characteristics of the algorithm, the commonly used health indicator SOC of lithium batteries was selected for comparison. According to the five indicators of monotonicity, trend, failure threshold variance, robustness, and RUL prediction accuracy proposed above, it is shown that the HI constructed by the algorithm has good performance in stability, versatility, and remaining life prediction.

[0258] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge, characterized in that: The steps include: Step S1, data preprocessing: Collect relevant sensor data in lithium battery charging and discharging experiments and perform data preprocessing; Step S2, feature extraction network based on physical model: The mathematical expression of the lithium battery mechanism model is used to guide the construction of the feature extraction network, extract unobservable signals containing the system degradation law, and enhance the input space of the health assessment and remaining life prediction algorithm; Step S3, constructing a network based on the health indicators of PINN: Based on the dynamic process of the lithium battery mechanism model and the dynamic expression of the degradation model, a PINN-based health assessment and remaining life prediction algorithm is constructed; the enhanced feature data is input into the PINN-based health assessment and remaining life prediction algorithm to optimize the model parameters for health indicator construction; Step S4, health assessment and remaining life prediction: Characteristic signals with physical significance are extracted based on the test data, and on this basis, health indicators integrating physical knowledge are constructed, ultimately achieving remaining life estimation based on health indicators.

2. According to the physical knowledge-based unsupervised health assessment and remaining life prediction method of lithium batteries in claim 1, it is characterized in that: In step S1, the data is preprocessed using the 2-sigma principle, including: Step S10, dividing the input sensor data into different groups of data according to a sliding window with a window length of len and a window interval of inter; Step S11, calculating the mean μ and standard deviation σ of each group of data; Step S12, according to the 2-sigma principle, data points falling in the interval [μ-2σ, μ+2σ] are normal data and are retained; data points falling outside [μ-2σ, μ+2σ] are outliers and are eliminated.

3. According to the physical knowledge-based unsupervised health assessment and remaining life prediction method of lithium batteries in claim 2, it is characterized in that: Step S1 also includes: first performing Min-Max normalization processing on normal data, and then performing data division processing.

4. According to the physical knowledge-based unsupervised health assessment and remaining life prediction method of lithium batteries in claim 3, it is characterized in that: In step S2, the specific method of the feature extraction network based on the physical model is: Step S21, lithium battery equivalent circuit model and mathematical expression Construct a second-order equivalent circuit model of a lithium battery. In the second-order equivalent circuit of a lithium battery, resistance R1-voltage C1 simulates nonlinear voltage drop; resistance R2-voltage C2 simulates ohmic drop; resistance R0 simulates the internal resistance of the battery; According to Kirchhoff's law, the mathematical expression of the second-order equivalent circuit model of lithium battery is as follows: U T =U oc (SOC)-U1-U2-R0i(T) Among them, T is the sampling time, T0 is the reference time, U1 and U2 are the terminal voltages of R1 and R2, and U oc is the circuit open circuit voltage, i is the loop current, Q N is the rated capacity, U oc There is a functional relationship with the remaining capacity SOC of the battery. The basic formula is as follows: Where, OCV is the open circuit voltage, exp(·) is the exponential function, E is the activation energy, R is the gas constant, and Temp is the ref is the reference temperature, Temp is the operating temperature, ω is the order of the approximate polynomial, a k is the coefficient of each order before sorting, a p is the coefficient of each order after arrangement, c0 is the corresponding polynomial coefficient in a single charge and discharge cycle; U1, U2 and SOC are selected as unobservable virtual sensor signals representing the health status of lithium batteries, i.e., target parameter signals of the feature extraction network; The discrete mathematical expressions of the three signals U1, U2 and SOC are as follows: Among them, k represents the sampling time, SOC k is the remaining power at the sampling time, is the voltage of R1 at the sampling moment, is the voltage of R2 at the sampling time, SOC k-1 is the remaining power at the last sampling moment, is the voltage of R1 at the last sampling moment, is the voltage of R2 at the last sampling moment; Step S22: Feature extraction network model-RNN based on recurrent neural network The main body of the feature extraction network model-RNN network is the basic RNN network, which is mathematically expressed as: h k =g(W hh h k-1 +W hxx k +b h ) y k =W yh h k +W yx x k Among them, h k is the hidden state at that moment, h k-1 is the hidden state at the previous moment, x k Enter y for this moment. k Output at this moment, W hh , W hx , W yh , W yx 、b h are model training parameters, g(·) is the network model; The internal structure of the RNN network is constrained according to the discrete equation of the equivalent circuit, and the model-RNN is used to extract unobservable feature signals; According to the discrete mathematical expression of the lithium battery equivalent circuit model, the hidden state of the feature extraction network model-RNN is defined as: According to the discrete equation, W hh Defined as: Select the diagonal matrix constraint to embed the RNN network, the simplified W hh Defined as: Among them, ψ1, ψ2, and ψ3 are the matrix weights to be trained; According to U oc The functional relationship between U and SOC uses the network model model-Uoc to model the third-order approximate function.

5. According to the physical knowledge-based unsupervised health assessment and remaining life prediction method of lithium batteries in claim 4, it is characterized in that: Model-Uoc consists of a power exponential calculation layer, two fully connected layers, a random dropout layer, and an activation layer; For the input sample in is a set of real numbers, batch_size is the number of training samples; after passing through the power exponential calculation layer, its output is: The input dimension of the first fully connected layer is 9 and the output dimension is 3. The specific formula is: Among them, W1 is the weight of the fully connected layer, b1 is the bias of the fully connected layer, and the output Then the data is sent to the random deactivation function, the specific formula is: y2=Dropout(y1) The data is then fed into the second fully connected layer, whose input dimension is 3 and output dimension is 1, and the output is Finally, the data is activated. The specific formula is: y4=ReLU(y3) According to Kirchhoff's second law, a fully connected layer is used to calculate the output y of the RNN network unit u , specifically defined as: y u =W u x u +b u in, W u , b u are the weights and biases of the fully connected layer, and i is the input current.

6. The method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge according to claim 5 is characterized in that: In step S3, the method of constructing a network based on the PINN and the health indicator of the degradation model includes: Step S31, establishment of degradation dynamic model Use multivariate functions to model general degradation processes: u=f(t,x) Where t is the number of discharge cycles, x is the vector constituting HI, and u is the constructed HI; The degradation dynamics equation of the system is expressed as: Where g(·) is the explicit partial differential equation of the degradation dynamics parameterized by θ, which is a nonlinear function of t, x, and u. Based on the explicit partial differential equation, a generalized function approximator G(·) is defined to characterize the nonlinear degradation dynamics equation. The specific equation is as follows: u t ≈G(t,x,u,u t ,u x ,u xx ,…;Θ) in, Θ is the network parameter of G(·); Based on the DeepHPM algorithm model, the lithium battery kinetic equation is modeled and a generalized function approximator is defined to characterize the nonlinear degradation kinetic equation. The specific equation is as follows: her t -DeepHPM(t,x,u,u t ,her x ,her xx ,...;Θ)=0 Where Θ represents the trainable model parameters of the DeepHPM network model; Step S32: constructing a network based on the health indicators of the PINN and degradation model The health indicator construction network based on PINN and degradation model mainly includes three modules: health indicator HI construction network N(·), degradation dynamics simulation network G(·), and degradation model embedding network E(·); The input dimension of network N(·) is 5, the output dimension is 1, and it consists of two fully connected layers and one sinusoidal nonlinear layer. The input dimension of network G(·) is 11, the output dimension is 1, and it consists of two fully connected layers and one sinusoidal nonlinear layer. According to the empirical law of lithium battery degradation process, an exponential degradation model is used to constrain the construction trend of HI. The exponential degradation model is defined as: Among them, z is the degradation index, φ, θ, α, β are degradation-related parameters, and ε is random noise; The corresponding exponential degradation kinetic equation is characterized as: The network E(·) is designed according to the exponential degradation dynamics equation, which uses three training parameters to constrain the dynamic changes of HI and continuously adjusts the direction of exponential fitting during the training process.

7. The method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge according to claim 6 is characterized in that: Step S3 also includes training a network model constructed based on the health indicators of the PINN and degradation model, as follows: Step S331: Loss function definition According to step S2, the feature extraction network based on physical knowledge is essentially an RNN model, which simulates the output voltage of the circuit through the current signal and the network structure embedded in the equivalent model, and is trained using a loss function based on mean square error. The specific equation of the loss function is as follows: in, is the output voltage regression loss of the lithium battery equivalent circuit, U k is the actual output voltage, is the output voltage predicted by the algorithm, Φ is the network parameter of the model, N is the total number of training samples, is a set of non-negative integers; According to step S3, the health indicator-based network constructed based on PINN and the degradation model includes three sub-networks, and the corresponding loss function is constructed based on MSE, which is specifically expressed as follows: in, is the first-order partial differential regression loss function of the health indicator degradation dynamics, is the regression loss function of the exponential degradation model of health indicator degradation dynamics, Θ is the network parameter of the G(·) model, and Ψ is the network parameter of the E(·) model. Step S332: Model training plan The training steps of the unsupervised health assessment and remaining life prediction algorithm of lithium batteries based on physical knowledge include two parts: pre-training of feature extraction network based on physical knowledge, and joint training of feature extraction network and health indicator construction algorithm; In the pre-training of the feature extraction network, the optimization goal is: In the joint training of the feature extraction network and the health indicator construction algorithm, the optimization goal is: Among them, λ1, λ2, λ3, λ4 are loss functions The corresponding weight coefficient is, is a monotonicity constraint; the lithium battery health assessment model is trained according to the optimization objective, and the influence of different loss functions on HI construction is balanced by adjusting the weight coefficient, and finally a stable health indicator that can characterize the degradation characteristics of lithium batteries is constructed.

8. The method for unsupervised health assessment and remaining life prediction of lithium batteries based on physical knowledge according to claim 7 is characterized in that: In step S4, health assessment and remaining life prediction specifically include: Step S41: Health status estimation and assessment Input the corresponding sensor signal of the test data into the feature extraction network based on physical knowledge involved in step S2 to obtain an unobservable sensor degradation signal based on the lithium battery equivalent circuit model, and use the health indicator construction network based on PINN and the degradation model involved in step S3 to complete the construction of the lithium battery health indicator and perform the lithium battery health status assessment; the indicators of the health indicator include monotonicity, trend, correlation, robustness, threshold consistency, and remaining life prediction accuracy; Step S42: Remaining life prediction and evaluation Based on the historical data of the full life cycle of multiple lithium batteries, corresponding health indicators are constructed, and appropriate failure thresholds are set based on the test time and trend characteristics of multiple health indicators combined with empirical rules; for the test set health indicators involved in step S41, the corresponding remaining life of the test unit is calculated according to the selected failure threshold, and the accuracy of the remaining life prediction is used to further illustrate the effectiveness of the algorithm.

Images