CN115684970A

CN115684970A - Method and apparatus for providing a total trace of aging status

Info

Publication number: CN115684970A
Application number: CN202210897463.7A
Authority: CN
Inventors: C·西莫尼斯; S·辛德勒
Original assignee: Robert Bosch GmbH
Current assignee: Robert Bosch GmbH
Priority date: 2021-07-29
Filing date: 2022-07-28
Publication date: 2023-02-03
Also published as: DE102021208227A1

Abstract

Method and device for fusing a plurality of aging state trajectories of an energy store, wherein each aging state trajectory is formed by an aging state as a model value of a corresponding aging state model at a reference point in time, which corresponds to a time of use of the energy store since commissioning; comprises the following steps: determining model values of the aging states of the first aging state trajectory and the second aging state trajectory from the first aging state model and the second aging state model, respectively, wherein a confidence value is assigned to each model value; fusing at least two model values of the aging state to fused model values according to time intervals between reference time points thereof, respectively, according to the confidence values of the fused model values, wherein the fused model values are assigned fused confidence values, which are derived from the confidence values or from probability density functions of the fused model values; an aggregate trajectory is created from all fused and non-fused model values of the aging state trajectory and their confidence values.

Description

Method and apparatus for providing a total trace of aging status

Technical Field

The invention relates to an electrical device, in particular an electrically drivable motor vehicle, which is operated independently of an electrical network and has an electrical energy store, and also to measures for determining a change in the aging state on the basis of various methods for determining the aging state.

Background

The energy supply of electrical devices and machines (e.g., electrically drivable motor vehicles) which operate independently of the electrical grid is carried out by means of an electrical energy store, which is usually a device battery or a vehicle battery. These electrical energy stores provide electrical energy to operate the apparatus. However, energy converters, such as fuel cell systems, including hydrogen tanks, may also be considered as electrical energy storages.

The electrical energy store or the energy converter degrades over its service life and depending on its load or use. This so-called aging leads to a continuous decrease in maximum performance and storage capacity. The state of health (SOH) corresponds to a measure for describing the aging of the energy storage. Conventionally, the aging state of a new energy storage is 100%, which decreases significantly during its service life. The measure of the aging of the energy storage (change in the aging state over time) depends on the individual load of the energy storage, i.e. in the vehicle battery of the motor vehicle on the usage behavior of the driver, on the external environmental conditions and on the vehicle battery type.

Determination and prediction of the state of aging of an energy storage (e.g., lithium ion battery) to determine remaining value and to make predictive diagnostics is one of the fundamental challenges of current energy storage management and monitoring systems. Basically, a distinction can be made between physical methods, which reflect past or future aging behavior of the energy storage by a causal physical description of the underlying aging mechanisms, and data-based methods, which use previous observations of the aging state for prediction.

In order to describe the aging state of the energy store by a combination of the above-described methods (physical method, data-based method), the aging state progression determined or predicted by the different methods must be combined to form a common aging state trajectory. The challenge in this case is that these methods are usually executed asynchronously in practice, so that the determined or predicted aging state of the energy store may be associated with different points in time. Furthermore, the model values of the respective methods may have different confidence levels, for example when values determined by physical methods are given less uncertainty than values determined by data-based methods.

Disclosure of Invention

According to the invention, a method for determining a total trajectory of the aging states from a plurality of aging state trajectories of the electrical energy store according to claim 1 and a corresponding device according to the parallel independent claim are provided.

Further developments are specified in the dependent claims.

According to a first aspect, a computer-implemented method for fusing a plurality of aging state trajectories of an energy store of a technical installation is provided, wherein each aging state trajectory is formed from model values of the aging state at a respective reference time point, which corresponds to the time of use of the energy store since the energy store was put into operation; comprises the following steps:

-determining model values of the aging states of the first aging state trajectory and the second aging state trajectory from the first aging state model and the second aging state model, respectively, wherein a confidence value is assigned to each model value;

-fusing at least two model values of the aging state to fused model values, respectively, according to their time intervals between reference time points, according to their confidence values, wherein the fused model values are assigned fused confidence values, which are derived from the confidence values or from a probability density function of the fused model values;

-creating an overall trajectory from all fused and non-fused model values of the two aging state trajectories and their confidence values.

Energy storage in the sense of this description includes device batteries, energy converter systems having an electrochemical energy converter with an energy carrier storage, for example fuel cell systems having a fuel cell and an energy carrier storage.

The state of aging of the electrical energy store, in particular of the battery of the device, is usually not measured directly. This would require a series of sensors inside the energy storage, which would make the manufacture of such an energy storage expensive and complicated and would increase the space requirement. Furthermore, no measurement method is available on the market which is suitable for everyday use for determining the state of ageing directly in an energy store. The current aging state of the electrical energy store is therefore usually determined by means of an aging model in a control device separate from the energy store. These aging state models may be inaccurate in general conditions or in specific conditions and may have model deviations of up to more than 5%.

The State of aging (SOH) is a key variable in the case of a device battery that accounts for remaining battery capacity or remaining battery charge. The state of aging is a measure of the aging of the device battery. In the case of a device battery or a battery module or a battery cell, the state of aging can be described as a Capacity Retention Rate (SOH-C). The capacity retention rate SOH-C is illustrated as the ratio of the measured instantaneous capacity to the initial capacity of a fully charged battery. Alternatively, the aging state may be described as an increase in internal resistance (SOH-R) relative to the internal resistance at the beginning of the useful life of the device battery. The relative change in internal resistance SOH-R increases as the battery ages.

The aging state of the electrical energy store can be determined periodically or on an event-by-event basis on the basis of operating variables of the energy store. For this purpose, different aging state models can be used, which determine the aging state by means of different methods and can also provide a confidence of the model value of the aging state in a suitable manner. Since the models are formed differently, model values of aging states cannot be easily combined into a trajectory because the model values have separate confidence and reference time points.

Common filter-based methods are known in the art, such as determining a sliding weighted average. However, most of these methods have a group latency that results in an undesirable time offset of the fused trajectory relative to the original model values.

Merging using classical filter-based schemes also poses the problem of selecting a suitable filter kernel or a suitable filter width, especially in the case of non-equidistant sampling model values, and the usage must always be re-evaluated depending on the data conditions.

The goal of the above method is to merge the model values of the aging states of multiple aging state trajectories into an overall trajectory, thereby completely preserving the original reference time points of the model values. Furthermore, a robust method for reliably fusing model values of the aging state independently of the underlying aging state model should be able to be implemented in this way. Furthermore, the aging state trajectories should be merged into a total trajectory without using parameterization rules.

To this end, the above method provides for fusing model values from different aging state models to determine an overall trajectory of the aging state of the electrical energy store. The fusion of the model values is performed according to the respective confidence degrees of the model values. In this way problems of group latency can be avoided. Furthermore, aging states with varying availability and accuracy due to the use of electrical energy stores may be used to determine the aging state trajectory. For example, in the case of a traction battery as an electrical energy store, the accuracy of a basic model for evaluating charge inflow and outflow at low vehicle usage rates is very poor, and model values are rarely determined because the basic state-of-charge stroke (ladezstandshub) is too small.

The above-described method thus makes it possible to combine aging state trajectories made up of individual model values of the aging state determined by different aging state models and at different points in time and to take into account the different confidence values of the model values when combining. The determined trajectories are then formed from the reference time points of the base model values of the aging state trajectories, so that the total trajectory provided generally has a greater number of reference time points than is the case for the individual aging state trajectories.

If the aging state model value of the first aging state trajectory and the aging state model value of the second aging state trajectory have a time interval between their reference time points that is smaller than a predefined first time interval, the aging state model value of the first aging state trajectory and the aging state model value of the second aging state trajectory can be fused. In particular, the aging state model value may be fused in correspondence with a confidence value assigned to the aging state model value according to bayesian theorem.

Furthermore, an aging state model value of one of the first and second aging state trajectories and two aging state model values of the corresponding other one of the first and second aging state trajectories may be fused, in that the reference time points of the two model values of the other aging state trajectory comprise the reference time points of the model values of the one aging state trajectory and are not temporally distant from each other by more than a predefined second time interval, wherein an interpolated value is determined, in particular by interpolating the two model values of the other aging state trajectory to the reference time points of the model values of the one aging state trajectory and an interpolated confidence value at the reference time points is determined, wherein the model values of the one aging state trajectory are fused with the interpolated values, in particular by means of bayesian theorem depending on the confidence value of the model value of the one aging state trajectory and the interpolated confidence value.

Furthermore, it can be provided that the model value of one of the first and second aging state trajectories for fusing with the other model value is no longer fused with the model value.

The above method allows rule-based merging of model values based on different aging state models or aging state trajectories with different data or model sources based on defined selection rules and subsequent formation of confidence weighted averages of the selected model value pairs. The method can be general and independent of the type of energy storage, so that it does not matter in which way the model values of the aging state trajectory are determined, as long as the confidence values belonging to the model values are normalized to one another and are associated with a common time axis.

In principle, various types of aging state models are known which can be used to determine the aging state of the electrical energy store. The aging state of the electrical energy store can thus be determined, for example, by means of a physical aging state model or a hybrid aging state model with a data-based correction model.

The aging model for determining the aging state of the electrical energy store may be provided in the form of a hybrid aging state model, i.e. a combination of a physical aging model and a data-based model. In the case of a hybrid model, the physical aging state can be determined by means of a physical or electrochemical aging model and a correction value derived from the data-based correction model can be applied to the physical aging state, in particular by addition or multiplication. The physical aging model is based on electrochemical model equations characterizing the electrochemical states of the nonlinear differential equations, the electrochemical model equations are continuously calculated and mapped to physical aging states for output, as SOH-C and/or SOH-R. These calculations may typically be performed in the cloud, for example once per week.

Furthermore, the correction model of the aging state model based on the mixed data may be constructed with a probabilistic or artificial intelligence based regression model, in particular a gaussian process model, and may be trained to correct the aging state obtained by the physical aging model. Therefore, for this purpose, a data-based aging state correction model for correcting the SOH-C and/or at least one further data-based aging state correction model for correcting the SOH-R are present. Possible alternatives to gaussian processes are further supervised learning methods, such as those based on random forest models, adaBoost models, support vector machines or bayesian neural networks.

Usually, with the aid of such an aging state model, model values, for example, of the aging state are determined at regular evaluation intervals on the basis of the course of change of the operating variables of the electrical energy store.

The surrogate model for determining the state of aging of the battery is the so-called base model. The aging state value can be determined here on the basis of the Capacity Retention Rate (SOH — C) by means of the coulomb counting method. For this purpose, it is recognized that a charging process is being carried out on the basis of the time profile of the operating variable. The charging process can be recognized, for example, when a constant current is supplied from a state in which the battery is completely discharged (in the case of a battery this can be recognized in the case of a discharge end voltage having been reached). The charging process can thus be determined on the basis of a constant current flowing into the vehicle battery. When the charging process proceeds to full charge, the total amount of charge delivered to the vehicle battery can be determined by integrating the current flowing into the battery. By comparison with the nominal charge capacity of the battery, the maximum charge can be assigned to the aging state value. The respective measurements of partial charge with a particular charge transport and cell voltage before and after partial charge can also be evaluated to determine an aging state value based on the capacity retention. Furthermore, coulomb counting can also be carried out during the discharge process, for example during the driving cycle, by determining the quantity of charge flowing out and evaluating the cell voltage before and after partial charging.

The determination of the aging state by means of the base model is event-triggered, so that the model value of the aging state can generally only be used at irregular points in time.

For the aging state model values from the two aging state models, in each case one can use the confidences which can be derived as confidences of the data-based correction model in the case of a hybrid aging state model and which are determined in particular as a function of sensor tolerances in the case of a base model.

The energy storage may be used for operating devices such as motor vehicles, electric bicycles, aircraft (in particular unmanned aerial vehicles), machine tools, consumer electronics devices such as mobile phones, autonomous robots and/or domestic appliances.

According to a further embodiment, a total trajectory can be determined for a large number of energy stores, wherein a data-based residual model is trained from a large number of training data sets, wherein the training data sets are assigned differences between model values of the total trajectory and model values of a specific aging state trajectory of the first and second aging state trajectories with respect to the observation variable and/or the measurement variable at the same reference point in time, wherein the residual model is used device-specifically for generating correction values to apply to the model values of the specific aging state model.

In particular, it can be provided that the residual model is used device-specifically only if the total deviation between the total trajectory and the aging state trajectory determined by the particular aging state model is improved by applying the particular residual model to the particular aging state model.

Furthermore, at least one of the first and second aging state models may be calibrated in a relaxed state after the stationary phase.

It can be provided that at least one aging state of the first or second aging state trajectory is determined by fitting the electrochemical state model with updated equilibrium parameters or by means of a capacity measurement by coulomb counting.

Furthermore, at least one aging state of the first or second aging state trajectory can be determined by means of a data-based aging state model comprising a gaussian process model.

According to one embodiment, the residual model may be verified by residual analysis using the Shapiro-Wilk hypothesis test, where the newly derived residual of the aging state of the total trajectory is normally distributed. This is used to verify the normal distribution.

Additionally, the residual model may be verified by residual analysis using fourier transform or wavelet transform to prove that the newly derived residual is not affected by seasonal effects. This is used to verify independence from seasonality.

Provision may be made for the first aging state model and/or the second aging state model to be corrected as a function of the overall trajectory and for the aging state to be calculated in the technical installation using the corrected first aging state model and/or second aging state model together with the residual model. For this reason, it is preferable to use an aging state model having a wider confidence band and higher sensitivity (e.g., higher sensitivity to temperature).

According to a further aspect, a device for fusing a plurality of aging state trajectories of an energy store of a technical device can be provided, wherein each aging state trajectory is formed by model values of an aging state at a respective reference point in time, which corresponds to a time of use of the energy store since commissioning; wherein the apparatus is configured to:

Drawings

Embodiments are explained in more detail below based on the drawings.

FIG. 1 shows a schematic diagram of a system for providing driver-specific and vehicle-specific operating variables in a central unit for determining the state of aging of a vehicle battery;

FIG. 2 shows a functional block diagram of a hybrid aging state model;

fig. 3 shows a flow chart for explaining a method of fusing the aging state trajectories of the vehicle batteries in the central unit;

FIG. 4 shows a schematic of the aging state trajectories of two different aging state models along with their respective confidence values, shown as error bars;

FIG. 5 shows a schematic of an aging state trace and the resulting total trace; and

fig. 6 is a diagram showing a variation process of a residual model trained by fleet data.

Detailed Description

The method according to the invention is described below on the basis of a vehicle battery as an electrical energy store in a large number of motor vehicles as devices of the same type. The method makes it possible to merge the aging states obtained by means of different aging state models into a total trajectory in the central unit, which can be used in the central unit for aging calculations, aging predictions, remaining service life calculations, etc.

The above examples represent a large number of stationary or mobile devices with a grid-independent energy supply, such as vehicles (electric vehicles, electric bicycles, etc.), facilities, machine tools, household appliances, internet of things devices, etc., which are connected via a corresponding communication connection (e.g., LAN, internet) to a central unit (cloud) outside the device.

Fig. 1 shows a system 1 for collecting and processing fleet data in a central unit 2 to create and run and evaluate aging status models. These aging state models are used to determine the aging state of an electrical energy store, for example a vehicle battery or a fuel cell in a motor vehicle, at a specific point in time. Fig. 1 shows a vehicle fleet 3 with a plurality of motor vehicles 4.

One of the motor vehicles 4 is shown in more detail in fig. 1. The motor vehicle 4 has a vehicle battery 41 as a rechargeable electrical energy store, an electric drive motor 42 and a control unit 43, respectively. The control unit 43 is connected to a communication module 44, which communication module 44 is adapted to transmitting data between the respective motor vehicle 4 and the central unit 2 (the so-called cloud).

The motor vehicle 4 transmits to the central unit 2 operating variables F which at least account for variables which influence the state of ageing of the vehicle battery 41. In the case of a vehicle battery, the operating variables F may describe the instantaneous battery current, instantaneous battery voltage, instantaneous battery temperature, and instantaneous State of Charge (SOC) at the pack level, module level, and/or cell level. The operating variable F is detected in a time grid of 0.1Hz to 100Hz and can be transmitted periodically to the central unit 2 in uncompressed and/or compressed form. For example, using a compression algorithm to minimize the data traffic to the central unit 2, the time series may be transmitted to the central unit 2 block by block at intervals of 10 minutes to several hours.

The central unit 2 has a data processing unit 21 which can perform the methods described below, and a database 22 for storing data points, model parameters, states, etc.

In the central unit 2, aging state models can be implemented which are based on different algorithms and are designed to determine the aging state of the vehicle battery 41 at specific points in time in each case.

The State of aging (SOH) is a key variable for describing the remaining battery capacity or remaining battery charge. The state of aging is a measure of the aging of the vehicle battery or the battery module or the battery cell, and can be described as a Capacity Retention Rate (SOH-C) or an internal resistance increase (SOH-R). The capacity retention rate SOH-C is illustrated as the ratio of the measured instantaneous capacity to the initial capacity of a fully charged battery. The relative change in internal resistance SOH-R increases as the battery ages.

For example, a first aging state model can be implemented in the central unit 2, which is based on data as a hybrid model part. The first aging state model can be used periodically, i.e. for example after the end of the respective evaluation duration, to determine the instantaneous aging state of the vehicle battery 41 of interest of the assigned fleet on the basis of the course of the time variation of the operating variables (in each case since the respective vehicle battery was put into operation) and the operating characteristics determined therefrom. In other words, the aging state of the vehicle battery 41 in question can be determined on the basis of the course of change of the operating variable of one of the vehicle batteries 41 of the motor vehicles 4 of the assigned fleet 3 and the operating characteristics resulting from said course of change of the operating variable.

Fig. 2 shows a schematic diagram of the functional structure of an embodiment of the data-based aging state model 9, which is constructed in a hybrid manner. The first aging state model 9 includes the physical aging model 5 and the correction model 6.

The physical aging model 5 is a nonlinear mathematical model based on differential equations. Evaluating the physical aging model of the aging state model with the operating variable progression, in particular the operating variable progression since the beginning of the service life of the system battery, leads to the following internal state of the system of equations of the physical differential equation, which internal state corresponds to the physical internal state of the system battery. Since the physical aging model is based on physical and electrochemical laws, the model parameters of the physical aging model are variables that describe physical properties.

The operating variable F therefore enters directly into the physical aging state model 5, which is preferably designed as an electrochemical model and describes the corresponding internal electrochemical states, such as layer thickness (for example SEI thickness), change in cyclizable lithium due to anode/cathode side reactions, rapid consumption of electrolyte, slow consumption of electrolyte, loss of active material in the anode, loss of active material in the cathode, etc., by means of nonlinear differential equations and multidimensional state vectors.

The physical aging model 5 thus corresponds to an electrochemical model of the cell and the cell chemistry. The model determines the internal physical cell state from the operating variables F in order to provide an at least one-dimensional, physically-based state of aging SOHph in the form of the above-mentioned electrochemical states, which are mapped linearly or nonlinearly to capacity retention rates (SOH-C) and/or internal resistance increase rates (SOH-R) to be provided as states of aging (SOH-C and SOH-R).

However, the model values of the physical state of aging SOHph provided by the electrochemical model are inaccurate under certain conditions, and therefore provision is made for correcting these model values using the correcting variable k. The correcting variables k are provided by a data-based correction model 6 which is trained by means of a training data set of vehicles 4 from the platoon 3 and/or by means of laboratory data.

The correction model 6 receives on the input side an operating characteristic M which is determined from the course of the change in the operating variable F and which may also comprise one or more internal electrochemical states of the system of differential equations of the physical model. Further, the correction model 6 may obtain the physical aging state SOHph obtained from the physical aging model 5 on the input side. The operating characteristics M of the current evaluation period are generated in the characteristic extraction block 8 on the basis of the time series of operating variables F. Furthermore, the operating characteristics M comprise the internal states of the state vector from the electrochemical physical aging model and advantageously comprise the physical aging state SOHph.

The operating characteristics M relating to the evaluation period can be generated already from the operating variables F in the central unit 2 for each fleet 3 or in other embodiments in the respective motor vehicle 4. To determine the state of aging, the evaluation period may be several hours (e.g., 6 hours) to several weeks (e.g., one month). The evaluation period is usually one week.

The operating characteristics M may comprise, for example, characteristics relating to an evaluation period and/or cumulative characteristics and/or statistical variables determined over the service life so far. In particular, the operating characteristics may include, for example: electrochemical states such as SEI layer thickness, changes in cyclable lithium due to anode/cathode side reactions, rapid uptake of electrolyte solvent, slow uptake of electrolyte solvent, lithium deposition, loss of anode active material and loss of cathode active material, information about impedance or internal resistance. Furthermore, histogram features are used as features, e.g. statistical variables from the histogram, such as mean, median, maximum, minimum, standard deviation, etc. The histogram may illustrate, for example, temperature versus state of charge, charge current versus temperature, and discharge current versus temperature. In this case, all histograms are multiplied by a penalty function to pre-evaluate multi-dimensional combinations of historical states with respect to the degenerative damage values and map the multi-dimensional combinations to one-dimensional features, respectively. Furthermore, the histogram and its penalty function may comprise and introduce information about the battery temperature distribution over the state of charge, the charging current distribution over the temperature and/or the discharging current distribution over the temperature. Other operating characteristics M may be current throughput in amp-hours, accumulated total charge (Ah), average capacity increase during charging (especially for charging with a charge increase above a threshold fraction (e.g. 20%) of total battery capacity), charging capacity, and a differentiated capacity extremum (e.g. maximum) during charging with a sufficiently large measured stroke of the state of charge (smoothly varying course of dQ/dU: charge change divided by battery voltage change) or accumulated mileage. These variables are preferably scaled in such a way that they characterize the real usage behavior as best as possible and are normalized in the feature space. The operating characteristic M may be used in whole or in part in the methods described below.

In order to determine the corrected state of aging SOH to be output, the outputs SOHph, k of the physical aging model 5 and the data-based correction model 6, which is preferably designed as a gaussian process model, are applied to one another. In particular, these outputs may be added or multiplied (not shown) in a summation block 7 to obtain the modeled state of aging SOH to be output at the current evaluation period. In the case of addition, the confidence of the gaussian process can still be used as the confidence of the corrected aging value SOH to be output of the mixture model. Thus, the confidence of the gaussian process model characterizes the modeling uncertainty of the mapping of the operating feature points or their principal components (when using PCA) to the aging state.

PCA (Principal Components Analysis) can be used, if necessary, to run scaling and dimensionality reduction of the features to correspondingly reduce the redundant linear correlation information in the feature space prior to training the correction model (unsupervised). This improves data coverage for each dimension of the feature space and counteracts "cursing of high dimensions". Alternatively, kernel PCA may be used to map out non-linear effects while reducing the complexity of the data. Before and in particular after the dimension reduction, the entire running feature space (or principal component space) is normalized, e.g., using min/max scaling or Z-transform.

Thus, for an energy storage device having at least one electrochemical unit, such as a battery cell, the aging state can be calculated and the aging state predicted. The method may also be applied to the entire system of energy storages via rule-based and/or data-based mapping. Therefore, taking a battery as an example, the aging prediction can be applied not only to the cell level, but also directly to the module level and the battery pack level. Additionally or alternatively, the weakest cell in the energy storage system may also be considered a limiting component of the battery pack.

The second aging state model can be provided as a so-called base model, which can be executed in vehicles and in the cloud. For this purpose, the process of the operating variable variation is monitored and, in particular, certain operating phases are evaluated, for example, a long standstill phase in the relaxation range and a repeatable charging phase. If it is determined during the charging state or during constant driving with a constant or almost constant battery current that the charging state stroke is above a predefined threshold value, the charging state stroke can be evaluated to identify the aging state.

Here, the aging state value can be determined based on the capacity retention rate (SOH-C) by means of the Coulomb counting method. For this purpose, it is recognized that a charging process is being carried out on the basis of the time profile of the operating variable. The charging process can be recognized, for example, when a constant current is supplied from a state in which the battery is completely discharged (in the case of a battery this can be recognized in the case of a discharge end voltage having been reached). The charging process can therefore be determined on the basis of a constant current flowing into the vehicle battery. When the charging process proceeds to full charge, the total amount of charge delivered to the vehicle battery can be determined by integrating the current flowing into the battery. By comparison with the nominal charge capacity of the battery, the maximum charge amount can be assigned to the aging state value. The respective measurements of cell voltage before and after partial charging and partial charging with a specific charge delivery can also be evaluated to determine the aging state value based on the capacity retention rate. In addition, coulomb counting can also be carried out during the discharge, for example during the driving cycle, by determining the quantity of charge flowing out and evaluating the cell voltage before and after partial charging.

Furthermore, during the resting phase, i.e. in the relaxed battery state, an electrochemical performance model can be fitted in order to determine both the equilibrium parameters and the state of aging with the aid of the collected resting voltage or OCV information. Observations through the electrochemical performance model or coulomb counting algorithm are then smoothed and converted to trajectory calculations with confidence to suppress numerical noise. The confidence is determined from the dispersion of the individual SOH observations, for example by means of the standard deviation. Thus, the base model may provide an aging state trace with confidence since commissioning, where it may be based on observations of electrochemistry, model-based, and measurement techniques.

Since the modeling conditions are only sporadically present, depending on the usage behavior, the model values for the second aging state model are available at irregular points in time, or the last trace calculation performed is some time ago and therefore not daily.

From the model values of the aging state model described above or of the alternative or additional aging state model, an aging state trajectory is derived which has the model values of the aging state at a predetermined point in time (calculated from the commissioning time point).

FIG. 3 shows a flow diagram illustrating a method of fusing two aging state trajectories, each having a confidence value, into an overall trajectory having model values of the aging states. The method is performed in the central unit 2, and the operating variables corresponding to a large number of vehicle batteries 41 are detected in the central unit 2.

By evaluating the operating variables F, in step S1, a first and a second aging state trajectory are provided for each vehicle, which can be determined by different aging state models (for example, the method described above).

Fig. 4 shows a diagram which illustrates the time profile of the aging state of the individual vehicle batteries 41 over the service life which has elapsed so far in relation to their respective reference time points. The reference time point represents the duration elapsed since the vehicle battery 41 was put into operation. The first and second aging state trajectories T1, T2 are represented by dashed lines, wherein their respective model values are represented by confidence intervals/confidence values (via error bars), respectively. The aging state trajectories T1, T2 are obtained by connecting model values determined for the aging state, which are determined by one of the aging state models.

To merge the aging state trajectories T1, T2 into a total trajectory, first a directly merged model value is selected in step S2. For this purpose, the model values of the first aging state trajectory T1 and the model values of the second aging state trajectory T2 are combined into a model value pair for direct fusion, the reference points in time of which are at most a predetermined first time interval apart from one another. The first time interval is a configurable parameter and may be set corresponding to existing data conditions.

For example, the fifth model value M5T1 of the first aging state trajectory T1 determined at the reference time point day 70 and the fifth model value M5T2 of the second aging state trajectory T2 having the reference time point day 65 are combined, corresponding to the aging state trajectories of fig. 4. Here, the first time interval is exemplarily set to six days. Thus, both the model value M5T1 from the first aging state trajectory T1 and the model value M5T2 from the second aging state trajectory T2 may be selected for direct fusion (represented by FD) as a model-value pair. In this way, the model value pairs for direct fusion are determined from the two aging state trajectories T1, T2 and stored temporarily.

This determination is made by checking for all model values of the first aging state trajectory whether there is a corresponding model value for a time interval of the second aging state trajectory that is less than the first time interval. If a model value pair is determined, the corresponding model value of the second aging state trajectory T2 is excluded from further determination of the model value pair.

In step S3, the aging state trajectory is checked for the model values for indirect fusion. Only model values that were not previously selected as model value pairs for direct fusion are considered here. For indirect fusion, it is checked whether the model values of the first aging state trajectory T1 (determined by the aging state and its reference point in time) are surrounded by two model values of the second aging state trajectory, wherein the model values of the second aging state trajectory are maximally spaced apart from each other by a second time interval. For this reason, it does not matter whether the further model value of the first aging state trajectory lies within the second time interval between the two model values of the second aging state trajectory.

The second time interval, like the first time interval, is a configurable parameter and may be set corresponding to existing data conditions. In particular, the second time interval may be selected to be greater than the first time interval.

If it is determined that the model value of the first aging state trajectory T1 is surrounded by two model values of the second aging state trajectory that are less than the second time interval from each other, the corresponding model value of the first aging state trajectory and the two second model values are combined into a model value triplet. The combined model value is not considered anymore.

The determination is carried out by checking for all model values of the first aging state trajectory T1 whether there is a corresponding model value of the second aging state trajectory T2 for a time interval which is smaller than the second time interval, which surrounds the reference time point of the first aging state trajectory T1 with respect to its reference time point. If such a triplet of model values is determined, the corresponding model value of the first aging state trajectory T1 and the corresponding model value of the second aging state trajectory T2 are excluded from further determination of the triplet of model values.

For example, the model value M2T1 of the first aging state trajectory T1 is surrounded by the second and third model values M2T2 and M3T2 of the second aging state trajectory T2 with respect to the reference time point. The second and third model values are separated from each other by a predetermined second time interval of at most 20 days. Thus, the second model value M2T1 of the first aging state trajectory T1 qualifies as a model value triplet (denoted FI) for indirect fusion with the corresponding model values M2T2, M3T2 of the second aging state trajectory T2.

The method is performed separately for model values of the first and second aging state trajectories, wherein the model values combined as a triplet of model values are excluded from further evaluation, respectively.

For direct fusion, it is assumed that the time intervals are negligible due to slow time-varying aging behavior.

As a result of the above steps S2 and S3, the pair of model values as model values for direct fusion, the triplet of model values as model values for indirect fusion, and the remaining model values that are not selected for either direct or indirect fusion are fused to each other.

To this end, in step S4, the model value pairs are fused according to bayes' theorem, the fusion being performed exemplarily on two normal distributions to be fused. SOH from first and second aged state traces, respectively ₁ 、SOH ₂ The weighted average of the model values of (A) yields the SOH value of the integrated total trajectory _fused ：

。

Weight factor w ₁ And w ₂ SOH by assigning model value pairs ₁ 、SOH ₂ The confidence values or model value variances σ 1, σ 2 of the respective model values (σ 1 is the model value variance of the model value from the model value pair of the first aging-state trajectory, and σ 2 is the model value variance of the model value from the model value pair of the second aging-state trajectory):

。

the confidence values of the individual model values may, for example, correspond to 90% confidence values, which are each determined by 1.645 times the standard deviation of the model values of the aging state trajectory.

In order to apply the above formula, the confidence values must be converted into respective model value variances according to their definition. In the case of direct fusion, the above formula for calculating the weighting factors is applied directly to the model value pairs SOH1, SOH2 determined from the two aging state trajectories to determine the aging state SOH of the fused model values _fused . This aging state of the fused model values is assigned to the two reference points in time of the model values of the model value pair.

In the case of a model-value triplet for indirect fusion, the two model values of the aging state on the same aging state trajectory are respectively interpolated, preferably linearly, between the two model values segment by segment, and the interpolated value at the reference point in time of the model value to be fused of the respective other aging state trajectory is determined. The model values and the interpolated values are then evaluated according to the above formula, wherein as confidence values of the interpolated values, the confidence values of the considered model values of the second aging state trajectory are also determined by (linear) interpolation to the reference time point.

In a subsequent step S5, SOH is derived from the corresponding fused model values _fused To determine an associated confidence value. This is done by re-assuming bayes' theorem, which yields the standard deviation of the fused model values:

the standard deviation of the fusion still has to be multiplied by a factor k, e.g. k =1.645, assuming a confidence of 90%, according to the definition of the confidence value of the original model value. The same calculation is performed for all fused model values.

Model values that are not selected for either direct or indirect fusion remain unchanged, including the associated confidence values, i.e., in this case, no bayesian theorem is applied.

Subsequently in step S6, the directly and indirectly fused model values and the unfused model values are combined into an overall trajectory. Here, the model values are sorted in ascending temporal order with respect to their reference time points, if necessary. A course of variation of the total trajectory Tges is obtained as described in figure 5.

If the above-described method is applied to more than two aging-state trajectories, then an iterative fusion can be carried out for each two of these aging-state trajectories, with the fused total trajectory being fused in each case with a further (third) aging-state trajectory which is likewise to be observed.

In a subsequent step S7, a residual analysis across vehicles may be performed. For this purpose, each vehicle uses the fused total trajectory Tges determined individually for the vehicle battery 41 and calculates the residual error for a sensitive observation or measurement variable, which can be assigned to one of the aging state trajectories, the aging state trajectory determined by means of the base model.

For example, due to the temperature dependence of the state of charge estimator, the calculation of the aging state in coulomb counting may be sensitive to battery or ambient temperature. Thus, the residuals are now analyzed by all available vehicle batteries and, as shown in fig. 6, the analysis is performed based on plotting the residuals of each vehicle battery over the relevant temperature.

It can be seen that for a large number of vehicle batteries, there is a system dependence of the residual on the battery temperature, which has an effect on the observed and measured variables. The residual change process on the sensitive observed or measured variable can now be trained in a data-based residual model.

The residual analysis is performed by training a data-based residual model (regression model), in particular a gaussian process model or integrator method, on the basis of the difference between the model value of the total trajectory fused at a reference point in time with respect to a specific battery temperature and/or an alternative operating characteristic (for example aging state range, temperature range, battery voltage or current value) and the model value of the aging state trajectory to be observed.

The temperature dependence of the learned residuals can be seen in fig. 6 based on curve M. The parameterized residual model thus provides an estimated system deviation that can be interpreted, in particular, for a fleet of vehicles or for a plurality of energy stores across vehicles by means of sensitive observation or measurement variables. This sensitivity analysis is always performed directly on the raw measurements of the underlying model without signal-technical processing such as smoothing or trajectory effects. These observations are evaluated for fleet cross-vehicle and used for training and validation by residual models.

In step S8, a residual model is applied to the aging state model on which it is based, by applying, in particular adding, the model values of the residual model that depend on the observed or measured variable as correction values to the model values derived from the aging state model concerned.

Furthermore, it can be checked for each vehicle battery 41 whether the correction by means of the corresponding residual model contributes to the model improvement. For this purpose, a correction term is applied a posteriori to all observations from the aging state trajectory, preferably added to the correction term, which is determined and provided by sensitive observation or measurement variables (i.e. for example battery temperature or state of charge) and a residual model. The system observation error can thus be learned and calculated at each measurement time point, which is triggered, for example, on an event-by-event basis by a stationary phase or by a charging event of the user. The residual of the now corrected model is now evaluated for suitability to determine if all systematic effects have been successfully eliminated. To this end, the Shapiro-Wilk test can be used as a hypothesis test to verify that the residuals of the fleet are normally distributed.

If the vehicle-specific residual is improved after the fusion method is re-executed and verified by residual analysis, in particular by recalculating the residual and comparing it with the model value of the originally learned residual model, and the corrected model value of the total trajectory is reasonable from a domain point of view, the residual model can be continuously used for determining the total trajectory.

After successful verification, the residuals are approximately normally distributed and are not affected by structural or seasonal effects. Thus, residual analysis demonstrated that there was no longer system sensitivity across vehicles for the entire fleet.

For example, the rationality may be checked for the absence of seasonal temperature dependence. For this purpose, it is checked on a regular basis whether the degradation profile of the vehicle battery is typical for the profile of the aging state, for example by means of gradient monitoring or wavelet-or FFT-based oscillation evaluation. So that, for example, no seasonal component should be present during the aging process, for example oscillations in the annual cycle which indicate the dependence of a particular aging state on the temperature present at a certain point in time.

If it is ascertained, based on the data and also by the domain rules, that the behavior learned by a large number of vehicle batteries leads to additional values in the aging state modeling, the aging state model considered is used for vehicle-specific modeling of the aging state in the further course of the change and operation of the residual model. In the future, this will also improve the vehicle-specific fusion of model values from different aging state models, since less uncertainty can be expected when modeling the aging state.

The validated correction model can be used in a vehicle to calculate the aging state and can also be run (temporarily) locally without a cloud or internet connection.

Claims

1. A computer-implemented method for fusing a plurality of aging-state trajectories (T1, T2) of an energy store (41) of a technical device (4), wherein each aging-state trajectory (T1, T2) is formed by an aging state as a model value of a corresponding aging-state model at a respective reference point in time, which reference point in time corresponds to a time of use of the energy store (41) since commissioning; comprises the following steps:

-determining (S1) model values of the aging states of the first aging state trajectory (T1) and the second aging state trajectory (T2) from the first aging state model and the second aging state model, respectively, wherein each model value is assigned a confidence value;

-fusing (S2, S3, S4) at least two model values of an aging state to the fused model values, respectively, according to the time intervals between their reference time points, according to their confidence values, wherein a fused confidence value is assigned to the fused model values, the fused confidence value being derived from the confidence value or a probability density function of the fused model values;

-creating (S6) a total trajectory (Tges) from all fused and non-fused model values of the aging state trajectory (T1, T2) and their confidence values.

2. The method according to claim 1, wherein the aging state model values of the first aging state trajectory (T1) and the aging state model values of the second aging state trajectory (T2) are fused if they have a time interval between their reference time points which is smaller than a predefined first time interval.

3. The method of claim 2, wherein the aging state model values are fused in correspondence with confidence values assigned to the aging state model values according to bayesian theorem.

4. Method according to one of claims 1 to 3, wherein an aging state model value of one of the first and second aging state trajectories (T1, T2) and two aging state model values of the corresponding other of the first and second aging state trajectories (T1, T2) are fused in such a way that the reference time points of the two model values of the other aging state trajectory (T1, T2) comprise the reference time points of the model values of the one aging state trajectory (T1, T2) and are not temporally distant from one another by more than a predefined second time interval, wherein in particular an interpolated value is determined by interpolating the two model values of the other aging state trajectory (T1, T2) to the reference time points of the model values of the one aging state trajectory (T1, T2) and the interpolated confidence value at the reference time points is determined, wherein the interpolated model values of the one aging state trajectory (T1, T2) are fused to the interpolated confidence values, in particular by means of a Bayesian algorithm on the basis of the interpolated confidence values of the one aging state trajectory and the interpolated confidence value at the reference time points.

5. The method according to claim 4, wherein the model value of one of the first and second aging state trajectories (T1, T2) for fusing with other model values is no longer fused with a model value.

6. Method according to one of claims 1 to 5, wherein a total trajectory (Tges) is determined for a large number of energy stores (41), wherein a data-based residual model is trained from a large number of training data sets, wherein the training data sets are assigned a difference between model values of the total trajectory (Tges) and model values of a specific aging state trajectory of the first and second aging state trajectories (T1, T2) with respect to observation variables and/or measurement variables at the same reference point in time, wherein the residual model is used device-specifically for generating correction values to apply to the model values of the specific aging state model.

7. The method of claim 6, wherein the residual model is used device-specifically only if a total deviation between the total trajectory and an aging state trajectory determined by a particular aging state model is improved by applying the particular residual model for the particular aging state model.

8. The method of any of claims 1 to 7, wherein at least one of the first and second aging state models is calibrated in a relaxed state after a quiescent stage.

9. The method of any one of claims 1 to 8, wherein at least one aging state of the first aging state trajectory (T1) or the second aging state trajectory (T2) is determined by fitting an electrochemical state model with updated equilibrium parameters or by means of a capacity measurement by coulomb counting.

10. The method according to one of claims 1 to 9, wherein at least one aging state of the first aging state trajectory (T1) or the second aging state trajectory (T2) is determined by means of a data-based aging state model comprising a gaussian process model.

11. The method of claim 6 or 7, wherein the residual model is validated by residual analysis using the Shapiro-Wilk hypothesis test, wherein the newly derived residual of the aging state of the total trajectory is normally distributed.

12. The method of claim 6 or 7, wherein the residual model is validated by residual analysis using fourier or wavelet transforms to prove that the newly derived residual is not affected by seasonal effects.

13. Method according to one of claims 1 to 12, wherein the first and/or second aging state model is corrected from the total trajectory (Tges), in particular in a central unit (2), and the corrected first and/or second aging state model is used together with a residual model in the technical installation to calculate an aging state.

14. A device for fusing a plurality of aging state trajectories (T1, T2) of an energy store (41) of a technical device (4), wherein each aging state trajectory (T1, T2) is formed by an aging state as a model value of a corresponding aging state model at a respective reference point in time, which corresponds to a time of use of the energy store (41) since a commissioning; wherein the apparatus is configured to:

-determining model values of the aging states of the first aging state trajectory (T1) and the second aging state trajectory (T2) from the first aging state model and the second aging state model, respectively, wherein a confidence value is assigned to each model value;

-fusing at least two model values of an aging state to a fused model value according to their respective time intervals between reference time points according to their confidence values, wherein the fused model value is assigned a fused confidence value, which is derived from the confidence value or a probability density function of the fused model value;

-creating a total trajectory (Tges) from all fused and non-fused model values of the aging state trajectory (T1, T2) and their confidence values.

15. A computer program product comprising instructions which, when the program is executed by at least one data processing apparatus, cause the data processing apparatus to perform the steps of the method according to any one of claims 1 to 13.

16. A machine-readable storage medium comprising instructions which, when executed by at least one data processing apparatus, cause the data processing apparatus to perform the steps of the method according to any one of claims 1 to 13.