Method and apparatus for operating a system providing a predicted state of aging of an electrical energy storage

Abstract

The invention relates to a computer-implemented method for predicting a modeled aging state of an electrical energy storage, having at least one electrochemical unit, in particular a battery cell unit, having the following steps: - providing a data-based aging state model trained to assign a modeled aging state to the electrochemical energy storage as a function of a time course of operating variables of the energy storage, - providing a time course of the operating variables characterizing the operation of the electrical energy storage; - determining a current or predicted modeled aging state from the generated time course of the operating variables by means of the aging state model, wherein data gaps in the time course of the operating variables due to inactive phases are supplemented based on a time course of the temperature of the energy storage, the time course of the temperature being derived from at least one provided environmental condition.

Description

Technical Field

[0001] This invention relates to electrical equipment with energy storage devices that operate independently of the power grid, particularly electrically driven motor vehicles, especially electric or hybrid vehicles, and also to measures for determining the State of Health (SOH) of said energy storage devices. In addition to mobile energy storage devices, the invention also relates to stationary energy storage devices. Background Technology

[0002] The energy supply for electrical equipment and machinery that operate independently of the power grid (such as electrically powered motor vehicles) is provided by means of electrical storage devices (typically device battery packs or vehicle battery packs). These electrical storage devices provide electrical energy to operate the equipment. However, fuel cells can also be used as electrical storage devices.

[0003] The aging condition of an energy storage device decreases significantly during its service life, leading to a reduction in maximum storage capacity. The aging metric of an energy storage device depends on its individual load, specifically, in the case of a vehicle's battery pack, on driver usage behavior, external environmental conditions, and the type of vehicle battery pack.

[0004] During parking or rest periods, battery pack data is typically not measured or provided, or only within a very small range. This creates time gaps where stress factors affecting the device's battery pack are absent or poorly indicated. These gaps make it difficult to fully track and predict the aging process. However, predicting the aging process of an energy storage device is an important technical variable because it allows for the economic assessment of the device's remaining value. Summary of the Invention

[0005] According to the present invention, a method for predicting the aging state of an electric energy storage device according to claim 1 and an apparatus in an electrically operable device according to the co-independent claims are provided.

[0006] Other designs are described in the dependent claims.

[0007] According to a first aspect, a computer-implemented method is provided for predicting the aging state of an energy storage device having at least one electrochemical unit, particularly a battery pack cell, or further, through rule-based and / or data-based mapping throughout the system, comprising the following steps: - Provide a data-based aging state model, which is trained to assign a modeled aging state (SOH) to the electrochemical energy storage (41) based on the change process of the operating variable (F) of the energy storage (41). - Provide the time-varying process of the operating variable (F) characterizing the operation of the energy storage device; - Using the aging state model, the current or predicted modeled aging state (SOH) is determined based on the change process of the generated operating variable (F). The temperature change process of the energy storage device fills the data gaps in the time-varying process of operating variables caused by the inactive phase, and the temperature change process is derived from the provided environmental conditions.

[0008] The aging state of rechargeable energy storage devices, especially battery packs, is not typically measured directly. This would require a series of sensors near the energy storage device, making its manufacturing costly, complex, and space-consuming. Furthermore, there are no commercially available methods suitable for everyday use to determine the aging state of devices. Therefore, physical aging models within the device are usually used to determine the current aging state. These physical aging state models are inaccurate under certain conditions and typically have a model bias of more than 5%.

[0009] Furthermore, due to the inaccuracy of physical aging models, which can only describe the instantaneous aging state of energy storage devices, the prediction of aging state depends particularly on how the energy storage device is operated (e.g., in the case of a battery pack, on the magnitude and amount of charge inflow and outflow) and therefore on usage behavior and operating parameters, leading to highly inaccurate predictions that are not currently available.

[0010] When a device's battery pack functions as an energy storage unit, State of Health (SOH) is a key variable used to describe the remaining battery pack capacity or charge. SOH is a measure of the aging of the energy storage unit. In the case of a device's battery pack, battery pack module, or battery pack cell, SOH can be described as Capacity Retention Rate (SOH-C) or as the increase in internal resistance (SOH-R). Capacity Retention Rate (SOH-C) is expressed as the ratio of the measured instantaneous capacity to the initial capacity of a fully charged battery pack. The relative change in internal resistance (SOH-R) increases with increasing battery pack aging.

[0011] Providing user-specific and usage-specific predictions of the aging state of energy storage devices based on data-driven aging state models is a promising approach. For example, a data-driven aging state model can be implemented in a central unit (cloud) and trained using operating variables from a large number of energy storage devices communicatively connected to the central unit.

[0012] Aging state models used to determine the aging state of energy storage devices can be provided in the form of hybrid aging state models, i.e., a combination of physical aging models and data-based models. In the case of hybrid models, the physical aging state can be determined by means of physical or electrochemical aging models, and a correction value can be applied to this physical aging state, which is generated by a data-based correction model, specifically by addition or multiplication. Physical aging models are based on electrochemical model equations that characterize the electrochemical state and map them to physical aging states in order to output the electrochemical state, including SOH-C and SOH-R.

[0013] Furthermore, a hybrid data-based correction model for the aging state can be constructed using probabilistic or AI-based regression models, particularly Gaussian process models, and this correction model can be trained to correct for physical aging states. Therefore, there exist data-based correction models for aging states to correct SOH-C and / or additional data-based correction models for aging states to correct SOH-R. Possible alternatives to supervised learning include random forest models, AdaBoost models, support vector machines, or Bayesian neural networks.

[0014] Predicting the aging state is helpful if the remaining lifespan of the energy storage device needs to be determined. For this purpose, a data-based aging state model can be continuously queried in conjunction with pre-defined usage patterns of users of devices with energy storage devices. This requires the continuous generation of changes in artificial operating variables, which is necessary for the physical aging model. These operating variables are generated based on load variables.

[0015] This predictive capability advantageously utilizes trained aging state models and usage patterns to generate input-side time series of running variables, thereby enabling more accurate computation or prediction of aging states compared to gapped data (computation) or pure extrapolation methods (prediction).

[0016] The load variable is generated in the usage model. The usage model is constructed to continuously output the variation process of at least one load variable based on vehicle-specific usage patterns. This allows usage behavior parameterized by the usage pattern to be converted into a time series of at least one load variable. The usage pattern can therefore describe the load type of the energy storage device by means of the load variable. The usage pattern can exist as a model parameter of the usage pattern model and / or can be provided to the usage pattern model from a database.

[0017] When the battery pack is used as an energy storage device, the load variable describes at least one current load. When the battery pack is used as an energy storage device, additional load variables may describe temperature load, load time frequency, and / or periodic load (periodicity).

[0018] The usage model can also provide inactive phases and data reconstructions for these inactive phases to simulate actual usage behavior. In real, inactive phases pre-defined by the usage pattern or behavior, the energy storage device is not used, and it is not subjected to any load by inputting or extracting electrical energy. However, a load can act on the energy storage device in the form of its temperature without requiring sensors to detect it. Especially when the battery pack acting as an energy storage device has a high state of charge, high ambient temperatures can impose additional loads on the battery pack and accelerate aging.

[0019] Since the battery pack temperature during the inactivity phase is largely dependent on environmental conditions, such as ambient temperature, air pressure, or humidity, operating variables can be estimated or pre-defined based on these environmental conditions using modeling and dynamic models during this inactivity phase. In this way, the aging effects of the inactivity phase can also be modeled, thereby calculating and predicting these effects.

[0020] It can be specified that a driver-specific, data-driven usage pattern model, drawing on historical usage behavior, provides time-series usage patterns for generating at least one load variable. These usage patterns are particularly used to reconstruct data gaps in calculations and predict aging conditions.

[0021] Furthermore, a dynamic model with a so-called battery pack model can be provided, which is advantageously constructed as an equivalent circuit model, electrochemical model, or single-event model of the energy storage device. In particular, the dynamic model can describe the response to at least one load variable, and especially considers temperature dependence and / or nonlinearity in dynamic transport behavior. The dynamic model generates one or more operating variables, such as voltage, based on at least one load variable. In the case of an energy storage device having battery pack cells, current is provided at the input side of the dynamic model. The output of the dynamic model is the voltage response of the battery pack cells. This voltage response is particularly dependent on temperature, and nonlinear effects in transport behavior can also be considered. The dynamic model can be used to deduce voltage and state of charge (SOC) from current and temperature.

[0022] In addition, to determine the temperature change process during the inactive phase, a data-based temperature model can be used, which is trained to provide the temperature of the energy storage device based on at least one environmental condition.

[0023] Specifically, the temperature model can model the temperature of the energy storage device based on its location (especially its geographical location) and how it is exposed to ambient temperature.

[0024] Furthermore, the environmental conditions can be derived from weather information, particularly by retrieving data from a weather database, or by directly measuring the ambient temperature.

[0025] It can be stipulated that during the initial stage of the inactive phase, the temperature change process of the battery pack is modeled according to a physical cooling model that also takes into account the ambient temperature.

[0026] Furthermore, the temperature of the energy storage device can represent the load variable of a dynamic model, wherein the dynamic model models the voltage of the energy storage device based on the temperature and the current during the inactive phase, wherein the dynamic model can be constructed as an equivalent circuit model of the energy storage device, or as an electrochemical model, or as a single-particle model.

[0027] In particular, the dynamic model can be adapted based on the aging state of the model, in particular by updating the model parameters or state of the dynamic model based on the aging state of the model.

[0028] Furthermore, when the battery pack is used as an energy storage device, the dynamic model can provide a state of charge change process in the inactive phase based on the modeled aging state, corresponding to the linear state of charge change process between the state of charge at the beginning of the inactive phase and the state of charge at the end of the inactive phase.

[0029] Data-driven models typically require a training process, which in turn requires a training dataset. This training dataset can be provided by evaluating one or more devices and / or test vehicles. The training dataset assigns aging states to the process of changing the running variables. These aging states can be determined based on known methods, such as using field diagnostic measurements, like coulomb counts, or alternatively, by reference measurements in a laboratory setting.

[0030] The data-based aging state model can be trained on a training dataset, wherein the training dataset is subdivided into a training set and an extended training set, wherein the aging model is parameterized using the training set, and wherein the calibration model is trained based on the entire extended training set, and wherein the data-based aging state model is tested on a test set unknown to the model to determine the effectiveness of the data-based aging state model.

[0031] It can be stipulated that the energy storage device is operated according to the predicted aging process, wherein in particular, the remaining service life of the energy storage device is signaled according to the predicted aging process or the service life is extended by measures to reduce stress factors.

[0032] According to another embodiment, the energy storage device can be used to operate devices such as motor vehicles, electric bicycles, aircraft (especially drones), machine tools, entertainment electronic devices (such as mobile phones), autonomous robots, and / or home appliances.

[0033] According to another aspect, a device for predicting the aging state of an electrical energy storage device having at least one electrochemical unit, or also through rule-based and / or data-based mapping throughout the system, wherein the device is configured as follows: - Provides a data-driven aging state model, which is trained to assign modeled aging states to the electrochemical energy storage device based on the changes in the device's operating variables. - Provide the time-varying process of the operating variables characterizing the operation of the energy storage device; - Using the aging state model, the current or predicted aging state is determined based on the changes in the generated operating variables. The temperature change process of the energy storage device fills the data gaps in the time-varying process of operating variables caused by the inactive phase, and the temperature change process is derived from the provided environmental conditions. Attached Figure Description

[0034] The implementation methods are explained in more detail below with reference to the accompanying drawings.

[0035] Figure 1 A schematic diagram of a system is shown that provides driver-specific and vehicle-specific operating variables for determining the aging state of the vehicle battery pack in a central unit. Figure 2 A schematic diagram of the functional structure of the hybrid aging state model is shown. Figure 3 A flowchart illustrating a method for training a data-based aging state model is shown; and Figure 4 A schematic diagram of the functional structure of a usage-dependent prediction hybrid aging state model with aging state is shown. Figures 5a-5c The changes in state of charge, battery pack temperature, and the dependence of battery pack temperature on ambient temperature during the inactive phase are illustrated; and Figure 6 A graph showing the characteristic curves of the temperature model learned from training data is presented. Detailed Implementation

[0036] The following describes a method according to the invention based on a vehicle battery pack, which serves as an electrical energy storage device in a large number of motor vehicles that are of the same type. In these motor vehicles, a data-based aging state model for the respective vehicle battery pack can be implemented in a control unit. Based on operating variables from the vehicle battery packs in the fleet, the aging state model can be continuously updated or retrained in a central unit. Alternatively, the aging state model can run entirely in the cloud and can provide results for calculating and predicting aging states. The aging state model runs in the central unit and is used for aging calculation and aging prediction.

[0037] The examples above represent a large number of fixed or mobile devices with power supplies independent of the power grid, such as vehicles (electric vehicles, electric bicycles, etc.), facilities, machine tools, home appliances, IoT devices, etc., which are connected to a central unit (cloud) via corresponding communication connections (such as LAN, Internet).

[0038] Figure 1 A system 1 is shown for collecting fleet data in a central unit 2 to create, run, and evaluate an aging condition model. The aging condition model is used to determine the aging condition of electrical energy storage devices, such as vehicle battery packs or fuel cells in motor vehicles. Figure 1 A convoy 3 with multiple motor vehicles 4 is shown.

[0039] exist Figure 1 One of the motor vehicles 4 is shown in more detail. Each motor vehicle 4 has a vehicle battery pack 41 as a rechargeable energy storage device, an electric drive motor 42, and a control unit 43. The control unit 43 is connected to a communication module 44, which is adapted to transmit data between the respective motor vehicle 4 and a central unit (the so-called cloud).

[0040] The vehicle 4 sends operating variables F to the central unit 2. These operating variables at least describe variables affecting or being affected by the aging state of the vehicle battery pack. In the case of the vehicle battery pack, operating variable F can describe instantaneous battery pack current, instantaneous battery pack voltage, instantaneous battery pack temperature, and instantaneous state of charge (SOC), including at the battery pack level, module level, and / or battery cell level. Operating variables F are detected using a fast time grid that typically depends on a signal between 0.1 Hz and 100 Hz, and these operating variables can be periodically transmitted to the central unit 2 in uncompressed and / or compressed form. For example, time series can be transmitted to the central unit 2 in blocks at intervals of 10 minutes to several hours.

[0041] The operational characteristics M relating to the evaluation period can be generated from the operational variable F in the central unit 2 or, in other embodiments, in the corresponding vehicle 4. To determine the aging state, the evaluation period can range from several hours (e.g., 6 hours) to several weeks (e.g., one month). A common value for the evaluation period is one week.

[0042] Operating characteristics may include, for example, characteristics relating to the evaluation period and / or cumulative characteristics and / or statistical variables determined throughout the entire service life to date. Specifically, the operating characteristics may include, for example: histogram data on state of charge changes, temperature, battery pack voltage, and battery pack current (particularly multidimensional histogram data relating to battery pack temperature distribution at state of charge, charging current distribution at temperature, and / or discharging current distribution at temperature), cumulative total charge (Ah), average capacity increase during charging (particularly for charging processes where the charge increase exceeds a threshold share of the total battery pack capacity (e.g., 20%)), maximum differential capacity (dQ / dU: charge change divided by battery pack voltage change) during a charging process with a sufficiently large charge stroke at the state of charge, etc.

[0043] State of Health (SOH) is a key variable used to describe the remaining capacity or charge of a battery pack. It measures the aging of a vehicle's battery pack, battery module, or battery cell, and can be expressed as Capacity Retention Rate (SOH-C) or as the increase in internal resistance (SOH-R). Capacity Retention Rate (SOH-C) is expressed as the ratio of the measured instantaneous capacity to the initial capacity of a fully charged battery pack. The relative change in internal resistance (SOH-R) increases with increasing battery pack aging.

[0044] An aging state model is implemented in the central unit 2, which is specifically based entirely or partially on data. This aging state model can be used periodically, i.e., after a corresponding assessment period has elapsed, to determine the instantaneous aging state of the vehicle battery pack 41 based on operating characteristics and / or operating variables. In other words, the aging state of the vehicle battery pack 41 involved, or the aging state of the modules or battery cells associated with the energy storage device, can be determined based on operating variables and / or operating characteristics derived from the changes in operating variables of one of the motor vehicles 4 in the fleet 3.

[0045] Furthermore, the operating variables, namely the future aging state of the vehicle battery pack 41, can be determined, for example, through linear or nonlinear extrapolation or by means of a predictive model. Preferably, data-based algorithms can be used to predict operating characteristics, such as deep learning algorithms or autoregressive methods using ARIMA models, which characterize the trends and periodicity in the changes of historical operating characteristics to predict said operating characteristics.

[0046] Figure 2 The functional structure of an implementation of a data-based aging state model constructed in a hybrid manner is illustrated schematically by way of example. The aging state model includes a physical aging model 5 and a correction model 6. They obtain the operating variables F or operating features M for the current evaluation time period. The operating features M for the current evaluation time period are generated in feature extraction block 8 based on the time series of the operating variables F.

[0047] The operating variable F directly enters the physical aging state model 5, which is preferably designed as an electrochemical model and uses nonlinear differential equations to describe the corresponding electrochemical states, such as layer thickness (e.g., SEI thickness), changes in cyclizable lithium due to anode / cathode side reactions, rapid consumption of electrolyte, slow consumption of electrolyte, loss of active material in anode, loss of active material in cathode, etc.

[0048] Physical aging model 5 corresponds to the electrochemical model of the battery pack cell and cell chemistry. This model determines the internal physical battery pack state based on the operating variable F in order to provide at least one-dimensional physical aging state SOHph in the form of the aforementioned electrochemical state, which is linearly or nonlinearly mapped to capacity retention rate (SOH-C) and / or internal resistance increase rate (SOH-R) to provide them (SOH-C and SOH-R).

[0049] However, the model values ​​for the physical aging state SOHph provided by the electrochemical model are inaccurate under certain conditions. Therefore, it is stipulated that a correction variable k be used to correct these model values. The correction variable k is provided by a data-based correction model 6, which is trained using a training dataset from vehicles 4 in fleet 3 and / or using laboratory data.

[0050] To determine the corrected aging state SOH to be output, the outputs SOHph,k of the physical aging model 5 and the corrected model 6, preferably designed as a Gaussian process model, are applied to each other. Specifically, these outputs can be added or multiplied (not shown) in a summing block 7 to obtain the modeled aging state SOH to be output for the current evaluation time period. In the case of addition, the confidence level of the Gaussian process can still be used as the confidence level of the corrected aging value SOH to be output for the mixed model.

[0051] If there is insufficient data near the currently running feature point during the calculation of the aging state, the Gaussian process for correcting Model 6 reverts to its prior, and the confidence interval, and thus the uncertainty of the prediction, expands due to the extrapolated data conditions. The prior is 0 or close to 0 because the Gaussian process is trained on the residuals of a purely physical aging model, which are typically normally distributed near 0, in order to correct the purely physical aging model. If the Gaussian process knows some known training data points near the currently running feature point, which are incorporated into the Gaussian process through the training process and thus into the mixed aging state model, then the Gaussian process can accurately correct the aging state. This makes the uncertainty, and therefore the confidence interval, smaller or narrower because the data-driven correction, and therefore the aging state model as a whole, is very reliable due to good data conditions.

[0052] Other designs for data-based aging state models are also possible. For example, a data-based aging state model can be constructed as a non-hybrid, purely data-based model based on a probabilistic regression model or an AI-based regression model (particularly a Gaussian process model or a Bayesian neural network). This aging state model can be trained to provide the modeled aging state SOH from running feature points determined by the current running feature M of the current evaluation time period, where the running feature M is determined based on the time series of the running variable F in feature extraction block 8.

[0053] Principal Components Analysis (PCA) can be used to scale and reduce the dimensionality of running features, thereby reducing redundant linearly related information in the feature space before training the calibration model (unsupervised). Alternatively, kernel PCA can be used to map nonlinear effects while reducing the complexity of the data. Before and especially after dimensionality reduction, the entire running feature space (or principal component space) is normalized, for example, using minimum / maximum scaling or Z-transform.

[0054] Therefore, for an energy storage device having at least one electrochemical unit such as a battery pack cell, the aging state can be calculated and predicted. This method can also be applied to the entire system of the energy storage device through rule-based and / or data-based mapping. Thus, taking a battery pack as an example, aging prediction can be applied not only at the battery cell level but also directly at the module and battery pack levels.

[0055] Figure 3A flowchart illustrating the method for training a hybrid aging state model in central unit 2 is shown. For this purpose, the process of assigning the variation of the running variables to the training dataset of the aging states is defined. These training datasets can be collected from a large number of vehicles in central unit 2, or alternatively or supplementarily collected through laboratory-based measurements.

[0056] These training datasets are divided into training and testing sets. The training set is used to train the hybrid aging state model, while the testing set is used to validate the hybrid aging state model with the help of new, unknown data.

[0057] In step S1, the physical aging model 5 is parameterized based on the first part of the training set, specifically through parameter optimization using methods such as least squares. Here, the physical aging state SOHph, which is the output of the physical aging model 5, is assumed to be the aging state of the corresponding training dataset.

[0058] In step S2, the physical aging model is applied to the entire training set of the hybrid model, i.e., the number of training datasets, which includes at least the number of training datasets parameterized by the physical aging model, or even more. The error of the physical aging model is evaluated as a histogram of model bias in the overall error with respect to the residuals. This residual, in combination with the running feature M or the running variable F, contains all relevant information about the systematic weaknesses of the physical aging model 5. Information on how the physical aging model 5 performs on new training datasets not used to parameterize the physical aging model 5 is also derived.

[0059] In the next step S3, a data-based calibration model 6 is trained on the complete training set of the hybrid model. This training set of the hybrid model includes at least the training set corresponding to the physical model from step S1. For training the calibration model 6, running features M are extracted from the running variable F, and the internal states of the physical aging model 5 are used as a subset of M to map all running features to the error between the model predictions (physical aging states) of the physical aging model and the aging states corresponding to the labels on the training dataset. Thus, the calibration model 6 can learn the weaknesses of the physical aging model 5, thereby enabling it to correct physical aging states in calibration blocks.

[0060] Training of the data-based calibration model 6 can be performed using cross-validation and bootstrap aggregation to improve robustness and accuracy. If the calibration model has been trained, the trained hybrid aging state model can be validated in step S4 using a test set unknown to the model, thereby verifying the overall performance of the aging state calculation.

[0061] Now, trained hybrid aging state models can be used to determine the current or predicted aging state based on the operating variable F. This is particularly important for reconstructing inactive phases when calculating the aging state, especially in modeling temperature variation processes. To predict the aging state based on usage data (e.g., vehicle volume and driver usage patterns N), models such as... Figure 4 The model shown. Therefore, the trained hybrid model can also be based on... Figure 2 This includes feature extraction blocks and data reduction blocks (e.g., with principal component analysis: PCA).

[0062] Training of the hybrid aging state model can always be initiated when new labeled data becomes available, especially when this data contains new relevant information. Therefore, when running in the central unit based on fleet data, the hybrid aging state model can be continuously retrained to determine and predict aging states.

[0063] Figure 4 based on Figure 2 A hybrid aging state model is used, in which a dynamic model 9 and a pattern model 10 are additionally employed to generate the changes in battery pack voltage U and state of charge (SOC) based on load variables L such as battery pack current I and battery pack temperature T. The dynamic model may include a battery pack model that generates operating variables from the load variables based on an equivalent circuit model. This allows it to be independent of providing the actual operating variables of the vehicle battery pack 41, since the physical aging model 5 requires a time series or change process of the operating variable F to model the aging state. This is done according to the aging state of the energy storage unit 41, which leads to updates to the dynamic model. Therefore, the transmission behavior of the dynamic model 9 varies according to the age of the energy storage unit 41. This is preferably accomplished by updating the parameters and / or state of the dynamic model 9 based on the calculated modeled aging state SOC.

[0064] To incorporate aging state information into the system dynamics, usage mode model 10 is run based on the modeled aging state (SOH). Therefore, for example, in the case of a vehicle powered by a battery pack, it can be considered that the driver must charge the battery three times a week, instead of the initial two times, to travel the desired distance when the battery pack is aging.

[0065] Model 10 uses a pre-defined usage mode N. The usage mode is defined by usage parameters that can be vehicle-specific or driver-specific. For example, usage parameters can be model parameters of model 10.

[0066] Usage pattern N is learned specifically by usage pattern model 10 from fleet data vehicles, preferably with the aid of a data-based approach, and is used to simulate the usage behavior of a user or drivetrain in relation to the vehicle battery pack 41 involved. Usage pattern N can be provided by usage pattern database 11, which provides previously learned usage patterns to the vehicle battery pack 41 in the form of usage parameters.

[0067] Using mode N, the battery pack current I and battery pack temperature T, which are output as load variables L, are generated by mode model 10. Through dynamic model 9, a set of operating variables (F) with time-varying battery pack voltage U and state of charge (SOC) is formed from these time-varying processes, thus completing the time series of operating variables required for the aging model. The mode model can be correspondingly constructed as a recurrent neural network (LSTM, GRU), trained using real, vehicle-specific operating variables based on a large amount of data from the vehicle battery pack and the corresponding aging. Therefore, the mode model, based on the aging state SOH and the time series of load variables generated using mode N, characterizes the vehicle-specific load curve and leads to the determination of the aging state corresponding to mode N in the hybrid aging state model.

[0068] Using mode N can illustrate the load type of the energy storage device, particularly the periodic load derived from historical operating behavior. Mode N defines the load of the vehicle battery pack 41 involved and determines the variation process of load variables such that these variations correspond to the degradation or aging process of the vehicle battery pack 41 involved. Thus, the periodic frequent fast charging through fast charging current and fast charging can be explicitly or implicitly described using the mode. Therefore, using mode N in usage mode model 10 yields time series containing weekly and seasonal periodic effects or linear or nonlinear trends, particularly for battery pack current and temperature.

[0069] Using pattern N, it is also possible to learn and model inactive phases and thus detect or pre-define, and particularly illustrate, calendar-based aging, especially according to environmental conditions. These environmental conditions can be derived, for example, from climate tables, illustrating the changes in battery pack temperature within diurnal rhythms, seasons, etc., preferably using GPS-related weather data from the central unit (cloud). Therefore, the calculated temperature change process represents the reconstructed battery pack temperature signal, which can be used as input in the hybrid model for calculating SOH.

[0070] The inactive phase is a phase in which neither charging nor discharging occurs by drawing current. In the case of electric vehicles, this is, for example, the parking phase. There is an idle current situation here, but environmental factors such as temperature remain stressors for battery pack degradation. Furthermore, during the inactive phase, operating variables are generally not detected by sensors or are only detected temporarily, thus lacking current information about the vehicle's battery pack.

[0071] Specifically, the aging state and aging of the vehicle battery pack depend on the state of charge (SOC) of the vehicle battery pack 41 and the temperature changes during the inactive phase. By using model 10, this inactive phase can also be considered in light of environmental conditions, and the missing time-series changes of load variables L (such as battery pack current and battery pack temperature) can be reconstructed. Thus, for example, the self-discharge of the vehicle battery pack 41 depends on environmental influences, such as ambient humidity, ambient temperature, and aging state at the beginning and during the inactive phase. Generally, the SOC changes during the inactive phase can be modeled as a linear process over time, so the SOC change can be determined by the SOC1 at the beginning of the inactive phase and the SOC2 at or immediately following the end of the inactive phase.

[0072] The time-varying process of the operating variable F is monitored using inactive model 12. If there is no inactive phase, the operating variable F received from the vehicle involved 4 is used to calculate the state of aging (SOH) as described above. If it is determined that the time-varying process of the operating variable has a long time gap indicating that the vehicle has stopped, the time-varying process of the artificial operating variable is generated using pattern model 10 and dynamic model 9. For this purpose, the time-varying process of the battery pack temperature T and the battery pack current (e.g., 0A) is generated using temperature model 10a as described below in pattern model 10, and the time-varying process of the state of charge (SOC) and the state of voltage is generated using dynamic model 9 as described below. The generated data then replaces the conventionally received time-varying process of the operating variable.

[0073] Figure 5a An example is shown illustrating the change in State of Charge (SOC) during an inactivity phase when the vehicle is parked overnight without charging. If an inactivity phase is identified, this change in SOC during the inactivity phase can be generated using Dynamic Model 9.

[0074] With the help of a driver-specific or battery pack-specific temperature model 10a, which is part of the usage mode model 10, the process of change of environmental conditions (especially ambient temperature UT) during the inactivity phase can be modeled based on the process of change of environmental conditions (especially ambient temperature UT) during the inactivity phase and, if necessary, based on a description of the parking location. For example, the process of change of environmental conditions, or especially the process of change of ambient temperature, during the inactivity phase can be determined by retrieving historical, current and / or future weather information from a weather database, using weather data for the geographical location where the vehicle is parked during the inactivity phase.

[0075] Exemplarily derived in Figure 5b The temperature change process is shown in the diagram. The temperature change process of the main phase H during the inactive phase can be seen, which depends on the ambient temperature change process UT.

[0076] exist Figure 5c The diagram illustrates the variation of the battery pack temperature T from the ambient temperature UT. The temperature model maps the ambient temperature UT to the battery pack temperature BT (=T) and is shown exemplarily in the diagram. Figure 6 The characteristic curves are shown. To train the temperature model, training data obtained from battery pack temperature and ambient temperature (especially under steady-state conditions, such as when ignition is on) are used to learn the relationship between ambient temperature and battery pack temperature for different temperature ranges and, if necessary, the location of residence. The temperature model is then trained to assign the battery pack temperature T to the characteristics of the ambient temperature UT.

[0077] The usage mode N assigned to the vehicle's battery pack can explicitly or implicitly specify the parking location where the vehicle will be parked during inactivity periods. This parking location can be defined as the geographical location of the vehicle and / or a description of the type of parking space. For example, the description of the parking space type could specify whether the vehicle is in a protected parking location, an vacant parking location, or parked in a garage. Based on the parking location, an associated temperature model can be provided, mapping the ambient temperature to the temperature acting on the vehicle's battery pack.

[0078] This allows the temperature model to take into account other dominant conditions at the parking location, which can be learned during the model's training. Thus, for example, for a vehicle parked in a parking lot, the battery pack temperature change may be less than the ambient temperature change in summer, and more than the ambient temperature change in winter. For instance, the temperature model can describe the battery pack temperature based on the vehicle's parking location and the ambient temperature. This model can be used to optimally map the stable portion of the battery pack temperature after the cooling phase of the vehicle's battery pack.

[0079] Furthermore, in the initial stage A of the inactive phase, starting from the battery pack temperature at the time of parking, a generally unstable cooling process can be considered based on the ambient temperature at the vehicle's parking location and the description of the parking location. The battery pack temperature in the initial stage A can be modeled using a physically driven cooling model. If the physically driven cooling model reaches the battery pack temperature determined by the temperature model, or if the cooling model approaches the battery pack temperature in a manner less than a tolerance threshold, thus reaching a stable state, the evaluation of the battery pack temperature T using the cooling model ends, and the battery pack temperature T in the main stage H of the inactive phase is evaluated using the temperature model.

[0080] For each vehicle battery pack, operational data is used to learn and provide a corresponding usage pattern. This usage pattern defines a temperature model for inactive phases based on frequently used parking locations, which is driver-dependent and, when necessary, location-dependent.

[0081] The temperature model can be trained using training data for the vehicle battery pack involved, which is derived from the stable battery pack temperature and stable ambient temperature at the end of the inactive phase.

[0082] After the constant ~0A current and temperature (and state of charge) are given as time series signals during the inactive phase, the voltage response can now be calculated as a time series signal in the battery pack model in Dynamic Model 9 to reconstruct the complete inactive phase as much as possible.

[0083] The above process enables the correspondence between the state of charge (SOC) change process and the battery pack temperature (T) change process during the inactive phase, which are pre-defined in mode N.

[0084] Therefore, the calculation of current or predicted aging status can be appropriately utilized by taking into account the inactivity phases pre-given by using pattern N, thus enabling improved modeling of aging status even in the absence of time-varying processes of running variables. Furthermore, driver-specific, learned usage patterns can also be used to predict aging status.

[0085] Model 10 and dynamic model 9 are adapted by updating the aging state (SOH), preferably by updating the parameters and / or states of the corresponding models. Dynamic model 9 can be designed in various ways, such as equivalent circuit model, electrochemical model, single-particle model of battery cell, etc.

[0086] The output of dynamic model 9 is the modeled battery pack voltage (terminal voltage) and state of charge (SOC), which are respectively affected by the modeled state of aging (SOH).

[0087] Due to the possibility of predicting and modeling the aging state SOH, driver-specific aging state trajectories can be created using pattern N. Pattern N can be derived from stress factors and / or learned driver-specifically and data-driven based on historical fleet data. For this purpose, an autoregressive model is preferred, or alternatively, deep learning methods for pattern recognition can be used.

Claims

1. A computer-implemented method for predicting the state of aging (SOH) of a modeled energy storage device (41) having at least one electrochemical unit, the method comprising the following steps: - Provide a data-based aging state model, which is trained to assign a modeled aging state (SOH) to the energy storage device (41) according to the change process of the operating variable (F) of the energy storage device (41). - Provides the time-varying process of the operating variable (F) characterizing the operation of the energy storage device (41); - Using the aging state model, the current or predicted modeled aging state (SOH) is determined based on the change process of the generated operating variable (F). The temperature (T) change process based on the energy storage device (41) supplements the data gaps in the time-varying process of the operating variable (F) due to the inactive phase, the temperature change process being derived from at least one provided environmental condition (UT). The temperature of the energy storage device (41) represents the load variable of the dynamic model (9), wherein the dynamic model (9) models the voltage of the energy storage device (41) based on the temperature (T) and the current during the inactive phase. In the case where the battery pack is used as an energy storage device (41), the dynamic model (9) provides a state of charge change process based on the modeled state of aging (SOH) during the inactive phase, which corresponds to the linear state of charge change process between the state of charge (SOC1) at the beginning of the inactive phase and the state of charge (SOC2) at the end of the inactive phase.

2. The method of claim 1, wherein a data-based temperature model is used to determine the temperature (T) change process during the inactive phase, the data-based temperature model being trained to provide the temperature of the energy storage device based on at least one environmental condition.

3. The method according to claim 2, wherein the temperature model (10a) models the temperature (T) of the energy storage device (41) based on the location of the energy storage device (41) and the way the energy storage device (41) is exposed to ambient temperature.

4. The method according to claim 2 or 3, wherein the temperature model (10a) is trained to assign the environmental conditions to the temperature of the energy storage device (41), wherein training data for training the temperature model (10a) is generated by determining the stable environmental conditions and stable temperature of the energy storage device at the end of the inactive phase.

5. The method according to any one of claims 1 to 3, wherein the environmental conditions are derived from weather information or by directly measuring the at least one environmental condition.

6. The method according to any one of claims 1 to 3, wherein the temperature (T) change process is modeled according to a physical cooling model during a pre-given initial phase of the inactive phase.

7. The method according to any one of claims 1 to 3, wherein the dynamic model (9) is configured as an equivalent circuit model of the energy storage device (41) or as an electrochemical model or as a single-particle model.

8. The method according to claim 7, wherein the dynamic model (9) is adapted according to the modeled aging state (SOH) by updating the model parameters or state of the dynamic model (9) based on the modeled aging state (SOH).

9. The method of claim 1, wherein the data-based aging state model is constructed as a hybrid model and includes a physical aging model (5) and a trainable data-based correction model (6), wherein the physical aging model is based on an electrochemical model equation and is constructed to output a physical aging state (SOHph), wherein the correction model (6) is trained to correct the physical aging state (SOHph) and provide the corrected physical aging state as the modeled aging state (SOH).

10. The method of claim 1, wherein the load variable (L) corresponds to current (I) and temperature (T), and the operating variable (F) corresponds to current (I), voltage (U), temperature (T) and state of charge (SOC).

11. The method according to any one of claims 1 to 3, wherein the trainable data-based correction model is constructed in the form of a regression model.

12. The method according to any one of claims 1 to 3, wherein the electric energy storage device (41) is used to operate motor vehicles, electric-assisted vehicles, aircraft, machine tools, entertainment electronic devices, autonomous robots and / or household appliances.

13. The method according to claim 1, wherein the electrochemical unit is a battery cell.

14. The method according to claim 2, wherein the at least one environmental condition includes ambient temperature, air pressure, or air humidity.

15. A device for predicting the state of aging (SOH) of an energy storage device (41) modeled with at least one electrochemical unit, or also through rule-based and / or data-based mapping throughout the system, wherein the device is configured as follows: - Provide a data-based aging state model, which is trained to assign a modeled aging state (SOH) to the energy storage device (41) according to the change process of the operating variable (F) of the energy storage device (41). - Provides the time-varying process of the operating variable (F) characterizing the operation of the energy storage device (41); - Using the aging state model, the current or predicted modeled aging state (SOH) is determined based on the change process of the generated operating variable (F). The temperature (T) change process based on the energy storage device (41) supplements the data gaps in the time-varying process of the operating variable (F) due to the inactive phase, the temperature change process being derived from at least one provided environmental condition (UT). The temperature of the energy storage device (41) represents the load variable of the dynamic model (9), wherein the dynamic model (9) models the voltage of the energy storage device (41) based on the temperature (T) and the current during the inactive phase. In the case where the battery pack is used as an energy storage device (41), the dynamic model (9) provides a state of charge change process based on the modeled state of aging (SOH) during the inactive phase, which corresponds to the linear state of charge change process between the state of charge (SOC1) at the beginning of the inactive phase and the state of charge (SOC2) at the end of the inactive phase.

16. A computer program product comprising instructions that, when executed by at least one data processing device, cause the at least one data processing device to perform the steps of the method according to any one of claims 1 to 14.

17. A machine-readable storage medium comprising instructions that, when executed by at least one data processing device, cause the at least one data processing device to perform the steps of the method according to any one of claims 1 to 14.

Images