Multi-objective charging strategy optimization method considering battery health degradation

Abstract

The application relates to the technical field of charging optimization, in particular to a multi-objective charging strategy optimization method considering battery health degradation. The application proposes the following scheme: firstly, collecting instant state parameters of a battery when the battery is connected to a battery swap cabinet, and mapping the instant state parameters to a preset battery group health model library to obtain an initial health state estimation value; before charging is started, voltage response and impedance characteristics are obtained through disturbance current diagnosis to correct the initial estimation, and a health state posterior estimation value and a corresponding uncertainty index are obtained; on this basis, a target optimization function is constructed by combining charging duration, battery health degradation cost, electricity price cost and thermal management cost to generate an individual charging strategy suitable for the battery; the application can improve health state estimation accuracy in a scenario lacking battery historical data, reduce charging strategy deviation, and balance safety, efficiency and economy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of charging optimization, in particular to a multi-objective charging strategy optimization method considering battery health degradation. BACKGROUND

[0002] In the prior art, the battery swap cabinet as the core equipment of centralized charging can simultaneously charge multiple batteries in parallel. However, due to the significant individual differences of the batteries, the swap cabinet often has difficulty in achieving differentiated control for different battery states when performing charging management, thereby causing health state recognition deviation and further affecting the charging effect. Under this condition, if a unified charging strategy is adopted, it is easy to cause some batteries to overheat and capacity degradation to intensify during the charging process, thereby reducing the cycle life of the batteries. On the other hand, a too conservative charging strategy will significantly prolong the charging time, increase the comprehensive cost of electricity price and thermal management, and reduce the overall operation efficiency and economy. In addition, the existing swap cabinet lacks optimization and coordination at the group level when multiple batteries are charged simultaneously, resulting in insufficient utilization of channel resources, deviation of the overall charging strategy from the optimal, and inability to balance safety, efficiency and cost control.

[0003] To solve the above problems, the present application designs a multi-objective charging strategy optimization method considering battery health degradation. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a multi-objective charging strategy optimization method considering battery health degradation, which first collects the instantaneous state parameters of the battery when it is connected to the swap cabinet, and maps them to a preset battery group health model library to obtain an initial health state estimate. Before charging starts, the voltage response and impedance characteristics are obtained through disturbance current diagnosis to correct the initial health state estimate to obtain a posteriori health state estimate and corresponding uncertainty index. On this basis, a target optimization function is constructed by combining the charging time, battery health degradation cost, electricity price cost and thermal management cost to generate an individual charging strategy suitable for the battery.

[0005] To achieve the above purpose, the present application provides the following technical solutions:

[0006] The multi-objective charging strategy optimization method considering battery health degradation is applied to a battery swap cabinet, the battery swap cabinet comprising a plurality of charging channels, and the method comprising:

[0007] Collecting the instantaneous state parameters of the battery, mapping the battery state to a preset battery group health model library according to the instantaneous state parameters, determining a reference battery model of the battery, and obtaining an initial health state estimate;

[0008] Before starting charging, the battery is subjected to a disturbance current diagnosis to obtain a voltage response and impedance characteristics, and the initial health state estimate is updated according to the voltage response and impedance characteristics to obtain a health state posterior estimate and a corresponding uncertainty index;

[0009] According to the health state posterior estimate and the corresponding uncertainty index, a target optimization function is constructed by combining the charging time, battery health decay cost, electricity price cost and thermal management cost to generate a charging strategy suitable for the battery.

[0010] The instant state parameters include the open-circuit voltage, terminal voltage and temperature of the battery after being connected to the battery swap cabinet, and the battery state is mapped to a preset battery population health model library according to the instant state parameters to obtain an initial health state estimate, including:

[0011] The terminal voltage is subjected to similarity calculation with a reference battery characteristic vector in the battery population health model library, and according to the result of the similarity calculation, a reference battery model of the battery is determined;

[0012] The open-circuit voltage and temperature are matched with a reference open-circuit voltage-temperature curve corresponding to the reference battery model in the battery population health model library to determine an initial state of charge correction value of the battery;

[0013] According to the reference battery model and the initial state of charge correction value, the initial health state estimate is output, wherein the initial health state estimate at least includes one of capacity retention rate and internal resistance growth rate.

[0014] The battery is subjected to a disturbance current diagnosis to obtain a voltage response and impedance characteristics, including:

[0015] In the static stage of the battery being connected to the battery swap cabinet and not starting charging, historical charging data of the reference battery model is called to generate a disturbance current excitation sequence, wherein the disturbance current excitation sequence satisfies a zero net electric quantity constraint;

[0016] The battery is excited by the disturbance current excitation sequence, and the battery terminal voltage is synchronously sampled to obtain a voltage original sequence;

[0017] The voltage original sequence is subjected to orthogonal demultiplexing and matched filtering array projection to obtain a corresponding voltage characteristic coefficient vector;

[0018] The voltage characteristic coefficient vector is subjected to feature extraction to obtain a voltage response, wherein the voltage response includes a voltage transient offset, a recovery slope and a steady-state difference value;

[0019] The impedance parameters in the reference battery model are corrected according to the voltage response, and impedance characteristics are obtained, wherein the impedance characteristics include a dynamic internal resistance, a polarization time constant and a temperature rise sensitivity.

[0020] The historical charging data of the reference battery model is called to generate a perturbation current excitation sequence, including:

[0021] The historical charging data is frequency domain decomposed to extract frequency components to generate a candidate perturbation sequence;

[0022] The candidate perturbation sequence is optimized under the condition of meeting the zero net electric quantity constraint to obtain first and second perturbation currents that are mutually orthogonal;

[0023] The first and second perturbation currents are corrected through amplitude scaling and phase adjustment to generate the perturbation current excitation sequence.

[0024] The voltage original sequence is orthogonally demultiplexed and projected by a matched filter array to obtain corresponding voltage characteristic coefficient vectors, including:

[0025] The voltage original sequence is decomposed in a preset characteristic space to obtain a plurality of mutually independent response components;

[0026] A projection matrix and a projection operator are calculated through historical charging data;

[0027] The response components are mapped into voltage characteristic coefficient vectors in combination with the projection matrix and the projection operator.

[0028] The initial health state estimate value is updated according to the voltage response and the impedance characteristics to obtain a health state posterior estimate value and a corresponding uncertainty index, including:

[0029] The perturbation current excitation sequence is forward simulated under the reference battery model to obtain a predicted voltage response and a predicted impedance parameter, and is matched with the voltage response and the impedance characteristics to calculate a time domain residual and a parameter deviation;

[0030] A likelihood function is constructed according to the time domain residual, the parameter deviation and a preset sampling noise covariance, the initial health state estimate value is taken as a prior, and the likelihood function and the prior are updated posteriorly through Bayesian regression to obtain a health state posterior estimate value and a posterior covariance matrix corresponding to the health state posterior estimate value;

[0031] Two groups of voltage characteristic coefficient vectors obtained through mutually orthogonal perturbation current excitation sequences are subjected to consistency test to calculate a cross-sequence consistency deviation;

[0032] Splice the impedance features into a feature vector, and perform distance measurement with a reference distribution in the battery population health model library to obtain a model domain distance index, the distance measurement being calculated by Mahalanobis distance;

[0033] Weighted fusion of the spectral radius of the posterior covariance matrix, the cross-sequence consistency bias and the model domain distance index to obtain an uncertainty index, wherein each weight is calculated according to the similarity between the battery and the reference battery model.

[0034] The target optimization function is constructed by combining the charging duration, the battery health degradation cost, the electricity price cost and the thermal management cost, including:

[0035] The feature factors corresponding to the charging duration, the battery health degradation cost, the electricity price cost and the thermal management cost are respectively taken as basic target items;

[0036] The basic target item corresponding to the charging duration and the battery health degradation cost is defined as a first function based on the capacity retention rate and the internal resistance growth rate, and the confidence interval boundary of the health state posterior estimate value is taken as the penalty factor of the first function;

[0037] The basic target item corresponding to the electricity price cost is expanded into a second function according to the time sequence, and the penalty factor of the second function is adjusted by the electricity price of different time periods;

[0038] The basic target item corresponding to the thermal management cost is associated with the preset battery temperature prediction trajectory in the reference battery model of the battery to obtain a third function, and the temperature rise allowed range is adjusted by the uncertainty index to obtain the penalty factor of the third function;

[0039] According to the first function, the second function and the third function, and the penalty factors corresponding to each function, the target optimization function is calculated.

[0040] The charging strategy adapted to the battery is generated, including:

[0041] The charging process is divided into multiple stages according to the state of charge interval, and a risk envelope function is constructed for each stage according to the uncertainty index, the risk envelope function being used for parameterized remodeling of the upper limit of the charging rate, the temperature rise range and the proximity penalty of the target state of charge of each stage;

[0042] The confidence boundary of the risk envelope function is taken as the basis to calculate the probabilistic safety constraint, wherein the probabilistic safety constraint represents the probability that the battery charging state falls into the safety domain within the uncertainty index value range;

[0043] In the case that the probabilistic safety constraint is met at each stage, an uncertainty evolution trajectory in the charging process is predicted according to the health state posterior estimation value, and an information gain of each stage to the uncertainty index having an expected descending effect is calculated, wherein the expected descending effect represents an expected contraction amplitude of the uncertainty index after the corresponding charging rate and timing setting is performed;

[0044] The information gain is used as a positive regulation factor to regulate the target optimization function, and a charging mode of each stage is calculated by an optimization algorithm according to the regulated target optimization function, and a charging strategy is obtained by summarizing the charging mode.

[0045] The information gain of each stage to the uncertainty index having an expected descending effect comprises:

[0046] Under the probabilistic safety constraint, simulation evolution is performed on different charging rate and timing combinations to obtain corresponding uncertainty contraction trajectories;

[0047] The entropy value of the uncertainty contraction trajectory is calculated, and the entropy reduction amplitude is used as a quantitative index of the information gain.

[0048] The method further comprises:

[0049] According to the charging strategy and the channel resource constraint of the corresponding charging channel of the battery swap cabinet, group charging scheduling optimization is performed to obtain an overall charging strategy.

[0050] Compared with the prior art, the beneficial effects of the present application are:

[0051] The present application realizes fine estimation of the battery health state under the condition of lacking complete historical data by introducing a battery group health model library combined with disturbance current diagnosis, and further quantifies the reliability by an uncertainty index to provide a verifiable basis for subsequent charging strategy formulation. By simultaneously introducing charging time, health degradation cost, electricity price cost and thermal management cost in the target optimization function, and combining a risk envelope function and a probabilistic safety constraint for dynamic regulation, the balance between charging efficiency and economy can be achieved under the premise of ensuring battery safety and service life. BRIEF DESCRIPTION OF DRAWINGS

[0052] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0053] Figure 1 A schematic diagram of the principle of the actual problem of the battery swap cabinet of the embodiments of the present application;

[0054] Figure 2 A schematic diagram of an exemplary application scenario of the embodiments of the present application;

[0055] Figure 3 Flowchart of the multi-objective charging strategy optimization method considering battery health degradation for the embodiments of the present application;

[0056] Figure 4 Flowchart of the calculation of the initial state of health estimation value for the embodiments of the present application;

[0057] Figure 5 Flowchart of the calculation of the voltage response and impedance characteristics for the embodiments of the present application;

[0058] Figure 6 Flowchart of another multi-objective charging strategy optimization method considering battery health degradation for the embodiments of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments.

[0060] In this document, the term “embodiment” means that a specific feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It will be explicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0061] The present application is applicable to multi-objective charging optimization control in the battery swap cabinet scenario, and its running environment has the characteristics of parallel charging channels, batch battery random access, and missing individual historical data of batteries. The charging optimization process in this kind of application scenario is limited by the external collection conditions of the swap cabinet and the individual differences of the batteries. In the absence of complete life curve support, directly building a battery health model often has estimation bias and strategy uncertainty interference.

[0062] The application scenarios of the present application include but are not limited to:

[0063] Vehicle power battery swap station with large-scale parallel channels;

[0064] The individual state of the user-accessed battery is significantly different, and the health condition cannot be traced back relying on the historical charging and discharging data of the single battery;

[0065] In the peak-valley electricity price environment, the charging scheduling needs to consider both cost constraints and battery life constraints;

[0066] Under the conditions of high-rate charging and limited cooling, battery thermal management becomes a key influencing factor in the design of the charging strategy.

[0067] The selection of application scenarios is determined based on the common features of the battery swap cabinet. The selected application scenarios need to have at least one of the following features:

[0068] Battery historical data is missing or insufficient, and health state estimation relies on real-time parameters and population model inference;

[0069] The charging process of the battery in the battery swap cabinet presents uncertainty propagation, resulting in a deviation interval in health assessment;

[0070] The charging strategy needs to be optimized among factors such as life attenuation, electricity price, and thermal management;

[0071] Individual charging strategies are coupled with group channel resources.

[0072] It should be noted that the multi-objective charging strategy optimization method proposed in the present application does not rely on a fixed life model of a specific battery model, nor does it require the battery to have complete historical charging and discharging data as a prerequisite. Instead, it is aimed at battery swap cabinet scenarios with real-time health diagnosis constraints, uncertainty quantification requirements, and multi-cost coupled optimization conditions.

[0073] It is worth noting that the multi-objective charging strategy optimization method considering battery health attenuation described in the present application is not designed for a specific charging protocol or a single life model, and can also be applied to:

[0074] Scenarios with battery model uncertainty and dynamic health assessment updates;

[0075] Energy price time-varying, cooling condition limited operation scenarios;

[0076] Optimization model space where the charging strategy needs to meet safety, life extension, and economic goals simultaneously.

[0077] Before further introducing the technical problem, the present application explains the principle of the battery swap cabinet in advance. In the actual application of the battery swap cabinet, the user's electric vehicle will remove the on-board battery and place it in the battery swap cabinet for centralized charging when the battery is low. The battery swap cabinet is usually provided with multiple charging channels, each of which can independently access the battery and perform charging management.

[0078] It can be understood that since the battery swap cabinet does not have complete historical charging and discharging data of the battery, the battery itself also does not have historical charging and discharging data, and the estimation of the health state often relies on empirical models or statistical average parameters, resulting in high uncertainty when the individual differences of the battery are large. This uncertainty will directly affect the generation of the charging strategy, leading to two typical problems:

[0079] On the one hand, if a more aggressive strategy is adopted, it may cause overheating or exacerbate capacity degradation when the battery is in a poor real health state; on the other hand, if a more conservative strategy is adopted, it will prolong the charging time or increase the comprehensive cost of electricity price and thermal management. Therefore, in actual application, the battery swap cabinet generally has the technical bottlenecks of inaccurate health state identification and deviation of the charging strategy from the optimal.

[0080] Reference Figure 1 , Figure 1 The actual problem principle diagram of the battery swap cabinet provided by the embodiment of the application.

[0081] As Figure 1 shown, the battery of the user's vehicle is placed in the charging channel of the swap cabinet when the power is insufficient, and the charging controller adopts a conservative charging strategy or an aggressive charging strategy to manage the charging of the battery. Under the existing technical conditions, the swap cabinet can only estimate the state based on limited instantaneous measurement parameters, and it is easy to cause deviation of the charging strategy due to lack of battery historical data.

[0082] It can be understood that Figure 1 is only an example of illustration, and the specific structure form and the number of channels do not constitute the limiting conditions of the application.

[0083] Reference Figure 2 , Figure 2 An example of an application scenario provided by the embodiment of the application.

[0084] Figure 2 The application scenario is shown to include a battery end and a charging control end, wherein:

[0085] The battery end includes a battery state estimation module and a disturbance diagnosis module, wherein the battery state estimation module is used to generate an initial health state estimation value based on the instantaneous state parameters after the battery is connected to the swap cabinet, and the disturbance diagnosis module is used to apply a disturbance current to the battery before charging is started and extract the voltage response and impedance characteristics to update the initial health state estimation value.

[0086] The charging control end includes a target function calculation module and a charging strategy generation module, wherein the target function calculation module constructs a target optimization function based on the health state posterior estimation value and the corresponding uncertainty index, combined with the charging time, the battery health degradation cost, the electricity price cost and the thermal management cost, and the charging strategy generation module outputs the charging rate, the stage timing and the resource allocation scheme adapted to the battery according to the target optimization function.

[0087] It can be understood that Figure 2 is only a logical division of functional modules, and does not limit the specific implementation form. Each module can be realized by independent software and hardware units, or can be realized by program instruction calling in the same controller.

[0088] Next, the multi-objective charging strategy optimization method considering battery health degradation provided by the embodiment of the application is introduced in combination with the drawings. The application is applied to a battery swap cabinet, and the method comprises the following steps. Figure 3

[0089] S1: Collecting instant state parameters of the battery, mapping the battery state to a preset battery colony health model library according to the instant state parameters, determining a reference battery model of the battery, and obtaining an initial health state estimation value;

[0090] In this embodiment, by introducing the matching mechanism of the colony health model library, the single-point inference problem caused by the lack of complete historical data of individuals is avoided, so that the battery can obtain a more reasonable initial health estimation value relying on the statistical characteristics of the colony model.

[0091] It can be understood that the instant state parameters in the application come from the static stage after the battery is placed in the battery swap cabinet. At this time, the static stage is different from the stage before the charging starts. Specifically, it refers to the time window after the battery completes the physical connection with the charging channel and before the charging is activated. In this stage, the electrochemical system of the battery is not disturbed by external current, and the terminal voltage and open circuit voltage can better reflect the true balance state. The temperature parameter is also in a steady state condition under natural conduction, so the collected data is more representative and robust.

[0092] Further, the instant state parameters in the application include the open circuit voltage, terminal voltage and temperature of the battery after being connected to the battery swap cabinet. The open circuit voltage is used to represent the preliminary level of the overall state of charge of the battery, the terminal voltage is used to reflect the polarization degree and internal state difference at the access moment, and the temperature is used to depict the initial thermal state of the battery under the action of the environment and self-heating.

[0093] S2: Before the charging starts, performing a disturbance current diagnosis on the battery to obtain a voltage response and impedance characteristics;

[0094] S3: Updating the initial health state estimation value according to the voltage response and impedance characteristics to obtain a health state posterior estimation value and a corresponding uncertainty index;

[0095] In this embodiment, the disturbance current diagnosis is combined with model correction and Bayesian regression, which not only improves the accuracy of the battery health state estimation, but also quantifies the uncertainty index in a statistical sense. The uncertainty index reflects the credibility of the estimation result, which can provide a risk boundary for subsequent charging strategy selection, thereby avoiding the problem of excessive aggressiveness or excessive conservatism caused by simply relying on experience parameters in the traditional method.

[0096] ​S4: Based on the posterior estimate of the health status and the corresponding uncertainty index, and combined with charging time, battery health degradation cost, electricity price cost and thermal management cost, construct an objective optimization function to generate a charging strategy adapted to the battery.

[0097] In this embodiment, by incorporating an uncertainty index into the objective optimization function, the strategy generation process not only possesses risk-adaptive capabilities but also achieves a dynamic balance between battery health degradation, cost consumption, and safety boundaries. For example, when uncertainty is high, the optimization results automatically favor conservative strategies to reduce risk; while when uncertainty decreases and electricity prices or thermal management conditions are relaxed, the strategy is allowed to be moderately aggressive to shorten charging time or reduce costs. This uncertainty-driven strategy generation method overcomes the limitations of traditional fixed optimization functions and solves the problem of charging strategy deviation caused by the lack of historical data in battery swapping cabinet scenarios.

[0098] Before detailing the specific steps, this application embodiment needs to explain the technical challenges in the application scenario of battery swapping cabinets.

[0099] In practical applications, battery swapping cabinets need to centrally manage charging for batteries from different vehicle models, usage habits, and degradation stages. Because battery swapping cabinets cannot access complete historical charge and discharge data, health status assessments often rely solely on current instantaneous measurements. This masks individual differences in group charging scenarios, leading to systemic problems in strategy generation that deviate from optimality. Further complicating matters, battery degradation is not a single variable but involves multiple dimensions such as capacity retention, internal resistance growth rate, polarization characteristics, and thermal sensitivity. Without statistical priors and dynamic corrections, estimating these states can easily lead to strategy imbalances among different batteries, ultimately affecting overall charging efficiency and safety.

[0100] The proposed solution in this embodiment goes beyond simply making single-point judgments based on real-time data. Instead, it introduces statistical priors from a battery population health model library to establish a mapping relationship between individuals and the population, thus enabling a reasonable initial health estimate even in the absence of historical data. Simultaneously, through disturbance current diagnosis and multi-dimensional feature extraction, the battery's dynamic voltage response and impedance behavior are incorporated into the posterior correction process, allowing the health state estimate to be iteratively updated by combining priors and actual measurements. More importantly, this embodiment quantifies the uncertainty in health state estimation into an observable index and uses this index as the core adjustment variable for charging strategy optimization. By constructing a multi-objective optimization function that includes charging time, degradation cost, electricity price cost, and thermal management cost, the charging strategy is guided to adaptively adjust between conservatism and aggressiveness, avoiding excessive risk-taking or excessive conservatism due to insufficient information.

[0101] It can be understood that the core idea of the method of the application is not a single modeling or diagnosis method, but to couple the health estimation and the uncertainty index as a unified framework to form a closed loop path. Compared with the traditional method, the application can dynamically balance the contradiction between safety, economy and life dimension when facing the differences of battery groups and information missing, especially suitable for battery replacement frequently, complex sources and lack of complete historical operation data of battery replacement cabinet scene, with higher stability and applicability.

[0102] Next, the technical content of the initial health state estimation value of the embodiment of the application is further expanded.

[0103] It can be understood that the initial health state estimation value of the application is obtained by mapping the instant state parameters of the battery to the population health model library. The basic principle is that in large-scale operation scenarios, although different batteries have differences in factory design, cell batch, running habits and environmental conditions, they often show limited typical trajectories in capacity retention rate and internal resistance growth rate and other core degradation modes. Therefore, by pre-constructing a population health model library, a large amount of historical observation data is summarized into several reference models, and each reference model carries the feature distribution of voltage, temperature, impedance and other features associated with a specific degradation mode.

[0104] Further, after the battery is placed in the battery replacement cabinet, the open-circuit voltage, terminal voltage and temperature and other instant parameters are collected, and the similarity between the instant parameters and the reference feature vectors in the population health model library in the feature space can be measured. Since these reference features have included the statistical rules formed by the evolution of health degradation, the instant parameters will be highly consistent with a certain type of degradation mode in this space, so that the battery can be reasonably mapped to the closest reference model. In other words, even if the complete historical operation data is lacking, the battery can still obtain the preliminary health state estimation by comparing with the population model.

[0105] As can be appreciated by those skilled in the art, by using the statistical prior of the battery population health model library, the estimation bias is effectively reduced by avoiding relying solely on the linear extrapolation of instantaneous voltage or temperature values; at the same time, the selection process of the reference battery model is essentially a process of finding the optimal approximation in a limited mode set, and thus has convergence and stability. Further, the reference battery model not only provides quantitative estimation of the capacity retention rate and the internal resistance growth rate, but also provides prior parameters for subsequent disturbance current diagnosis, making the posterior update more targeted. Compared with the existing scheme relying on average parameters or empirical formulas, the application realizes the initial health state estimation with statistical reason under incomplete information by introducing the population health model library.

[0106] In some optional embodiments, the construction of the battery population health model library includes the following three aspects. To avoid over-limitation, the required numerical values, algorithms and thresholds can be replaced or equivalently deformed as needed.

[0107] In the first aspect, the detection data of the battery at the factory stage, the aging data of the bench test and part of the long-term charge and discharge data collected in operation are first used as data sources. Different sources of data often have differences in measurement environment, sampling frequency and statistical distribution, so before construction, benchmarking processing such as temperature normalization, standing verification and drift compensation is needed to ensure that all kinds of data have comparability in a unified feature space. For example, for the curve characteristics of open circuit voltage and terminal voltage, the standard open circuit voltage-state of charge surface can be fitted after temperature correction; for the direct current resistance and its evolution law over time, extreme outliers can be removed through filtering and denoising methods. Finally, a static feature vector for population model retrieval can be generated, which can be quickly mapped under the condition of immediate measurement after the battery is placed in the battery replacement cabinet, avoiding the deviation caused by the lack of complete historical data of the individual.

[0108] In the second aspect, during the construction of the battery population health model library, the modeling process of dynamic response and impedance parameterization is also included. Specifically, based on the pulse current fragments in the bench test or historical operation records, the voltage transient change under small perturbation condition can be extracted, and the impedance parameters such as dynamic resistance, polarization time constant and thermal sensitivity can be fitted by combining the equivalent circuit model. In order to ensure that the matching between the battery individual and the reference model can be quickly completed during subsequent diagnosis, impedance templates are established for each typical state of charge and temperature point in the battery population health model library, and features such as voltage transient offset, recovery slope and steady-state difference value are stored in a vectorized manner. In this way, after the battery enters the battery replacement cabinet, it can first be positioned through static features, and then corrected dynamically through disturbance current excitation and impedance templates, so as to obtain more accurate health status estimation value.

[0109] In the third aspect, during the long-term maintenance of the battery population health model library, clustering induction and online evolution logic are also included. Specifically, during the initial library construction, the benchmarked static and dynamic features can be fused and clustered to obtain several typical reference model prototypes, and the degradation curves of capacity retention rate and internal resistance growth rate with cycle number or temperature stress can be fitted for each prototype. In actual operation, as the battery replacement cabinet gradually accumulates more data of connected batteries, new data can be backfilled into existing prototype models after consistency test, or trigger new prototype splitting and merging when the deviation is large, to maintain the adaptability of the model library to different batches of batteries and new working conditions.

[0110] Reference Figure 4 , Figure 4A flowchart of calculating the initial health state estimation value provided by the embodiments of the present application.

[0111] In one example, the battery state is mapped to a preset battery colony health model library according to the instant state parameters to obtain an initial health state estimation value, including:

[0112] S1.1: Similarity calculation is performed on the terminal voltage and a reference battery feature vector in the battery colony health model library, and a reference battery model of the battery is determined according to the result of the similarity calculation.

[0113] S1.2: The open-circuit voltage and temperature are matched with a reference open-circuit voltage-temperature curve corresponding to the reference battery model in the battery colony health model library to determine an initial state of charge correction value of the battery.

[0114] S1.3: The initial health state estimation value is output according to the reference battery model and the initial state of charge correction value, wherein the initial health state estimation value at least includes one of capacity retention rate and internal resistance growth rate.

[0115] Next, the technical content of the embodiments of the present application on disturbance current diagnosis is further expanded.

[0116] Reference Figure 5 , Figure 5 A flowchart of calculating the voltage response and impedance characteristics provided by the embodiments of the present application.

[0117] In one example, the battery is subjected to disturbance current diagnosis to obtain voltage response and impedance characteristics, including:

[0118] S2.1: In the static stage of the battery being connected to the battery replacement cabinet and not starting charging, historical charging data of the reference battery model is called to generate a disturbance current excitation sequence, wherein the disturbance current excitation sequence satisfies a zero net electric quantity constraint.

[0119] Specifically, when the battery is just connected to the battery replacement cabinet but has not yet entered the regular charging link, the battery is in a relatively stable static state, and the terminal voltage and temperature of the battery are not affected by external large current impact, so this stage is suitable for performing small amplitude disturbance to obtain diagnostic signals. In order to ensure that the diagnostic process does not cause significant changes in the state of charge of the battery, the disturbance current excitation sequence must satisfy the zero net electric quantity constraint when designed, that is, the state of charge of the battery remains basically unchanged after the excitation process is completed.

[0120] In this embodiment, the generation of the disturbance current excitation sequence is not randomly set, but is calculated based on the historical charging data of the reference battery model. The current response segments under different working conditions are first decomposed in the frequency domain, and then the frequency components most sensitive to impedance characteristics are selected as candidate excitations. Subsequently, the final current sequence is formed through amplitude scaling and phase adjustment, and it is ensured that it is mutually offset in the positive and negative current directions.

[0121] In one example, the specific steps of S2.1 are as follows:

[0122] S2.1.1: Frequency domain decomposition is performed on the historical charging data to extract frequency components and generate candidate disturbance sequences;

[0123] In this embodiment, the historical charging current sequence of the reference battery model is subjected to fast Fourier transform or wavelet decomposition, converting the time domain signal into energy distribution in the frequency domain. In this way, the energy proportion corresponding to different frequency components can be clearly separated, and it can be identified which frequency components are most sensitive to voltage response. Subsequently, the key frequency interval that can effectively excite the polarization effect and diffusion dynamics of the battery is selected as the preferred candidate interval, and the components with too low energy or susceptible to noise interference are removed in combination with the empirical threshold. Finally, these frequency components are recombined into a group of representative candidate disturbance sequences, each of which can highlight a certain dynamic characteristic of the battery within a limited time scale.

[0124] S2.1.2: The candidate disturbance sequences are optimized under the condition of satisfying the zero net electric quantity constraint to obtain a first disturbance current and a second disturbance current that are orthogonal to each other;

[0125] In this embodiment, by applying constraints in the frequency domain and the time domain at the same time, the integral result of the disturbance sequence in one complete cycle is zero, thereby ensuring that the overall electric quantity of the battery does not change significantly after the diagnosis process is completed. In addition, in order to improve the independence of the diagnosis information, multiple candidate sequences are designed to be orthogonalized, so that they do not overlap in frequency distribution and phase characteristics. Finally, the first disturbance current and the second disturbance current are obtained, which have strong complementarity in energy distribution and signal response.

[0126] S2.1.3: The first disturbance current and the second disturbance current are modified through amplitude scaling and phase adjustment to generate the disturbance current excitation sequence;

[0127] In this embodiment, the amplitude scaling process is to limit the peak value of the perturbation current according to the current bearing capacity and safety threshold of the battery, so that it can excite a sufficient voltage response without causing overheating or overcurrent risk; the phase adjustment process is to control the superposition relationship of different frequency components on the time axis by changing the relative displacement of the positive and negative half cycles, so that the excitation signal is more smooth in time domain and avoids sudden changes. The modified perturbation current excitation sequence can cover the target frequency components and has good energy distribution characteristics and safety. The finally generated sequence can apply a series of small perturbations to the battery in a very short time, ensuring the distinguishability of the subsequent voltage response and the accuracy of parameter extraction, while not affecting the actual state of charge and working life of the battery.

[0128] S2.2: excite the battery by the perturbation current excitation sequence, and synchronously sample the battery terminal voltage to obtain a voltage original sequence;

[0129] In this embodiment, after the perturbation current excitation sequence is loaded to the battery port, it will cause a slight fluctuation of the voltage in a very short time scale. In order to ensure the accuracy of the data, a high-precision synchronous sampling circuit is used to collect the battery terminal voltage, and the sampling frequency needs to cover the upper limit of the excitation frequency component, so as to avoid aliasing effect. Unlike traditional constant current charging and discharging, this excitation method can provide rich transient information in a short time and will not significantly change the temperature or state of charge of the battery. The voltage original sequence obtained by synchronous sampling completely retains the response curve of the battery under controlled small perturbation, which contains signal characteristics of the superposition of various mechanisms such as battery internal polarization, charge transfer and thermal effect, providing a basis for subsequent decomposition and feature extraction.

[0130] S2.3: orthogonal demultiplexing and matched filter array projection are performed on the voltage original sequence to obtain a corresponding voltage feature coefficient vector;

[0131] Specifically, since the voltage original sequence often contains noise components and the coupling of multiple frequency components, it is difficult to directly distinguish the signal contributions of different physical mechanisms without processing, so it is necessary to decompose the signal into a group of independent response components by orthogonal demultiplexing method.

[0132] In this embodiment, first, a group of orthogonal basis functions is constructed, and the original sequence is projected into multiple independent channels based on the frequency domain orthogonality, thereby realizing the decoupling of different frequency and phase components. Then, combined with the filter array calculated in advance based on the historical charging data of the reference battery model, the decoupled signal is projected with the filter operator, so as to extract a stable voltage feature coefficient vector in a multi-dimensional feature space.

[0133] In one example, the specific steps of S2.3 are as follows:

[0134] S2.3.1: decompose the voltage raw sequence in a preset feature space to obtain a plurality of independent response components;

[0135] Specifically, the collected voltage raw sequence is input into the preset feature space, which is constructed by time domain, frequency domain or time-frequency joint analysis method.

[0136] In this embodiment, wavelet packet decomposition or empirical mode decomposition can be used to decompose the voltage raw sequence into a plurality of frequency band components, each component corresponding to the response characteristics of the battery at different time scales or different frequency ranges. In this way, independent response components can be obtained to ensure that the subsequent processing can separate the signal components corresponding to different mechanisms such as polarization process, diffusion process and steady state process of the battery.

[0137] S2.3.2: calculate a projection matrix and a projection operator based on historical charging data;

[0138] In this embodiment, after obtaining the decomposed response components, a training sample set is constructed using the historical charging data of the reference battery model. By performing principal component analysis or canonical correlation analysis on the voltage change curves in the training sample set, the main feature patterns are extracted, and the projection matrix is calculated accordingly. The projection matrix can maximize the separability between different response components, ensuring that different features have good distinguishability in low-dimensional space. Meanwhile, the corresponding projection operator is further obtained based on the above projection matrix. The projection operator can be regarded as a mapping rule for projecting the original component sequence into a unified feature coordinate system.

[0139] S2.3.3: map the response components to a voltage feature coefficient vector based on the projection matrix and the projection operator;

[0140] Specifically, after the projection matrix and the projection operator are determined, each response component is substituted into the projection operation one by one to obtain the numerical expression in the feature space, i.e. the voltage feature coefficient vector. The coefficients corresponding to each component represent the contribution and pattern characteristics of the component in the overall response. The final voltage feature coefficient vector is a set of low-dimensional but high-information-density features, which can be directly used for impedance parameter correction or subsequent state of health estimation calculation.

[0141] S2.4: extract features from the voltage feature coefficient vector to obtain a voltage response, wherein the voltage response includes voltage transient offset, recovery slope and steady state difference value;

[0142] Specifically, the voltage transient offset describes the instantaneous voltage change of the battery at the beginning of excitation. This index is closely related to the polarization effect and contact internal resistance of the battery. The recovery slope is used to characterize the speed at which the battery voltage recovers to the equilibrium state after the excitation stops. Its magnitude reflects the efficiency of ion diffusion and charge migration. The steady-state difference value represents the average change of voltage under long-term excitation and can be used to infer the thermal effects and slow polarization phenomena inside the battery.

[0143] In this embodiment, the voltage characteristic coefficient vector is first divided into time-domain windows, dividing the data from the three stages of excitation start, excitation maintenance, and excitation end into transient, steady-state, and recovery segments, respectively. By searching for the initial abrupt change point of the voltage within the transient segment and calculating the difference between this abrupt change point and the equilibrium voltage, the voltage transient offset value can be obtained. The processing typically employs a combination of differential operations and extreme value detection to avoid abrupt change point identification errors caused by noise disturbances.

[0144] Furthermore, in the recovery phase, the voltage curve is subjected to linear regression or sliding window fitting to extract the trend of voltage gradually approaching equilibrium over time, and the slope of the fitted line is calculated as the recovery slope. To ensure the stability of the slope calculation, the initial transition region can be excluded in the recovery phase, and the least squares method can be used for global fitting in the remaining interval to obtain more stable gradient information.

[0145] Furthermore, in the steady-state region, the voltage curve is smoothed using a moving average or low-pass filtering method. The average voltage difference during the excitation duration is calculated and compared with the resting voltage to obtain the steady-state differential value. The steady-state differential value reflects the overall voltage change amplitude under long-term excitation. Its extraction process focuses on reducing high-frequency noise interference; therefore, frequency-domain based filtering methods are often used to ensure the reliability of the steady-state voltage change.

[0146] Furthermore, the transient offset, recovery slope, and steady-state differential value are weighted and summed to obtain the voltage response.

[0147] S2.5: Correct the impedance parameters in the reference battery model based on the voltage response to obtain impedance characteristics, wherein the impedance characteristics include dynamic internal resistance, polarization time constant and temperature rise sensitivity;

[0148] Specifically, the correction of the dynamic internal resistance is based on the ratio of the voltage transient offset to the current excitation amplitude, and the equivalent series resistance value of the battery is adjusted by comparing the difference between the reference model prediction and the actual observation; the correction of the polarization time constant is combined with the fitting of the recovery slope, so that the model can be closer to the actual ion diffusion rate and polarization dynamic process of the battery; the correction of the temperature rise sensitivity is realized by the joint correction of the steady-state difference value and the temperature sensor data, which is used to reflect the thermal response level of the battery under a specific load.

[0149] In the embodiment, first, the dynamic internal resistance of the battery is corrected based on the ratio of the voltage transient offset to the excitation current amplitude.

[0150] It can be understood that, in the initial excitation stage, the mutation of the voltage is mainly caused by the internal resistance of the battery, and by dividing the transient offset by the corresponding current mutation, the initial estimated value of the dynamic internal resistance can be directly obtained. In order to reduce the influence of random noise and contact resistance, the estimated value can also be filtered and corrected in combination with the reference internal resistance range in the reference battery model library, so as to obtain the updated parameter of the dynamic internal resistance.

[0151] Further, the polarization time constant of the battery is corrected based on the recovery slope. The polarization time constant usually reflects the dynamic process of charge migration and ion diffusion, and by matching the recovery slope with the preset polarization model, the characteristic time required for the battery to recover from the polarization state to the equilibrium state can be calculated. In actual calculation, an exponential regression model can be used to fit the voltage curve of the recovery section to obtain the correction value of the time constant, and the correction value is updated to the reference battery model to improve the fitting accuracy of the impedance model to the actual electrochemical process.

[0152] Further, based on the steady-state difference value, the temperature rise sensitivity of the battery under long-time excitation is further evaluated. Since the steady-state voltage offset is closely related to the internal heat accumulation and slow polarization of the battery, by combining the steady-state difference value with the temperature change amount collected by the battery swap cabinet in real time, the coupling relationship between temperature and voltage change can be quantified. This relationship can be used to correct the thermal parameters in the reference model, so as to obtain the updated value of the temperature rise sensitivity. In order to improve the reliability of the calculation, cross verification can also be performed through a plurality of disturbance current tests with different amplitudes, so as to ensure that the estimation of the temperature rise sensitivity is not limited by a single working condition.

[0153] Next, the technical content of the embodiment of the application about updating the initial health state estimation value is further expanded.

[0154] In one example, the initial health state estimation value is updated according to the voltage response and impedance characteristics to obtain a health state posterior estimation value and a corresponding uncertainty index, including:

[0155] S3.1: forward simulating the perturbation current excitation sequence under the reference battery model to obtain a predicted voltage response and a predicted impedance parameter, and matching the predicted voltage response and the predicted impedance parameter with the voltage response and the impedance feature to calculate a time-domain residual and a parameter deviation;

[0156] Specifically, in order to accurately depict the dynamic response of the actual battery under the action of the perturbation current, the same current excitation needs to be forward simulated by the reference battery model to obtain the predicted voltage change curve and the corresponding impedance feature. The forward simulation is not a simple numerical calculation, but a comprehensive solution based on factors such as the equivalent circuit structure, the polarization branch and the temperature coupling parameter in the reference battery model, so that the prediction result can cover the dynamic response characteristics of multiple types of parameters such as ohmic resistance, polarization capacitance and charge transfer impedance.

[0157] In the present embodiment, in the forward simulation process, the perturbation current sequence is first discretized to match the time resolution of the battery voltage sampling, and then the discretized current is injected into the electrochemical equivalent circuit of the reference battery model as an input. The electrochemical equivalent circuit includes a direct current resistance branch, a parallel diffusion impedance branch and a temperature-dependent polarization branch. The voltage response is solved at different time constants during simulation, and the predicted voltage curve is obtained. At the same time, according to the thermal effect parameters embedded in the model, the temperature rise under current injection can be obtained, and the predicted impedance parameters are further derived. Then, the predicted voltage response is compared with the actual sampled voltage response point by point, and the time-domain residual curve is calculated using a sliding window, so as to quantify the deviation between the prediction and the measurement. Then, the predicted impedance is compared with the impedance feature extracted from the voltage feature, and a parameter deviation vector is obtained.

[0158] S3.2: constructing a likelihood function according to the time-domain residual, the parameter deviation and a preset sampling noise covariance, taking the initial health state estimation value as a priori, updating the likelihood function and the priori by Bayesian regression to obtain a health state posterior estimation value and a posterior covariance matrix corresponding to the health state posterior estimation value;

[0159] Specifically, the estimation of the battery health state belongs to an uncertainty inference problem, and a single point estimation cannot cover the volatility that may occur in the actual operation of the battery. Therefore, the residual and the deviation need to be converted into a statistical likelihood function, and the existing initial health state estimation value is combined as prior information to form an updateable probability inference framework. Through the Bayesian regression method, the estimation of the health state can be dynamically updated under the consideration of noise and model uncertainty, so as to obtain a more realistic posterior estimation value and its variance quantification.

[0160] In this embodiment, firstly, the observation likelihood is constructed by using the time-domain residual and parameter deviation, which is modeled as a multi-dimensional Gaussian distribution, where the noise covariance is set by the resolution of the sampling device and the known measurement accuracy. Then, the posterior distribution of the state-of-health parameters such as capacity retention rate and internal resistance growth rate is obtained by Bayesian inference combined with the observation likelihood, taking the initial state-of-health estimation value as the prior input. The posterior distribution not only outputs the corrected estimation mean value, but also gives the covariance matrix as an uncertainty measure.

[0161] Further, in this embodiment, a regularization factor is also introduced to prevent overfitting. When the residual distribution is too concentrated, the weight of the noise covariance is adjusted to avoid excessive dependence on the measurement data of a batch. At the same time, for the multi-battery scenario across channels, the posterior distribution of different batteries can be jointly corrected by multi-source Bayesian updating, so that the consistency and stability of the estimation can be maintained in the overall battery swap cabinet application.

[0162] S3.3: Perform consistency test on the two groups of voltage feature coefficient vectors obtained by the mutually orthogonal disturbance current excitation sequence, and calculate the cross-sequence consistency deviation;

[0163] Specifically, a single disturbance current sequence may be affected by noise or local nonlinear effects, resulting in deviation in the estimation result. Therefore, two groups of mutually orthogonal disturbance current sequences are used to excite the battery respectively, and the voltage feature coefficient vectors are extracted, and the consistency of the results is tested. If the difference between the two groups of data is too large, it means that there is strong random fluctuation or non-ideal factor in the response of the battery, which needs to be included in the uncertainty index.

[0164] In this embodiment, firstly, the voltage feature vectors obtained under the first disturbance current and the second disturbance current excitation are normalized respectively to eliminate the influence of amplitude and sampling resolution. Then, the cosine similarity and Euclidean distance between the two groups of vectors are calculated, and the cross-sequence consistency deviation is obtained. If the deviation is within a reasonable threshold, it means that the voltage features under the two independent excitations are consistent, and it can be considered that the response of the battery is stable; otherwise, it is considered that there is strong uncertainty.

[0165] S3.4: The impedance features are spliced into a feature vector, and the distance measure is calculated by Mahalanobis distance, and the model domain distance index is obtained;

[0166] Specifically, impedance features can not only be used alone to reflect the electrochemical characteristics of a battery, but also serve as a holistic feature vector for comparison with a population model to quantify the degree of difference between the target battery and the reference model. Mahalanobis distance is introduced in this process because it can calculate distances while considering the correlation between different feature dimensions, making it more suitable for data scenarios with strong multidimensional correlations than simple Euclidean distance.

[0167] In this embodiment, impedance characteristics such as dynamic internal resistance, polarization time constant, and temperature rise sensitivity are first concatenated sequentially to form a multidimensional feature vector. This vector is then compared with a reference distribution in a population model library to calculate its Mahalanobis distance. A smaller distance indicates that the target battery's behavior is highly consistent with the population distribution, which can improve the confidence level of the estimate. A larger distance suggests that the battery may be in an abnormal state or deviating from the mainstream healthy trajectory, requiring it to be marked as a high-uncertainty individual.

[0168] S3.5: The spectral radius of the posterior covariance matrix, the cross-sequence consistency deviation, and the model domain distance index are weighted and fused to obtain the uncertainty index, wherein each weight is calculated based on the similarity between the battery and the reference battery model;

[0169] Specifically, uncertainty indicators need to consider multiple information sources, as a single dimension cannot fully reflect the reliability of the estimate. Therefore, by weighting and fusing the spectral radius of the posterior covariance matrix, cross-sequence consistency bias, and model domain distance, a comprehensive characterization of the confidence level of the battery state estimate can be achieved.

[0170] In this embodiment, the spectral radius is used to measure the diffusion of the estimated distribution after the Bayesian update, the consistency deviation is used to measure the stability of the estimation results under different perturbation stimuli, and the model domain distance is used to measure the deviation of the target battery from the population distribution. These three are fused together using a weighted summation method to form the final uncertainty index. The weights are not fixed values ​​but are dynamically calculated based on the similarity between the target battery and the reference battery model. Higher similarity indicates greater reliance on the residuals and covariance results, while lower similarity increases the weights of the model domain distance and consistency deviation.

[0171] Next, we will further elaborate on the technical content of the objective optimization function calculation in the embodiments of this application.

[0172] It is understandable that generating a charging strategy requires simultaneously considering multiple factors such as charging efficiency, battery health degradation, electricity costs, and thermal management energy consumption. However, when the battery health status is uncertain, directly using a single objective function often leads to bias. Therefore, this application introduces an uncertainty-driven objective optimization function, incorporating both the posterior estimate of the health status and the uncertainty index into the modeling process to ensure the reliability and robustness of the optimization results under multi-objective conditions.

[0173] In this embodiment, the construction of the target optimization function first takes the health state posterior estimate value as the reference range of the battery's bearable charging intensity, ensuring that the current safety limit of the battery will not be exceeded when the strategy is generated. At the same time, the uncertainty index is taken as a risk adjustment factor, which acts on each cost item and constraint condition. For example, when the uncertainty is high, the weight of the battery health degradation cost is automatically increased, thereby inhibiting aggressive charging behavior; when the uncertainty is reduced, the weights of the charging time item and the electricity price cost item can be moderately increased to promote more efficient charging strategies. In this way, the target optimization function not only embodies the effect of multi-objective weighting, but also realizes the dynamic adjustment of risk adaptation.

[0174] Further, the specific composition of the target optimization function in this embodiment includes four aspects: first, the charging time item is used to constrain the total charging time and achieve peak-valley time-of-use optimization through association with the grid time period price; second, the battery health degradation cost item is calculated through the capacity degradation rate and impedance growth parameter in the posterior estimate value, quantifying the contribution value of a charging process to the life; third, the electricity price cost item is weighted and integrated according to the real-time electricity price curve to control the user's economic expenditure; fourth, the thermal management cost item estimates the heat dissipation power consumption according to the temperature rise prediction model and is associated with the cooling unit energy consumption, thereby reducing the operating energy consumption while ensuring controllable battery temperature.

[0175] In one example, the construction of the target optimization function combining the charging time, the battery health degradation cost, the electricity price cost, and the thermal management cost includes:

[0176] S4.1: The characteristic factors corresponding to the charging time, the battery health degradation cost, the electricity price cost, and the thermal management cost are respectively taken as basic target items;

[0177] Specifically, in the charging strategy optimization, multiple conflicting objectives need to be considered at the same time, such as the contradiction between shortening the charging time and reducing the health degradation. If only a single optimization target is set, other performance indicators will deviate from the actual demand. Therefore, this step first abstracts the charging time, the battery health degradation cost, the electricity price cost, and the thermal management cost into measurable basic target items, so that each optimization direction can be independently expressed in mathematical form.

[0178] In this embodiment, the charging time target item is defined by the cumulative time required for the battery to charge to the target state of charge; the battery health degradation cost target item is quantified by the predicted values of the capacity retention rate decline rate and the internal resistance growth rate; the electricity price cost target item is calculated by weighting the battery's charging power in different time periods and the electricity price curve; and the thermal management cost target item is defined by associating the auxiliary energy consumption of cooling fans, liquid cooling devices, etc. with the temperature rise model.

[0179] S4.2: define a basic target item corresponding to the charging duration and the battery health degradation cost as a first function based on the capacity retention rate and the internal resistance growth rate, and take the confidence interval boundary of the health state posterior estimation value as a penalty factor of the first function;

[0180] Specifically, there is a natural conflict between battery health degradation and charging speed. If the charging duration is simply shortened, it is easy to cause rapid capacity degradation or abnormal increase of internal resistance. This step establishes a first function through two indicators of capacity retention rate and internal resistance growth rate, to depict the balance between charging duration and health degradation cost.

[0181] In this embodiment, the health state posterior estimation value provides the most likely interval of battery capacity and internal resistance, and the confidence interval boundary reflects the uncertainty range of health estimation. By introducing the confidence interval boundary as a penalty factor, the value range of the first function can be dynamically constrained. For example, when the uncertainty is high, the penalty factor will be expanded, forcing to inhibit the charging rate too aggressively; and when the health state estimation is reliable, the penalty factor is weakened, allowing a higher charging rate.

[0182] S4.3: expand the basic target item corresponding to the electricity price cost into a second function according to the time sequence, and adjust the penalty factor of the second function through the electricity price of different time periods;

[0183] Specifically, the volatility of electricity price determines that the charging cost is not a fixed value, but depends on the charging power distribution of the battery at different times. Therefore, it is necessary to expand the electricity price cost into a function that changes with time, to accurately depict the coupling relationship between charging time sequence and electricity cost.

[0184] In this embodiment, the second function is defined by the time sequence integral of the electricity price curve and the charging power. That is, during the peak electricity price period, the charging behavior will correspond to a higher cost, while during the valley period, the cost is relatively low. In order to further enhance the flexibility of the strategy, this step introduces a penalty factor about the electricity price, to dynamically adjust the priority of charging in different time periods. When the electricity price is high, the penalty factor is increased, forcing the optimization algorithm to reduce the charging power in the corresponding period; when the electricity price is low, the penalty factor is reduced, so that the optimization result is more inclined to concentrate charging in the low-cost period.

[0185] S4.4: correlate the basic target item corresponding to the thermal management cost with the preset battery temperature prediction trajectory in the reference battery model of the battery, to obtain a third function, and adjust the temperature rise allowed range through the uncertainty indicator, to obtain a penalty factor of the third function;

[0186] Specifically, the battery charging process is accompanied by energy conversion and side reactions, which can cause internal temperature rise. If not effectively controlled, it will lead to material performance degradation and even trigger thermal runaway. This step couples the battery temperature with the cooling energy consumption by introducing a thermal management cost term, so that the temperature rise prediction directly affects the charging strategy optimization.

[0187] In this embodiment, the third function is established by coupling the temperature prediction trajectory in the battery model with the cooling power consumption model. The temperature prediction trajectory reflects the temperature evolution under a given charging rate, while the cooling power consumption increases with the temperature rise amplitude. To further improve safety, this step introduces an uncertainty index as a penalty factor. When the uncertainty is high, the temperature rise allowable range automatically shrinks, increasing the weight of cooling cost, thereby avoiding potential overheating risk; when the uncertainty is low, the allowable range is relaxed to reduce unnecessary cooling energy consumption.

[0188] S4.5: Calculate the target optimization function according to the first function, the second function and the third function, and the penalty factors corresponding to each function;

[0189] In this embodiment, the target optimization function, in the calculation process, will first meet the probabilistic safety boundary constrained by the uncertainty index, and then realize the balanced optimization of multiple objectives under this constraint. In this way, the optimization result not only converges to the optimal solution in the mathematical sense, but also ensures the comprehensive balance between battery life, electricity cost and cooling energy consumption in engineering practice.

[0190] Next, the technical content of the application embodiment for generating a charging strategy is further expanded.

[0191] In one example, the generating a charging strategy adapted to the battery comprises:

[0192] S4.6: Divide the charging process into multiple stages according to the state of charge interval, and construct a risk envelope function for each stage according to the uncertainty index, the risk envelope function being used to parameterize and reshape the upper limit of the charging rate, the temperature rise range and the proximity penalty of the target state of charge for each stage;

[0193] Specifically, the electrochemical behavior of the battery differs in different state of charge intervals. For example, at low state of charge, the polarization effect is small, and a higher rate of charging can be allowed, while at high state of charge, the side reactions are active, and if a high rate is continued, irreversible damage will occur. Therefore, dividing the charging process into multiple stages helps to constrain the charging conditions in each stage. Further, due to the uncertainty in the state of health estimation, the risk envelope function is used to dynamically correct the charging rate, temperature rise range and proximity of the terminal state of charge for each stage, in order to offset the potential safety hazards caused by prediction bias.

[0194] In the embodiment, the construction of the risk envelope function is based on the numerical size of the uncertainty indicator. When the uncertainty indicator is high, the upper limit of the charging rate of the corresponding stage is automatically contracted, the temperature rise allowed range is narrowed, and the penalty term for approaching the target state of charge is increased, thereby forcing the optimization algorithm to select a more robust charging method; when the uncertainty indicator is low, the risk envelope function relaxes the restrictions, allowing the strategy to have room to improve charging efficiency while meeting safety constraints.

[0195] S4.7: Calculate a probabilistic safety constraint based on the confidence boundary of the risk envelope function, wherein the probabilistic safety constraint represents the probability that the battery state of charge falls within the safety domain within the uncertainty indicator value range;

[0196] Specifically, a single deterministic safety threshold often cannot cope with the uncertainty predicted by the model, so it is necessary to introduce a probabilistic safety constraint based on the risk envelope function. This constraint no longer requires all charging states to strictly fall within the safety domain, but is defined in a probabilistic form, i.e., as long as within the uncertainty indicator value range, the charging state satisfies the safety condition with a certain probability.

[0197] In the embodiment, the confidence boundary of the risk envelope function serves as the core reference for the probabilistic safety constraint, and the posterior distribution of the health state estimate is statistically simulated for the rate, voltage and temperature trajectory of each stage. Then the probability that the battery state point falls within the safety domain under a given uncertainty range is calculated and compared with the preset safety threshold. If the probability is insufficient, the charging rate or the timing is automatically adjusted to improve safety.

[0198] S4.8: Under the condition that the probabilistic safety constraint is met at each stage, predict the uncertainty evolution trajectory during the charging process according to the health state posterior estimate, and calculate the information gain of each stage on the uncertainty indicator, wherein the expected reduction indicates the expected contraction amplitude of the uncertainty indicator after the corresponding charging rate and timing settings are executed;

[0199] Specifically, the charging process is not only a process of energy input, but also a process of obtaining battery state information. Different rate and timing settings can significantly affect the evolution of subsequent uncertainty indicators. For example, under moderate disturbance conditions, the battery response can reveal more characteristic information related to the health state, thereby reducing the estimated uncertainty. Therefore, the concept of information gain is introduced in this step to measure whether the strategy selection of each stage helps to reduce the overall uncertainty.

[0200] In this embodiment, first, the charging process is simulated based on the health state posterior estimate value and the reference battery model to predict the evolution trend of the uncertainty at each stage. Then, the contraction amplitude of the uncertainty index in the expected sense is calculated for different rates and timing configurations, and is taken as the information gain to quantify the expression. The greater the information gain, the more helpful the strategy at this stage is to improve the accuracy of future state estimation, so it should be given higher priority in the optimization process.

[0201] Further, by introducing the information gain evaluation, the optimization algorithm can actively select strategies that are both safe and reduce uncertainty while meeting safety constraints, thereby achieving dual improvement of charging efficiency and model learning effect. This mechanism is particularly suitable for the problem of lack of long-term individual historical data in the battery swap cabinet scene, and can continuously correct the model cognition and improve the dynamic adaptation ability of the strategy in operation.

[0202] S4.9: Adjust the target optimization function by taking the information gain as a positive adjustment factor, and calculate the charging mode at each stage according to the adjusted target optimization function, and summarize the charging mode to obtain a charging strategy;

[0203] Specifically, the target optimization function mainly balances multiple target factors such as time length, attenuation, electricity price, and thermal management when originally constructed, but does not directly consider the contribution of the charging process to information acquisition. Therefore, it is necessary to introduce the information gain as a positive adjustment factor into the target optimization function to promote the strategy to consider multiple target optimization while promoting the rapid contraction of uncertainty.

[0204] In this embodiment, the adjustment method is to quantify the information gain as a positive factor and adjust the weight of each basic target item. For example, when a strategy at a certain stage can significantly reduce uncertainty, the weight of the corresponding target function item is reduced, thereby encouraging the algorithm to select that strategy; otherwise, when the information gain is insufficient, the weight remains unchanged or the proportion of other cost items is increased to avoid ineffective resource waste. In this way, the target optimization function maintains a dynamic balance driven by information gain during the calculation process.

[0205] Further, the optimization algorithm uses an iterative method based on multiple target constraints, such as the combination of gradient descent and evolutionary algorithm, to ensure rapid convergence under complex constraints. Finally, the charging modes obtained at each stage are unified and summarized to form a complete charging strategy, including the rate trajectory, timing arrangement, and temperature control coordination plan.

[0206] In one example, the information gain that the calculation of each stage has an expected decreasing effect on the uncertainty index includes:

[0207] Under the probabilistic safety constraint, simulation evolution is performed on different charging rates and timing combinations to obtain corresponding uncertainty contraction trajectories;

[0208] Entropy value calculation is performed on the uncertainty contraction trajectories, and the entropy reduction amplitude is taken as a quantitative indicator of information gain.

[0209] Reference Figure 6 , Figure 6 A flowchart of another multi-objective charging strategy optimization method considering battery health degradation provided by the embodiments of the present application is shown.

[0210] In one example, the specific steps of the method are as follows:

[0211] S1: Collect the instantaneous state parameters of the battery, map the battery state to the pre-set battery population health model library according to the instantaneous state parameters, determine the reference battery model of the battery, and obtain the initial health state estimate value;

[0212] S2: Before charging starts, perform a disturbance current diagnosis on the battery to obtain a voltage response and impedance characteristics;

[0213] S3: Update the initial health state estimate value according to the voltage response and impedance characteristics to obtain a health state posterior estimate value and a corresponding uncertainty indicator;

[0214] S4: According to the health state posterior estimate value and the corresponding uncertainty indicator, construct a target optimization function combining the charging time, battery health degradation cost, electricity price cost, and thermal management cost to generate a charging strategy suitable for the battery;

[0215] The specific contents of S1 to S4 have been described in detail in the foregoing content, and the present application will not be repeated here.

[0216] S5: According to the charging strategy and the channel resource constraints of the charging channel corresponding to the battery swap cabinet, perform group charging scheduling optimization to obtain an overall charging strategy;

[0217] It can be understood that in the application scenario of the swap cabinet, multiple batteries will be simultaneously connected to different charging channels. Although each channel has independent running capability in hardware, from the overall running perspective, it is still subject to global constraints such as upper limit of grid access power, upper limit of heat dissipation capacity, and fluctuation of electricity price period. If each battery is independently run according to the individual optimal strategy, problems such as overall power overlimit, excessive heat dissipation load, or missing low electricity price period may occur, ultimately leading to increased running cost or rising equipment risk at the group level. Therefore, group scheduling optimization needs to be further introduced on the basis of the single battery strategy, so that the charging processes of different batteries can be coordinated and allocated under global constraints, ensuring individual safety and efficiency, and realizing optimal utilization of overall resources.

[0218] In the embodiment, the scheduling optimization first needs to determine the resource constraint model of the channel level, which includes three parts of total power constraint, heat dissipation capability constraint and electricity price period constraint. Among them, the total power constraint is used to limit the maximum power that the battery swap cabinet can extract from the power grid in unit time, so as to avoid local overload and cause power supply instability or power quality decline; the heat dissipation capability constraint is based on the rated heat dissipation capability of the internal air cooling or liquid cooling module of the battery swap cabinet, and establishes the balance condition between the group temperature rise and the total heat dissipation capability, so as to ensure that the group operation will not produce temperature overshoot; the electricity price period constraint is to combine the time sequence electricity price data of the external power market to adjust the power distribution of the group charging in period, so that the battery swap cabinet can complete most of the energy input in the low electricity price period, thereby reducing the overall cost.

[0219] Further, in the group scheduling optimization process, the differences between individual batteries also need to be considered. For example, when the posteriori estimation value of the state of health of some batteries shows that the attenuation level is high or the uncertainty index is large, the charging rate of the battery should be preferentially limited in the group allocation to avoid dragging the overall safety due to individual weakened batteries. In the embodiment, by adding a differentiated weight coefficient in the target optimization function, the battery in good health state can bear more power allocation under the condition of meeting the probability safety constraint, and the battery with high health risk charges at a lower rate, which can improve the energy efficiency and reliability of the group operation under the condition of constant overall power.

[0220] Further, the calculation process of the group scheduling optimization adopts a hierarchical optimization method, first completes the convergence calculation of the individual target optimization function in the individual charging strategy stage, and then constructs a unified global scheduling problem at the group level, which is solved by a multi-objective optimization algorithm. In this process, the target function not only includes the health attenuation cost and uncertainty constraint of individual battery, but also additionally increases the energy consumption cost and equipment load cost of group operation. By transferring the constraint conditions and intermediate variables between the two levels of optimization, the two-way cooperation between individual and group can be realized, which not only avoids the problem of high calculation dimension in global optimization, but also can ensure that the generated group charging strategy has executability.

[0221] Finally, the overall charging strategy obtained in this embodiment gives the upper limit of the charging power and the timing arrangement of each charging channel in each time period. The battery swap cabinet executes group control according to the scheduling result, which can ensure that the total power distribution is reasonable, the heat dissipation load is balanced, and the overall operation cost is minimized while meeting the charging demand of each battery individual. This individual-group double-layer optimization method helps to break through the limitation of the existing battery swap cabinet, which can only perform single safety protection and lacks group coordination, so that the battery swap cabinet can still maintain efficient, safe and economic operation characteristics in the scene of high concurrency and diversified battery access.

[0222] Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A multi-objective charging strategy optimization method considering battery health degradation, applied to battery swapping cabinets, characterized in that... The method includes: Collect real-time state parameters of the battery, map the battery state to a preset battery family health model library based on the real-time state parameters, determine the reference battery model of the battery, and obtain an initial health state estimate. Before charging starts, the battery is subjected to disturbance current diagnosis to obtain voltage response and impedance characteristics. The initial health state estimate is updated based on the voltage response and impedance characteristics to obtain the posterior health state estimate and the corresponding uncertainty index. Based on the posterior estimate of the health status and the corresponding uncertainty index, a target optimization function is constructed by combining charging time, battery health degradation cost, electricity price cost and thermal management cost to generate a charging strategy adapted to the battery. The battery is subjected to disturbance current diagnosis to obtain voltage response and impedance characteristics, including: During the resting phase when the battery is connected to the battery swapping cabinet but charging has not been started, the historical charging data of the reference battery model is called to generate a disturbance current excitation sequence, wherein the disturbance current excitation sequence satisfies the zero net charge constraint. The battery is excited by the perturbation current excitation sequence, and the battery terminal voltage is sampled synchronously to obtain the original voltage sequence; The original voltage sequence is orthogonally demultiplexed and projected onto a matched filter array to obtain the corresponding voltage characteristic coefficient vector; Feature extraction is performed on the voltage characteristic coefficient vector to obtain the voltage response, wherein the voltage response includes voltage transient offset, recovery slope and steady-state difference value; The impedance parameters in the reference battery model are corrected based on the voltage response to obtain impedance characteristics, wherein the impedance characteristics include dynamic internal resistance, polarization time constant and temperature rise sensitivity. The initial health state estimate is updated based on the voltage response and impedance characteristics to obtain a posterior health state estimate and corresponding uncertainty indices, including: The perturbation current excitation sequence is simulated forward under the reference battery model to obtain the predicted voltage response and predicted impedance parameters, which are then matched with the voltage response and impedance characteristics to calculate the time-domain residual and parameter deviation. A likelihood function is constructed based on the time-domain residual, parameter bias, and preset sampling noise covariance. The initial health state estimate is used as the prior. Bayesian regression is used to update the likelihood function and the prior posteriorly to obtain the posterior estimate of the health state and the posterior covariance matrix corresponding to the posterior estimate of the health state. The consistency of two sets of voltage characteristic coefficient vectors obtained by mutually orthogonal perturbation current excitation sequences is checked, and the cross-sequence consistency deviation is calculated. The impedance features are concatenated into a feature vector, and the distance between the vector and the reference distribution in the battery population health model library is measured to obtain the model domain distance index. The distance measurement is calculated using Mahalanobis distance. The uncertainty index is obtained by weighting and fusing the spectral radius of the posterior covariance matrix, the cross-sequence consistency deviation, and the model domain distance index, wherein each weight is calculated based on the similarity between the battery and the reference battery model.

2. The multi-objective charging strategy optimization method considering battery health degradation according to claim 1, characterized in that, The real-time status parameters include the open-circuit voltage, terminal voltage, and temperature of the battery after it is connected to the battery swapping cabinet. Based on these real-time status parameters, the battery status is mapped to a preset battery family health model library to obtain an initial health status estimate, including: The similarity between the terminal voltage and the feature vector of the reference battery in the battery family health model library is calculated, and the reference battery model of the battery is determined based on the result of the similarity calculation. The open-circuit voltage and temperature are matched with the reference open-circuit voltage-temperature curves corresponding to the reference battery models in the battery family health model library to determine the initial state of charge correction value of the battery. Based on the reference battery model and the initial state of charge correction value, the initial health state estimate is output, wherein the initial health state estimate includes at least one of the capacity retention rate and the internal resistance growth rate.

3. The multi-objective charging strategy optimization method considering battery health degradation according to claim 1, characterized in that, By calling the historical charging data of the reference battery model, a perturbation current excitation sequence is generated, including: The historical charging data is decomposed in the frequency domain to extract frequency components and generate candidate perturbation sequences. Under the condition of satisfying the zero net charge constraint, the candidate perturbation sequence is optimized to obtain a first perturbation current and a second perturbation current that are orthogonal to each other. The first and second disturbance currents are corrected by amplitude scaling and phase adjustment to generate the disturbance current excitation sequence.

4. The multi-objective charging strategy optimization method considering battery health degradation according to claim 1, characterized in that, The original voltage sequence is orthogonally demultiplexed and projected onto a matched filter array to obtain the corresponding voltage characteristic coefficient vector, including: The original voltage sequence is decomposed in a preset feature space to obtain multiple independent response components. The projection matrix and projection operator are calculated using historical charging data. By combining the projection matrix and the projection operator, the response components are mapped to a voltage characteristic coefficient vector.

5. The multi-objective charging strategy optimization method considering battery health degradation according to claim 1, characterized in that, The objective optimization function, which combines charging time, battery health degradation cost, electricity price cost, and thermal management cost, includes: The characteristic factors corresponding to charging time, battery health degradation cost, electricity price cost, and thermal management cost are respectively used as basic objective items; The basic objective terms corresponding to charging time and battery health degradation cost are defined as a first function based on capacity retention rate and internal resistance growth rate, and the confidence interval boundary of the posterior estimate of the health state is used as the penalty factor of the first function. The basic objective term corresponding to the electricity price cost is expanded into a second function according to the time series, and the penalty factor of the second function is adjusted by the electricity price in different time periods; The basic target item corresponding to the thermal management cost is associated with the preset battery temperature prediction trajectory in the reference battery model of the battery to obtain the third function, and the allowable temperature rise range is adjusted by the uncertainty index to obtain the penalty factor of the third function. Calculate the objective optimization function based on the first function, the second function, and the third function, as well as the penalty factor corresponding to each function.

6. The multi-objective charging strategy optimization method considering battery health degradation according to claim 1, characterized in that, The generation of a charging strategy adapted to the battery includes: The charging process is divided into multiple stages based on the state of charge range, and a risk envelope function is constructed for each stage based on the uncertainty index. The risk envelope function is used to parametrically reshape the upper limit of the charging rate, the temperature rise range, and the approach penalty of the target state of charge for each stage. The probabilistic safety constraint is calculated based on the confidence boundary of the risk envelope function, wherein the probabilistic safety constraint represents the probability that the battery charging state falls into the safety region within the range of the uncertainty index. Under the condition that the probabilistic safety constraints are met at each stage, the uncertainty evolution trajectory in the charging process is predicted based on the posterior estimate of the health state, and the information gain that has the expected reduction effect on the uncertainty index at each stage is calculated, wherein the expected reduction effect represents the expected shrinkage of the uncertainty index after executing the corresponding charging rate and timing settings. The information gain is used as a positive adjustment factor to adjust the objective optimization function. Based on the adjusted objective optimization function, the charging method at each stage is calculated using an optimization algorithm, and the charging methods are summarized to obtain the charging strategy.

7. The multi-objective charging strategy optimization method considering battery health degradation according to claim 6, characterized in that, The information gains that are expected to reduce the uncertainty index at each stage of the calculation include: Under the aforementioned probabilistic safety constraints, simulations were performed on different combinations of charging rates and timing sequences to obtain the corresponding uncertainty contraction trajectories. The entropy value of the uncertain contraction trajectory is calculated, and the entropy reduction magnitude is used as a quantitative indicator of information gain.

8. The multi-objective charging strategy optimization method considering battery health degradation according to claim 1, characterized in that, The method further includes: Based on the charging strategy and the channel resource constraints of the corresponding charging channel of the battery swapping cabinet, group charging scheduling optimization is performed to obtain the overall charging strategy.

Images