Lithium ion battery health state lightweight detection method based on physical information neural network

Abstract

The invention provides a lithium ion battery health state lightweight detection method based on a physical information neural network, and the method comprises the steps: collecting the time, voltage, current, temperature and state-of-charge data of a battery in a takeoff and landing stage discharge process, processing the data into takeoff and landing stage discharge time sequence data, and carrying out the detection of the lithium ion battery health state based on the takeoff and landing stage discharge time sequence data. The method comprises the following steps: designing characteristic factors related to battery aging, screening the characteristic factors by utilizing a Pearson's correlation coefficient and a grey relational degree algorithm to obtain optimal characteristic sequence data, inputting the optimal characteristic sequence data into a physical information neural network model constructed by two serially connected neural networks for training, and in the training process, obtaining the optimal characteristic sequence data. And performing hyper-parameter tuning on the two neural networks by adopting a Bayesian optimization algorithm, then performing fine tuning on the second neural network by adopting a hierarchical transfer learning strategy, and finally applying the trained physical information neural network model to battery health state detection. The method improves the quality of feature data, reduces the calculation complexity of features and models, and achieves the accuracy and reliability of the detection of the health state of the battery under the airborne condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium battery, and particularly relates to a lithium ion battery health state lightweight detection method based on a physical information neural network. BACKGROUND

[0002] With the rapid development of low-altitude economy, electric vertical take-off and landing (eVTOL) aircraft are widely used in the fields of logistics transportation, emergency rescue, urban transportation, etc. Lithium ion batteries have become the main power source of eVTOL aircraft due to their advantages of high energy density, long cycle life and environmental friendliness. However, during the key flight stages such as take-off and landing, eVTOL aircraft not only puts forward strict requirements on high-rate discharge of power batteries, but also puts forward higher standards for the accuracy and reliability of battery management. Lithium ion batteries themselves are complex electrochemical systems, and the internal state evolution mechanism is highly nonlinear and difficult to directly observe, which makes it difficult to accurately estimate the state of health (SOH). At the same time, different aircraft work in different temperature conditions, different working modes and take different charge and discharge rates, which constitute the complex working conditions (referred to as working conditions) of the battery. These problems restrict the optimal design and application implementation of the battery management system, and more seriously affect the operation reliability and flight safety of the eVTOL aircraft. Therefore, there is an urgent need for a battery health state detection technology that is fast, accurate and lightweight for eVTOL aircraft.

[0003] The existing methods for evaluating battery health mainly include methods based on quantitative calculation, model-based and data-driven. However, the highly coupled thermal-electric- mass transfer inside the battery makes the degradation mechanism difficult to be directly calculated and quantified, so the method based on quantitative calculation needs to consume a large amount of computing resources to accurately obtain the aging parameters, which makes it difficult to popularize. The method based on physical model needs to have a large amount of battery mechanism related knowledge in advance, and be familiar with various physical models, but this method still leads to the accumulation of deviations across temperature and rate scenarios, which puts high requirements on the modeling personnel. At the same time, as the battery ages, the model parameters may change, leading to a weakening of the model's discriminability to parameters, resulting in a large error in the estimation result. In contrast, the data-driven method does not need explicit physical modeling, only needs to obtain features representing battery aging (such as voltage, current and temperature, etc.), and then uses artificial intelligence algorithms to run the aging features to accurately evaluate the state of health (SOH) of the battery. However, the data-driven method generally lacks explainability in the feature level and model inference process, making it difficult to establish an explainable and stable mapping with the aging mechanism. In addition, the battery historical data across types and working conditions are heterogeneous, and the battery capacity label is difficult to obtain, which makes the data inconsistency problem more prominent, making the model training and feature extraction more difficult.

[0004] In addition, the existing battery state of health estimation model is mainly based on the historical data of the vehicle-mounted battery, and only a small amount of research is directed to the estimation of the state of health (SOH) of the eVTOL aircraft battery. Unlike the traditional vehicle-mounted lithium battery operation data, the eVTOL aircraft needs high-rate discharge during takeoff and landing, so that the time series data obtained presents non-stationary, strong transient and multi-scale coupling characteristics. Based on these characteristics, the current mining of local mutations, recovery processes and common patterns across temperature and rate in the operation of eVTOL aircraft batteries is still insufficient, and the existing technology mainly has the following problems:

[0005] The existing battery state of health detection method is mainly applied to vehicle-mounted lithium batteries, which is suitable for low discharge rate and is difficult to apply to high-rate discharge, strong transient and non-stationary working conditions such as takeoff and landing of electric vertical takeoff and landing aircraft;

[0006] The traditional data-driven method usually has a complex model and a large number of parameters, which needs to consume a large amount of computing resources, which is not conducive to deployment on aircraft;

[0007] Due to manufacturing differences and different operating conditions, different single batteries have inconsistent aging trajectories, making it difficult for the model to be applied to all working condition batteries. SUMMARY

[0008] To solve the above problems, the present application provides a lithium ion battery health state lightweight detection method based on physical information neural network, through the data collected during the discharge process of the battery in the take-off and landing stage, the take-off and landing stage discharge timing data is designed, the feature factors related to battery aging are designed based on the timing data, and the feature factors are screened by using the Pearson correlation coefficient and the grey correlation degree algorithm, the optimal feature sequence data is obtained, the optimal feature sequence data is input into the physical information neural network model constructed by two series neural networks for training, in the training process, the hyperparameter optimization of the two neural networks is carried out by using the Bayesian optimization algorithm, after the first neural network is trained, the second neural network is fine-tuned by using the hierarchical transfer learning strategy, and finally the trained physical information neural network model is used for battery health state detection, which can improve the accuracy and reliability of the battery health state detection under the airborne condition.

[0009] A lithium ion battery health state lightweight detection method based on physical information neural network, the method comprises the following steps:

[0010] Step S1, feature segment extraction and label definition: analyze the lithium battery data set, extract the current sequence I and voltage sequence U in the data, and introduce the current change rate dI / dt and voltage change rate dV / dt as transient characteristics, then calculate the real-time power P=UI and its change rate dP / dt, integrate the power and the current and voltage change rate, construct an adaptive boundary recognition algorithm, and extract the take-off and landing stage segment data from the data set using the algorithm, then calculate the battery capacity Q of the battery in the battery data using the ampere-hour integral method, and use the battery capacity as the label value of the battery health state, finally repair the missing values in the take-off and landing stage data using the linear interpolation method, and smooth the abnormal data points using the neighbor point mean method, complete the data preprocessing;

[0011] Step S2, calculate the optimal feature sequence: based on the preprocessed take-off and landing segment data, design the feature factors related to battery aging, and verify the correlation of these features with the battery health state by using the Pearson correlation coefficient, then remove the features with high autocorrelation by using the grey correlation degree algorithm, and obtain the optimal feature sequence;

[0012] Step S3, construct a physical information neural network model PINN: PINN is composed of two series neural networks, the first neural network is a solution function network, which captures the local features in the battery aging process, and is used to learn the data loss and consistency loss of the model, the second neural network is a dynamics network, which is used to learn the physical constraint loss of the partial differential equation, and realizes the time sequence evolution rule of simulating the battery aging characteristics, then the optimal feature sequence data is divided into training set and test set, and the training set data is input into the model to train the parameter set of the two neural networks;

[0013] Step S4, the hyperparameter learning rate and network layer number of the PINN are optimized by using a Bayesian algorithm based on TPE;

[0014] Step S5, the PINN is fine-tuned by using a hierarchical transfer learning strategy, and the fine-tuned PINN is used to detect the health state of the battery data.

[0015] Further, the adaptive boundary recognition algorithm comprises:

[0016] Step S11, loading a battery cycle data set , setting a take-off segment list and a landing segment list and initializing to 0, wherein, denotes the th sampling point, k ∈ [ 1 , 2 , ⋯ , N ] , denotes the sampling time of the th sampling point, is the total number of sampling points, is the time collected voltage, is the time collected current, is the time collected temperature, is the time collected state of charge;

[0017] Step S12, filtering the voltage and current data by using a sliding window method;

[0018] Step S13, calculating the power P=UI, the power change rate dP / dt, the voltage change rate dU / dt, and the current change rate dI / dt according to the filtered voltage U and current I;

[0019] Step S14, if |dP / dt|> , and dU / dt , dI / dt , the counter_high value is increased by 1, is the entering high-rate threshold, is the strong transient value, and counter_high is a counter representing the entering high-rate;

[0020] Step S15, if counter_high≥N1, it indicates that this is the take-off stage start point, and the start point position is saved, N1 is the entering high-rate confirmation number, which is set when initialized, N1, is the sampling point number at the beginning of take-off;

[0021] Step S16, if |dP / dt| < dP / dt and dU / dt < dU / dt and dI / dt < dI / dt , then counter_low is increased by 1, counter_low is the counter representing the exit of high rate;

[0022] Step S17, if counter_low N2, indicates the end point of take-off, and the end point position is saved N2 is the number of confirmations of exit of high rate, which is set at initialization, N1 is the sampling point number at the end of take-off;

[0023] Step S18, the discrete segment data from the beginning time to the end time in the battery cycle data is extracted, and saved to the take-off segment list S takeoff = { ( t k , U ( t k ), I ( t k ), T ( t k ), SOC ( t k )) ， | t k ∈ [ t R 1 , t R 2 ] } , and counter_high=0, counter_low=0 are reset;

[0024] Step S19, the landing stage repeats steps S12 to S17, when counter_high≥N1 appears again, it indicates the beginning point of landing, and the beginning point position is saved , when counter_low N2 appears again, it indicates the end point of landing, and the end point position is saved , the discrete segment data from the beginning time to the end time in the battery cycle data is extracted, and saved to the landing segment list S landing = { ( t k , U ( t k ), I ( t k ), T ( t k ), SOC ( t k )) ， | t k ∈ [ t R 3 , t R 4 ] } ;

[0025] Step S20, on the battery cycle data set, the above steps S12 to S19 are repeated until all take-off and landing segment data are extracted, and the program ends.

[0026] Further, the battery capacity value of each cycle is calculated by using the ampere-hour integration method , and the battery capacity sequence Q = [ Q 1 , Q 2 , ⋯ Q n ] , is the nth cycle, is the total number of cycles, and the calculation method is:

[0027] ​ ;

[0028] in, For the first The battery capacity of the next cycle, Δt is the sampling interval, I(t) is the current data, t1 and t2 are the start and end times of the discharge process in this cycle, respectively, and SOC. t1 This is the initial state of charge (SOC) for this discharge cycle. t2 This terminates the SOC (State of Charge) for this discharge cycle.

[0029] Further, step S2 specifically includes:

[0030] Step S21: Based on the preprocessed takeoff and landing segment data, statistical features are designed for voltage, current, and temperature. These features include the maximum, minimum, and average voltage values, as well as the voltage difference at the start and end times. After incorporating all statistical features into the candidate feature pool, the Pearson correlation coefficient between each candidate feature and the battery capacity Q is calculated. Candidate features with Pearson correlation coefficients higher than a preset threshold are retained and sorted in descending order by Pearson correlation coefficient;

[0031] Step S22: For the feature sequence after descending order, take the first ranked feature as the reference feature, and use the grey relational algorithm to calculate the similarity between the remaining features and the reference feature in turn. Delete redundant features with similarity higher than a set threshold, and finally obtain the optimal feature sequence X, which is used as the input for the next step.

[0032] Furthermore, the Pearson correlation coefficient The calculation method is as follows:

[0033] ;

[0034] in, Features In the The value taken in the next loop. Features of The average value of each cycle. This represents the average value of the battery capacity sequence. For the first Battery capacity per cycle, This is for the cumulative summation operation;

[0035] The gray relational degree is calculated as follows:

[0036] ;

[0037] ;

[0038] in, Let the gray relational degree be the m-th feature. For the m-th feature in the th... The correlation coefficient of the next cycle, The total number of cycles, For the reference feature sequence, the first The statistical characteristic value of the next cycle, For the m-th feature The statistical characteristic value of the next cycle, The resolution coefficient is defined by min and max, which represent the minimum and maximum values, respectively.

[0039] Furthermore, step S3 specifically includes:

[0040] Constructing the optimal features of the battery through a solution function network The mapping relationship to battery capacity is set as follows. Then we have: ,in, In order to be in The predicted battery capacity is as follows. For time, The input feature vector sequence is the optimal feature sequence. For the parameter set of the solution function network;

[0041] Parameter set This includes the weight parameters and bias terms of each layer of the solution function neural network, where the first... The learning process for the layer's weight parameters and bias terms is represented as follows:

[0042] h j F = { σ ( W j F ⋅ [ X , t ] + b j F ) , j = 1 σ ( W j F ⋅ [ h j − 1 F , t ] + b j F ) , j > 1 ;

[0043] in, To solve the function network F(·) of the th The output of the layer, Let t be the input feature vector sequence, and t be time. For the first Layer weight parameters, For the first The layer's bias term, where σ is the activation function;

[0044] Then, the parameters of the learned solution function network are used to apply the input. Make a prediction to obtain the predicted battery capacity. This predicted value is then used as one of the inputs to the dynamics network to continue learning the network's parameter set. Parameter set Including weight parameters and bias terms, the learning dynamics network is in its first stage. The learning process for the layer's weight parameters and bias terms is as follows:

[0045] h j G = { σ ( W j G ⋅ [ t , X , u , ∂ u ∂ t , ∂ u ∂ X ] + b j G ) , j = 1 σ ( W j G ⋅ h j − 1 G + b j G ) , j > 1 ;

[0046] wherein, is the output of the layer of the dynamics network, t is the time, is the input feature vector sequence, is the predicted battery capacity by the solution function network, ∂u / ∂t is the rate of change of the predicted battery capacity, ∂u / ∂X is the partial derivative of the predicted battery capacity with respect to the feature vector, is the weight parameter of the layer of the dynamics network, is the bias term of the layer of the dynamics network, and σ is the activation function;

[0047] The physical constraint loss of the two networks is realized by the residual function :

[0048] ;

[0049] wherein, F(t, X; Φ) is the output of the solution function neural network, is the output of the dynamics neural network, and are the parameter sets of the two networks, respectively;

[0050] The total loss function of the PINN network is composed of the data loss function , the consistency loss function , and the physical constraint loss function , and is calculated as follows:

[0051] ;

[0052] wherein, is the total number of battery cycles, is the battery capacity of the th cycle, is the predicted battery capacity of the th cycle, is the predicted battery capacity of the th cycle, is the physical constraint loss of the th cycle, ReLU is the activation function, and α and β are the weighting coefficients, is the total loss function, and when the total loss value reaches the preset minimum value, the model is optimal.

[0053] Further, step S4 specifically includes: randomly setting multiple initial learning rates and obtaining the predicted battery capacity corresponding to each learning rate; comparing with the actual battery capacity to obtain the one among these multiple predicted battery capacities that is closest to the actual battery capacity, which is denoted as the current optimal observation value; constructing two density functions based on this optimal observation value to model the conditional probability distribution of battery capacity at different learning rates; subsequently, calculating the expected improvement EI based on the conditional probability distribution, and obtaining the optimal learning rate by maximizing EI; learning another hyperparameter network layer in the same way to obtain the optimal model.

[0054] Further, step S5 specifically includes: keeping the parameters of the solution function network unchanged, and then fine-tuning the parameters in the dynamic network layer by layer on the target domain battery data, setting the parameters in the dynamic network at the first... The parameters of the layer are The loss function of this layer is expressed as follows:

[0055] ;

[0056] in, For the first The loss function of the layer, Let MSE be the total number of samples, and MSE be the mean squared error loss function. For dynamic networks to input features and the Layer parameters The predicted output, For the first The battery capacity in the next cycle is determined by the loss function. Layer-by-layer fine-tuning optimizes the parameters of each layer in the dynamics network, resulting in the optimal dynamics network. Finally, a physical information neural network model constructed from two optimized neural networks is used to detect the battery health status.

[0057] This invention effectively solves the limitations of existing health estimation methods, such as difficulty in applying them to high-rate discharge scenarios, large data requirements, poor interpretability, and low estimation accuracy. This invention has the following beneficial technical effects:

[0058] Intelligent high-rate working condition feature extraction and screening: In view of the high-rate and strong transient working condition characteristics of the low-altitude aircraft during the take-off and landing stage, the application proposes an adaptive boundary recognition algorithm based on the power density change rate. By monitoring the power density P = UI in real time, when |dP / dt| exceeds the preset threshold τ, the segment boundary is automatically marked, and the high-rate discharge characteristic interval is accurately captured. Meanwhile, the current change rate dI / dt and the voltage change rate dV / dt are introduced as transient features to construct a preliminary feature factor set that can map the battery aging mechanism. Linear interpolation is used to optimize the time resolution, and the Pearson correlation coefficient and grey correlation degree algorithm are used to calculate the high-quality feature set from the preliminary features, which is highly correlated with the battery capacity and has low self-correlation between features. This method effectively solves the problem of insufficient feature extraction in traditional methods under non-stationary and multi-scale coupled working conditions, avoids subjectivity in data acquisition, greatly reduces data requirements, and balances high estimation accuracy and low data acquisition difficulty.

[0059] Physical information constrained neural network modeling: To solve the problems of lack of physical interpretability and insufficient generalization ability in traditional data-driven methods, the application introduces a physical information neural network (PINN) architecture. The physical information neural network consists of two serial neural networks. The first neural network captures local features in the battery aging process from the optimal feature sequence. The second neural network is used to learn the time evolution law of battery aging features from the local features in the battery aging process and the optimal feature sequence data, and to construct the constraint relationship between the two based on the physical mechanism of battery aging. The loss function combines data fitting error and physical consistency residual to ensure that the model learning result conforms to the physical law of battery aging. A composite loss function is constructed, and the Adam optimizer and cosine annealing learning rate scheduling strategy are used for model training.

[0060] Hyperparameter optimization: The Bayesian algorithm based on TPE is used to optimize the learning rate and network layers of PINN. TPE divides the historical evaluation results into two categories by establishing a non-parametric probability model, corresponding to different density functions. This modeling method avoids the limitations of Gaussian process assumptions, making TPE more adaptable to noisy, non-smooth, and non-convex objective functions, and less susceptible to complex function patterns. In addition, the next hyperparameter point selected by TPE is determined by maximizing the expected improvement (EI), which enables it to intelligently jump to a potentially better area. Therefore, under the same computational budget, TPE can significantly reduce the number of trials and quickly converge. Compared with traditional neural network hyperparameter optimization methods, TPE has higher efficiency, faster convergence speed, and stronger search ability, making it particularly suitable for complex hyperparameter spaces.

[0061] The layered transfer learning strategy is used for fine-tuning the trained model: the layered transfer learning strategy is used for fine-tuning the pre-trained model in the physical information neural network (PINN), which can effectively improve the convergence efficiency and prediction accuracy of the model in the new task. By gradually fine-tuning the network parameters layer by layer, the progressive adaptation from low-level features to high-level structures is realized. The layered transfer learning can avoid the knowledge forgetting or unstable training caused by one-time global fine-tuning on the basis of fully preserving the universal physical features learned in the original task. The low-level network is usually responsible for learning the universal physical laws, partial differential equation structural features and basic function representation ability. These parts are often cross-scene universal. The high-level parameters are more focused on the boundary conditions, parameter shifts or field distribution patterns of specific problems. Therefore, layer-by-layer fine-tuning can make the model more stably adapt to new physical scenes. Compared with training PINN from scratch or directly fine-tuning the overall parameters, the layered transfer strategy can significantly reduce the training cost, improve the optimization efficiency, and perform more stably in small sample or complex physical constraint scenarios. It is an effective method to improve the generalization ability and transferability of PINN.

[0062] In summary, the present application extracts the segment data of the eVTOL in the take-off and landing stage based on the unique working mode of the electric vertical take-off and landing aircraft, and filters out a feature set strongly related to battery aging for modeling, improving the relevance and accuracy of the model. At the same time, the model framework designed by the present application has less feature calculation, less storage space occupation, high model detection precision and simple network structure. These characteristics enable the method proposed by the present application to be quickly deployed on the aircraft for real-time SOH detection, realizing the accuracy and reliability of battery health state detection under airborne conditions. BRIEF DESCRIPTION OF DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creating any inventive labor.

[0064] Figure 1 is a flowchart of a lithium ion battery health state lightweight detection method based on a physical information neural network provided by an embodiment of the present application;

[0065] Figure 2 is a schematic diagram of different tasks in one cycle of an eVTOL battery provided by an embodiment of the present application;

[0066] Figure 3 is a flowchart of feature segment extraction in the take-off and landing stages of an eVTOL battery provided by an embodiment of the present application;

[0067] Figure 4 is a feature screening process schematic diagram provided by an embodiment of the present application;

[0068] Figure 5 is a PINN model structure and training strategy schematic diagram provided by an embodiment of the present application;

[0069] Figure 6 is a SOH estimation result schematic diagram of the PINN model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0071] The present application provides a lithium-ion battery health state lightweight detection method based on a physical information neural network. The method is aimed at the specific working mode of the battery of an electric vertical take-off and landing aircraft. By analyzing the operation data, the characteristic factors highly related to battery aging are extracted, a neural network mapping model from the characteristics to the battery capacity is constructed, and finally the learned model is used to realize the rapid detection of the battery health state. The overall process is as shown in takeoff The method comprises the following steps.

[0072] Step S1: feature segment extraction and label definition

[0073] Step S11: analyze the lithium battery data set, extract the current sequence I and voltage sequence U in the data, and introduce the current change rate dI / dt and voltage change rate dV / dt as transient characteristics. Then, the real-time power P=UI and its change rate dP / dt are calculated. The adaptive boundary recognition algorithm is constructed by comprehensively considering the above power and current, voltage change rate, and the algorithm is used to extract the segment data of the take-off and landing stages from the data, wherein the lithium battery data set is obtained from the battery management system.

[0074] In order to accurately extract the characteristics representing battery aging, the battery operation data needs to be analyzed first, and then the characteristics are designed accordingly. SOC The changes of voltage, current and temperature with time of the battery of the electric vertical take-off and landing aircraft in a complete cycle are shown. As can be seen from the figure, during the take-off and landing stages, the voltage, current and temperature all fluctuate significantly, reflecting that the internal mechanism of the battery changes significantly during this process. Therefore, the feature data of these two stages has stronger representation ability for the battery aging state. In comparison, the data of other stages changes smoothly and cannot effectively reflect the battery aging information.

[0075] To this end, the present application firstly extracts the segment data of take-off and landing stage from the complete cycle data of aircraft battery, as shown in landing The method is realized by constructing an adaptive boundary recognition algorithm based on the change rates of voltage, current and power, and the main steps are as follows:

[0076] Step 1, load the battery cycle data set , set the take-off segment list and the landing segment list and initialize them to 0; , represents the th sampling point, k ∈ [ 1 , 2 , ⋯ , N ] , , represents the sampling time of the th sampling point, is the total number of sampling points, are the voltage, current, temperature and state of charge collected at time, respectively.

[0077] Step 2, filter the voltage and current data by using the sliding window method, and the window width is set to 5 sampling points;

[0078] Step 3, according to the filtered voltage U and current I, calculate the power P=UI, the power change rate dP / dt, the voltage change rate dU / dt, and the current change rate dI / dt;

[0079] Step 4, if |dP / dt|>τ_high, and dU / dt , dI / dt , then counter_high value is increased by 1, τ_high is the entering high-rate threshold, is the strong transient value, counter_high is the counter representing entering high-rate, and τ_high and values are set according to experience;

[0080] Step 5, if counter_high≥N1, it indicates that this is the starting point of take-off stage, and the starting point position is saved, N1 is the confirmation number of entering high-rate, which is set when initialized; is the sampling point number at the start of take-off.

[0081] Step 6, if |dP / dt|<τ_low and dU / dt and dI / dt counter_low is incremented by 1, and τ_low is the exit high-rate threshold, and counter_low is the counter representing the exit high-rate;

[0082] Step 7, if counter_low N2, indicating that this is the end point of the take-off phase, and the end point position is saved N2 is the exit high-rate confirmation number, which is set at initialization; is the number of sampling points at the end of take-off

[0083] Step 8, extract the discrete segment data from the start time to the end time in the cycle data and save it to the take-off segment list S SOC = { ( t k , U ( t k ), I ( t k ), T ( t k ), Figure 2 ( t k )) ， | t k ∈ [ t R 1 , t R 2 ] } , reset counter_high = 0, counter_low = 0.

[0084] Step 9, the battery operation mode in the landing phase is still characterized by strong transient high power, so steps 2 to 7 are repeated, and when counter_high ≥ N1 appears again, it indicates that this is the start point of the landing phase, and the start point position is saved When counter_low N2 appears again, it indicates that this is the end point of the landing phase, and the end point position is saved , extract the discrete segment data from the start time to the end time in the cycle data and save it to the landing segment list S Figure 4 = { ( t k , U ( t k ), I ( t k ), T ( t k ), Figure 5 ( t k )) ， | t k ∈ [ t R 3 , t R 4 ] } .

[0085] Step 10, repeat steps 2 to 9 above on the cycle data set until all take-off and landing segment data are extracted, and the program ends.

[0086] Step S12: Calculate the battery capacity Q of the battery in the battery data using the ampere-hour integration method, and use it as the label value of the battery state of health.

[0087] From the battery data set, a sample is constructed for each cycle data, and the battery capacity of the cycle is used as the true value label of the battery state of health (SOH). Then, all "sample-label" pairs are divided into a training set and a test set, and one flight task corresponds to one cycle of the battery.

[0088] In the field of batteries, a generally accepted criterion is that when the capacity of a lithium-ion battery decays to 80% of the initial capacity, it is considered to have reached the end of its service life. Therefore, the detection of the state of health (SOH) of a lithium-ion battery is focused on accurately estimating its current capacity. To achieve this goal, a precise mapping relationship needs to be established from the characteristics that characterize battery aging to the battery capacity. This process mainly includes the following three steps: First, obtain the battery capacity of each cycle in the battery historical data and use it as the label of the training data; second, design features that are strongly correlated with battery aging; and finally, establish a mapping model between the features and the capacity.

[0089] In model training, the battery capacity as the label value of the sample is the benchmark for evaluating the prediction accuracy of the model: the closer the predicted value of the model to the label value, the better the performance of the model. Therefore, accurate acquisition of battery capacity data has a decisive influence on the training effect of the model.

[0090] In the battery management system (BMS), the data acquisition module can obtain real-time data such as time, voltage, current, temperature, and state of charge (SOC) during the operation of the battery. To accurately construct the sample label required for model training, an improved ampere-hour integral method is used to calculate the battery capacity value of each cycle , and the battery capacity sequence Q = [ Q 1 , Q 2 , ⋯ Q n ] , for the th cycle, is the total number of cycles, and the specific calculation method is as follows:

[0091] ;

[0092] wherein, is the battery capacity of the th cycle, Δt is the sampling interval time, I(t) is the current data, t1 and t2 are the start and end times of the discharge process in the cycle, respectively, SOC t1 is the initial SOC of the cycle, and SOC t2 is the final SOC of the cycle.

[0093] Step S13: Linear interpolation is used to repair missing values in the data of the rising and falling stages, and the mean method is used to smooth abnormal data points to complete data preprocessing.

[0094] Step S2: Calculate the optimal feature sequence. Based on the extracted takeoff and landing data segments, firstly, design feature factors related to battery aging, and use the Pearson correlation coefficient to verify the correlation between these features and battery health status. Then, use the grey relational analysis algorithm to remove features with high autocorrelation, including:

[0095] Step S21: Based on the preprocessed takeoff and landing segment data, design statistical features for voltage, current, and temperature. The statistical features include the maximum, minimum, and average voltage values, as well as the voltage difference at the start and end times. After including all statistical features in the candidate feature pool, calculate the Pearson correlation coefficient between each candidate feature and the battery capacity Q. Candidate features with Pearson correlation coefficients higher than a preset threshold are retained and sorted in descending order of Pearson correlation coefficient.

[0096] Step S22: For the feature sequence after descending order, take the first ranked feature as the reference feature, and use the grey relational algorithm to calculate the similarity between the remaining features and the reference feature in turn. Delete redundant features with similarity higher than a set threshold, and finally obtain the optimal feature sequence X, which is used as the input for the next step of the model.

[0097] Based on σ Based on the analysis of voltage, current, and temperature variation characteristics during the take-off and landing phases, and combined with industry experience, this embodiment designed 28 candidate features related to voltage, current, temperature, internal resistance, and capacity, as shown in Table 1. These features can reflect the battery aging state to varying degrees, therefore, it is necessary to select the key features with the greatest characterizing ability.

[0098] Table 1. Candidate Battery Features for Takeoff and Landing Phases of Vertical Electric Aircraft

[0099]

[0100] The specific process of optimal feature selection is as follows: σ As shown:

[0101] First, on the takeoff or landing segment data, calculate each candidate feature and the battery capacity sequence respectively. Q = [ Q 1 , Q 2 , ⋯ Q n ] The Pearson correlation coefficient, with a certain feature set as , K = [ k 1 , k 2 , ⋯ k n ] , Features In the The value taken in the next loop;

[0102] Next, calculate and Pearson correlation coefficient The calculation method is as follows:

[0103] ;

[0104] in, Features of The average value of each cycle. It is the average value of the battery capacity sequence. For the first Battery capacity per cycle, This is for the cumulative summation operation.

[0105] After calculation, candidate features with Pearson correlation coefficients higher than a preset threshold are retained and sorted in descending order of correlation coefficient. In this embodiment, the preset threshold is set to 0.6, resulting in Pearson correlation coefficients for 16 candidate features. If the value is greater than 0.6, save these features in the optimal feature sequence. X = [ f 1 , f 2 , ⋯ f 16 ] In the middle, then... f 1 = [ f 1 ( 1 ), f 1 ( 2 ), ⋯ , f 1 ( n )] As a reference feature sequence, calculate middle f m = [ f m ( 1 ), f m ( 2 ), ⋯ , f m ( n )] and Grey relational degree , m = [ 2 , 3 , ⋯ 16 ] The gray relational degree is calculated as follows:

[0106] ;

[0107] ;

[0108] in, Let the gray relational degree be the m-th feature. For the m-th feature in the th... The correlation coefficient of the next cycle, The total number of cycles, For the reference feature sequence, the first The statistical characteristic value of the next cycle, For the m-th feature The statistical characteristic value of the next cycle, The resolution coefficient is set to 0.5 in this embodiment, and min and max are used to find the minimum and maximum values, respectively.

[0109] Then, redundant features with a gray correlation degree higher than a set threshold are deleted. In this embodiment, the threshold is set to 0.9, meaning that if the gray correlation degree of a certain feature is higher than a set threshold, then the redundant features are deleted. If the similarity is greater than 0.9, it is considered highly similar to the reference feature, and the feature is removed. Ultimately, eight optimal features are retained, forming the optimal feature sequence, denoted as... X = [ f 1 , f 2 , ⋯ f 8 ] , as the input of the next model.

[0110] Step S3, constructing a physical information neural network model

[0111] The physical information neural network model (hereinafter referred to as the model) in the present application is composed of two neural networks in series. The first neural network is a solution function network, which is used to learn the data loss and consistency loss of the model by capturing the local features in the battery aging process. The second neural network is a dynamics network, which is used to learn the physical constraint loss of the partial differential equation, so as to realize the time sequence evolution law of simulating the battery aging characteristics. The total loss of the model is composed of the data loss, the consistency loss and the physical constraint loss. When the total loss value of the model reaches the minimum, the model reaches the optimum, and the minimum value is pre-set according to experience;

[0112] Then the optimal feature sequence data is divided into a training set and a test set, and the training set data is input into the model to train the parameter set of the two neural networks.

[0113] In the present embodiment, the optimal feature sequence X = [ f 1 , f 2 , ⋯ f 8 ] is divided into a training set and a test set in a ratio of 7:3, and the training set data is input into the model for training, and the learning rate and the network layer layer are set as the hyperparameters of the model, and the initial learning rate =0.001, the initial layer=3, and the Adam optimizer and the cosine annealing strategy are used for model training.

[0114] The structure of the physical information neural network model is shown in σ , which is described in detail as follows:

[0115] The physical information neural network model includes two core components: a solution function network F(·) and a dynamics network G(·). The solution function network is responsible for establishing a nonlinear mapping relationship from the input feature to the battery capacity, and its structure includes an input layer, a fully connected hidden layer and an output layer. The structure of the dynamics network is consistent with that of the solution function network, except that the weight parameters are different. The dynamics network is used to simulate the dynamics behavior in the battery degradation process to capture the change law of the battery capacity over time. The two networks are connected through the physical constraint loss to form a unified physical information neural network model.

[0116] Firstly, the solution function network F(·) is a neural network used to simulate a nonlinear function, so as to construct the mapping relationship from the battery optimal feature to the battery capacity. Assuming that the mapping relationship is , then the process can be expressed in mathematical expression as:

[0117] ;

[0118] in, In order to be in The predicted battery capacity is as follows. For time, The input feature vector sequence is the optimal feature sequence. This is the set of parameters for the solution function network.

[0119] Obtain the mapping relationship The process of training a neural network is the same as the process of learning a set of parameters. The weight parameters and bias terms for each layer.

[0120] For the solution function network F(·), its first... The learning process for the layer's weight parameters and bias terms is represented as follows:

[0121] h j F = { σ ( W j F ⋅ [ X , t ] + b j F ) , j = 1 Figure 6 ( W j F ⋅ [ h j − 1 F , t ] + b j F ) , j > 1 ;

[0122] in, To solve the function network F(·) of the th The output of the layer, Let t be the input feature vector sequence, and t be time. For the first Layer weight parameters, For the first The layer's bias term, σ, is the activation function.

[0123] Once all the weight parameters and bias terms in the solution function network F(·) have been learned, the solution function network F(·) becomes a definite mapping relationship, which can be used to map the input... Make a prediction to obtain the predicted battery capacity. This predicted value is then used as one of the inputs to the dynamic network G(·), which continues to learn the parameter set of the dynamic network G(·). Parameter set Including weight parameters and bias terms, since the structure of the dynamic network G(·) is the same as that of the solution function network F(·), its first... The learning process for the layer's weight parameters and bias terms is represented as follows:

[0124] h j G = { ​ ( W j G ⋅ [ t , X , u , ∂ u ∂ t , ∂ u ∂ X ] + b j G ) , j = 1 ​ ( W j G ⋅ h j − 1 G + b j G ) , j > 1 ;

[0125] in, For the dynamic network G(·), the th The layer's output, where t is time. Given the input feature vector sequence, For the battery capacity predicted by F(·), ∂u / ∂t is the rate of change of the predicted battery capacity, and ∂u / ∂X is the partial derivative of the predicted battery capacity with respect to the feature vector, is the weight parameter of the i-th layer in the dynamics network G(·), is the bias term of the i-th layer in the dynamics network G(·), and σ is the activation function. is the weight parameter of the i-th layer in the dynamics network G(·), The dynamics network G(·) captures the dynamic characteristics and intrinsic laws of the battery degradation process by integrating the time, features, current battery capacity state, and their partial derivative information.

[0126] The physical constraint loss of the two networks is realized by the residual function

[0127]

[0128] where F(t, X; Φ) is the output of the solution function neural network F(·), is the output of the dynamics neural network G(·), and are the parameter sets of the two networks, respectively.

[0129] Finally, the PINN network adopts a polynomial compound loss function to optimize the prediction accuracy of the model, which is composed of a data loss function , a consistency loss function , and a physical constraint loss function , and the calculation method is as follows:

[0130]

[0131] where is the total number of battery cycles, is the battery capacity of the i-th cycle, is the predicted battery capacity of the i-th cycle, is the predicted battery capacity of the i-th cycle, is the physical constraint loss of the i-th cycle, ReLU is the activation function, and α and β are the weighting coefficients. is the total loss function, and when the total loss value reaches a minimum value, the model is optimal.

[0132] ​​​​​​​Step S4, the hyperparameters learning rate and network layer of PINN are optimized by using the Bayesian algorithm based on TPE: a plurality of initial learning rates are randomly set, and the corresponding predicted battery capacity under each learning rate is obtained, by comparing with the real battery capacity, the one closest to the real battery capacity in the plurality of predicted battery capacities is obtained, which is recorded as the current optimal observation value, two density functions are constructed based on the optimal observation value to model the conditional probability distribution of the battery capacity under different learning rates, then, the expected improvement EI is calculated based on the conditional probability distribution, and the optimal learning rate is obtained by maximizing the EI, in the same way, the network layer is learned as another hyperparameter, the optimal model is obtained, and the battery capacity is estimated on the test set to detect the battery state of health.

[0133] In step S3, the PINN network model mainly learns the weight parameters and bias terms in the training process, and the learning rate and the network layer layer as the hyperparameters of the model are still crucial to the model, and the model itself cannot learn them, therefore, the Bayesian optimization algorithm based on the tree structure Parzen estimator (Tree based Parzen estimator, TPE) is used to search for the optimal hyperparameters.

[0134] The general description of the TPE algorithm is as follows:

[0135] The TPE algorithm is a parameter optimization method based on a probability model, assuming that the parameter to be optimized is x, and the target function is y=f(x), the goal of the algorithm is to find the optimal parameter x* so that the target function value y is optimal. Since the probability distribution of the target function y cannot be directly obtained, the TPE algorithm uses two density functions to model the conditional probability distribution of the target function, the specific implementation process is as follows: first, a plurality of points are randomly sampled in the value space of the parameter x, and these sampling points are substituted into the target function y=f(x), by observation and comparison, the optimal value of the random sampling points can be determined, and is recorded as the current optimal observation value y*, then two density functions are constructed based on the optimal observation value as follows:

[0136] Assuming that is the parameter configuration set of all , there are points, is the parameter configuration set of all , there are points, then:

[0137] ;

[0138] wherein, is the parameter to be optimized, is the target function value, is the current optimal observation, is the conditional probability density function, and are two different density functions, is the parameter dimension, is a balance parameter, is the Euclidean distance.

[0139] The optimization goal of the TPE algorithm is to find the optimal parameter such that is as close as possible to , for which the expected improvement is used to find the optimal parameter , when the expected improvement reaches the maximum, the corresponding is the optimal parameter, and the expression of the expected improvement

[0140] .

[0141] is the expected improvement of under the condition , the integral is from negative infinity to , and and are conditional probability density functions, , is the probability density function, where is the weighting coefficient, and the expected improvement function is simplified as follows:

[0142] . The final simplified expression is:

[0143]

[0144] . where

[0145] is proportional to .

[0146] According to the general formula of the TPE algorithm, the parameter is replaced by the learning rate , then two density functions of the learning rate are constructed respectively, and the maximum expected improvement is calculated according to the two density functions, so as to obtain the optimal learning rate , and in the same way, the optimal hyperparameter network layer layer is learned.

[0147] Step S5, the physical information neural network model is improved by adopting a hierarchical transfer learning strategy: the network model optimized by the above TPE-based Bayesian algorithm is fine-tuned by adopting the hierarchical transfer learning strategy, so that the optimal dynamic network is obtained, and finally the battery health state detection is performed by the physical information neural network model constructed by the two optimized neural networks.

[0148] For battery data of different battery types and working conditions, there is a source domain (one working condition battery data) and target domain (another battery data) inconsistency problem, therefore, the pre-trained PINN model is fine-tuned by adopting the hierarchical transfer learning strategy, and the main implementation process is freezing the parameters of the solution function neural network F(·), that is, keeping the parameters in the solution function neural network F(·) unchanged, then layer-by-layer fine-tuning the parameters in the dynamic network G(·) on the target domain battery data, and using the learned new features for the final target domain battery capacity prediction.

[0149] Suppose the parameters of the first layer in the dynamic network G(·) are , then the loss function of this layer is represented by the following formula:

[0150] ;

[0151] Wherein, is the loss function of the first layer, N is the total number of samples, MSE is the mean square error loss function, is the predicted output of the dynamic network model to the input features and the first layer parameters , is the battery capacity of the first cycle.

[0152] The loss function is layer-by-layer fine-tuned so that the parameters of each layer in the dynamic network G(·) reach the optimal value, thereby obtaining the optimal dynamic network, and finally the battery health state detection is performed by the physical information neural network model constructed by the two optimized neural networks.

[0153] Finally, the method disclosed in the application is verified on the public battery data set of eVTOL aircraft, and the battery health state estimation result is as follows ​ ​As shown in the figure, the X axis is the true SOH value (discharge capacity of the battery) of the prediction sample, the Y axis is the prediction value of the prediction sample by the method, the blue curve represents that the prediction value is consistent with the true value, the red scattered points are generally distributed along the blue line, which shows that the prediction value by the method is highly consistent with the true value, and the individual red scattered points deviate from the blue solid line, which shows that there is a certain error between the individual prediction value and the true value, which is caused by the poor quality of individual data in the prediction sample, but the overall prediction trend is highly consistent with the true value, which shows that the method has high accuracy and stability in predicting the battery SOH.

[0154] The application accurately captures the key feature interval of the battery under high rate and strong transient working conditions in the take-off and landing stages by constructing an adaptive boundary recognition algorithm based on the power change rate, and introduces a physical information neural network (PINN) to jointly fit the data error, physical loss function error and consistency error in the loss function, overcomes the defects of lack of physical mechanism mapping and insufficient interpretability of traditional data-driven methods, ensures that the estimation result conforms to the physical law of battery aging, and significantly improves the generalization ability of the model; the hyperparameters in the PINN network model are quickly optimized by using the Bayesian optimization algorithm based on the TPE structure, realizing the collaborative optimization of estimation accuracy, calculation efficiency and physical consistency, so that the model can complete high-precision battery capacity estimation only by using the key short-time discharge data in the take-off and landing stages, thereby realizing rapid detection of the battery health state, and the model in the application uses two serially connected neural networks for feature learning, so that the model has fewer parameters and has a lightweight design, meeting the stringent requirements of onboard real-time applications. At the same time, the feature sequence calculation is relatively simple, which greatly reduces the data demand and the occupation of calculation resources compared with traditional methods; the transfer learning strategy based on layer-by-layer fine-tuning is adopted, which effectively solves the inconsistency problem between the target domain data and the source domain data, and improves the model's cross-condition robustness. The method proposed in the application not only efficiently solves the problem of aircraft battery health detection, but also provides a scientific basis for optimizing the battery health management strategy, and comprehensively improves the intelligent level and safety protection ability of the low-altitude aircraft battery management system.

[0155] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks, characterized in that, The method includes: Step S1, Feature Segment Extraction and Label Definition: Analyze the lithium battery dataset, extract the current sequence I and voltage sequence U from the data, and introduce the current change rate dI / dt and voltage change rate dV / dt as transient features. Then calculate the real-time power P=UI and its change rate dP / dt. Combining power with current and voltage change rates, construct an adaptive boundary recognition algorithm, and use this algorithm to extract the segment data of the takeoff and landing phases from the dataset. Then use the ampere-hour integration method to calculate the battery capacity Q in the battery data, and use the battery capacity as the label value of the battery health status. Finally, use the linear interpolation method to repair the missing values ​​in the takeoff and landing phase data, and use the neighbor mean method to smooth the abnormal data points to complete the data preprocessing. Step S2, calculate the optimal feature sequence: Based on the preprocessed takeoff and landing segment data, design feature factors related to battery aging, and use Pearson correlation coefficient to verify the correlation between these features and battery health status. Then, use the grey relational algorithm to remove features with high autocorrelation to obtain the optimal feature sequence. Step S3: Construct the Physical Information Neural Network Model (PINN): PINN consists of two cascaded neural networks. The first neural network is a solution function network, which captures local features during the battery aging process to learn the model's data loss and consistency loss. The second neural network is a dynamics network, which learns the physical constraint loss of partial differential equations to simulate the temporal evolution of battery aging characteristics. Then, the optimal feature sequence data is divided into a training set and a test set, and the training set data is input into the model to train the parameter sets of the two neural networks. Step S4: Use a TPE-based Bayesian algorithm to optimize the hyperparameters learning rate and number of network layers of PINN; Step S5: Fine-tune PINN using a hierarchical transfer learning strategy, and use the fine-tuned PINN to detect the health status of battery data.

2. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 1, characterized in that, The construction of the adaptive boundary recognition algorithm includes: Step S11, Load the battery cycling dataset Set the takeoff clip list and landing clip list And initialized to 0, where, Indicates the first One sampling point, , Indicates the first Sampling time of each sampling point This represents the total number of sampling points. for Voltages collected at all times for The current collected at all times, for Temperatures collected at all times for The state of charge collected at any time; Step S12: The voltage and current data are filtered using the sliding window method; Step S13: Calculate the power P=UI, the power change rate dP / dt, the voltage change rate dU / dt, and the current change rate dI / dt based on the filtered voltage U and current I. Step S14, if |dP / dt|> And dU / dt dI / dt Then the value of counter_high is incremented by 1. To reach the high magnification threshold, It is a strong transient value, and counter_high is a counter that indicates entering a high-multiplication rate; Step S15: If counter_high ≥ N1, it indicates that this is the start point of the takeoff phase, and the starting point position is saved. N1 represents the number of times to confirm entering high-magnification mode, which is set during initialization. The number of sampling points at the start of takeoff; Step S16, if |dP / dt| < And dU / dt < and dI / dt< Then the value of counter_low is incremented by 1. The threshold for exiting high-magnification mode is defined by `counter_low`, which is a counter representing the exit from high-magnification mode. Step S17, if counter_low N2 indicates that this is the end point of the takeoff phase, and the end point position is saved. N2 represents the number of times to confirm exiting high-multiplier mode, which is set during initialization. This represents the number of sampling points at the end of takeoff. Step S18, extract the battery cycle data from the start time. By the end time The discrete fragment data is saved to the takeoff fragment list. And reset counter_high=0, counter_low=0; Step S19: During the descent phase, repeat steps S12 to S17. When counter_high≥N1 occurs again, it indicates that this is the start point of the descent phase, and the start point position is saved. When counter_low occurs again When N2 is reached, it indicates the end point of the descent phase; the end point position is saved. Extract battery cycle data from the start time By the end time The discrete fragment data is saved to the landing fragment list. ; Step S20: Repeat steps S12 to S19 on the battery cycle dataset until all takeoff and landing segment data are extracted, and then end the program.

3. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 1, characterized in that, The battery capacity for each cycle is calculated using the ampere-hour integration method. And obtain the battery capacity sequence. , For the first The next loop. The total number of cycles is calculated as follows: ； in, For the first The battery capacity of the next cycle, Δt is the sampling interval, I(t) is the current data, t1 and t2 are the start and end times of the discharge process in this cycle, respectively, and SOC. t1 This is the initial state of charge (SOC) for this discharge cycle. t2 This terminates the SOC (State of Charge) for this discharge cycle.

4. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 1, characterized in that, Step S2 further includes: Step S21: Based on the preprocessed takeoff and landing segment data, statistical features are designed for voltage, current, and temperature. These features include the maximum, minimum, and average voltage values, as well as the voltage difference at the start and end times. After incorporating all statistical features into the candidate feature pool, the Pearson correlation coefficient between each candidate feature and the battery capacity Q is calculated. Candidate features with Pearson correlation coefficients higher than a preset threshold are retained and sorted in descending order by Pearson correlation coefficient; Step S22: For the feature sequence after descending order, take the first ranked feature as the reference feature, and use the grey relational algorithm to calculate the similarity between the remaining features and the reference feature in turn. Delete redundant features with similarity higher than a set threshold, and finally obtain the optimal feature sequence X, which is used as the input for the next step.

5. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 4, characterized in that, Pearson correlation coefficient The calculation method is as follows: ； in, Features In the The value taken in the next loop. Features of The average value of each cycle. This represents the average value of the battery capacity sequence. For the first Battery capacity per cycle, This is for the cumulative summation operation; The gray relational degree is calculated as follows: ； ； in, Let the gray relational degree be the m-th feature. For the m-th feature in the th... The correlation coefficient of the next cycle, The total number of cycles, For the reference feature sequence, the first The statistical characteristic value of the next cycle, For the m-th feature The statistical characteristic value of the next cycle, The resolution coefficient is defined by min and max, which represent the minimum and maximum values, respectively.

6. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 1, characterized in that, Step S3 further includes: Constructing the optimal features of the battery through a solution function network The mapping relationship to battery capacity is set as follows. Then we have: ,in, In order to be in The predicted battery capacity is as follows. For time, The input feature vector sequence is the optimal feature sequence. For the parameter set of the solution function network; Parameter set This includes the weight parameters and bias terms of each layer of the solution function neural network, where the first... The learning process for the layer's weight parameters and bias terms is represented as follows: ； in, To solve the function network F(·) of the th The output of the layer, Let t be the input feature vector sequence, and t be time. For the first Layer weight parameters, For the first The layer's bias term, where σ is the activation function; Then, the parameters of the learned solution function network are used to apply the input. Make a prediction to obtain the predicted battery capacity. This predicted value is then used as one of the inputs to the dynamics network to continue learning the network's parameter set. Parameter set Including weight parameters and bias terms, the learning dynamics network is in its first stage. The learning process for the layer's weight parameters and bias terms is as follows: ； in, For the dynamic network The layer's output, where t is time. Given the input feature vector sequence, Let be the battery capacity predicted by the solution function network, ∂u / ∂t be the rate of change of the predicted battery capacity, and ∂u / ∂X be the partial derivative of the predicted battery capacity with respect to the eigenvector. For the dynamic network Layer weight parameters, For the dynamic network The layer's bias term, where σ is the activation function; The physical constraint loss of the two networks is expressed through the residual function. accomplish: ； Where F(t, X; Φ) is the output of the solution function neural network. The output of the dynamic neural network, and These are the parameter sets for the two networks, respectively. The total loss function of the PINN network is composed of the data loss function. Consistency loss function and physical constraint loss function The composition and calculation method are as follows: ； in, This represents the total number of battery cycles. For the first Battery capacity per cycle, For the first Predicted battery capacity for the next cycle For the first Predicted battery capacity for the next cycle For the first The physical constraint loss for the next iteration uses ReLU as the activation function and α and β as weighting coefficients. Let this be the total loss function, when the total loss value The model is optimal when the preset minimum value is reached.

7. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 1, characterized in that, Step S4 further includes: randomly setting multiple initial learning rates and obtaining the predicted battery capacity corresponding to each learning rate; comparing the predicted battery capacity with the actual battery capacity and obtaining the one that is closest to the actual battery capacity among these multiple predicted battery capacities, which is denoted as the current optimal observation value; constructing two density functions based on this optimal observation value to model the conditional probability distribution of the battery capacity at different learning rates; then calculating the expected improvement EI based on the conditional probability distribution and obtaining the optimal learning rate by maximizing EI; and learning another hyperparameter network layer in the same way to obtain the optimal model.

8. The lightweight detection method for the health status of lithium-ion batteries based on physical information neural networks according to claim 1, characterized in that, Step S5 further includes: keeping the parameters of the solution function network unchanged, and then fine-tuning the parameters in the dynamic network layer by layer on the target domain battery data, setting the parameters in the dynamic network at the first... The parameters of the layer are The loss function of this layer is expressed as follows: ； in, For the first The loss function of the layer, Let MSE be the total number of samples, and MSE be the mean squared error loss function. For dynamic networks to input features and the Layer parameters The predicted output, For the first The battery capacity in the next cycle is determined by the loss function. Layer-by-layer fine-tuning optimizes the parameters of each layer in the dynamics network, resulting in the optimal dynamics network. Finally, a physical information neural network model constructed from two optimized neural networks is used to detect the battery health status.

Images