DE102020212280A1 - Method and device for training a state model based on artificial intelligence for determining a state variable of an electrical energy store - Google Patents

Abstract

The invention relates to a computer-implemented method for training a state model for determining a state of an electrical energy storage device of a device operated therewith, with the following steps: - providing a data-based state model that has an operating feature point a modeled state variable, in particular an aging state of the energy storage device, and a state -Assigns uncertainty measure;- Providing a database with progressions of real operating characteristic points from a plurality of energy storage devices for successive evaluation periods;- Training or updating the state model depending on at least one training data set,wherein the at least one training data set is generated with the following steps:◯ Determining an improvement measure for each of the current operating feature points of the energy storage determined for the current evaluation period, with the measure of improvement indicating how the state Unsi improve security at all real operating characteristic points if a state variable is determined for the respective current operating characteristic point;◯ selecting at least one operating characteristic point with the best improvement measure;◯ determining a corresponding state variable at the at least one selected operating characteristic point as a label; at least one selected operating feature point and the determined corresponding state variable.

Description

technical field

The invention generally relates to the characterization of system states of electrical energy storage devices, such as in electrically operated devices such as motor vehicles, in particular electric vehicles or hybrid vehicles, and also to measures for determining a state of an electrical energy storage device, such as. B. a device battery or a fuel cell system.

Technical background

Electrical energy stores, such as batteries, are generally used to supply technical devices with energy that is not connected to the grid. Energy stores within the meaning of the following description also include energy converters, such as fuel cell systems. For example, the power supply of electrically driven motor vehicles using z. B. a vehicle battery or a fuel cell system. This supplies electrical energy for the operation of vehicle systems and in particular the drive system. The aging condition of the electrical energy store deteriorates over the course of its service life, resulting in a decreasing maximum storage capacity. A measure of the aging of the device battery depends on an individual load on the device battery, i. H. on the usage behavior of a user and the type of device battery.

Although the current state of aging can be determined using a purely physical aging model based on historical operating variable trends, this model is often imprecise. This inaccuracy of the conventional aging model makes it difficult to predict the course of the aging state. However, the prediction of the course of the aging state of the vehicle battery is an important technical variable, since it allows an economic evaluation of the residual value of the vehicle battery.

Other states of a vehicle battery, such as a state of charge, cannot usually be determined with high accuracy using conventional physical models.

Both physical and data-based or hybrid state models require precisely recorded state labels for training data, since they represent the basis of the model parameterizations. A continuous label generation after its start of use in the application enables a constant improvement of the parameterization of the state model in order to determine and predict the state variable to be characterized.

Disclosure of Invention

According to the invention, a method for providing training data for training a state model for modeling a state of an electrical energy store according to claim 1 and a corresponding device according to the independent claim are provided.

Further developments are specified in the dependent claims.

• - Bereitstellen eines datenbasieren Zustandsmodells, das einem Betriebsmerkmalspunkt eine modellierte Zustandsgröße, insbesondere einen Alterungszustand des Energiespeichers, und ein Zustands-Unsicherheitsmaß zuordnet;
• - Bereitstellen einer Datenbank mit Verläufen von realen Betriebsmerkmalspunkten von einer Vielzahl von Energiespeichern für aufeinanderfolgende Auswertungszeiträume;
• - Trainieren oder Aktualisieren des Zustandsmodells abhängig von mindestens einem Trainingsdatensatz,

wobei der mindestens eine Trainingsdatensatz mit folgenden Schritten generiert wird:

• ◯ Ermitteln eines Verbesserungsmaßes für jeden der für den aktuellen Auswertungszeitraum ermittelten aktuellen Betriebsmerkmalspunkt der Energiespeicher, wobei das Verbesserungsmaß angibt, wie sich Zustands-Unsicherheiten an allen realen Betriebsmerkmalspunkten verbessern, wenn eine Zustandsgröße für den jeweiligen aktuellen Betriebsmerkmalspunkt bestimmt wird;
• ◯ Auswählen mindestens eines Betriebsmerkmalspunkts mit dem besten Verbesserungsmaß;
• ◯ Ermitteln einer entsprechenden Zustandsgröße an dem mindestens einen ausgewählten Betriebsmerkmalspunkt als Label;
• ◯ Generieren des mindestens einen Trainingsdatensatzes aus dem mindestens einen ausgewählten Betriebsmerkmalspunkt und der ermittelten entsprechenden Zustandsgröße.

According to a first aspect, a computer-implemented method for training a state model for determining a state of an electrical energy store of a device operated therewith is provided, with the following steps:

• - Providing a data-based status model, which assigns a modeled status variable, in particular an aging status of the energy store, and a status uncertainty measure to an operating feature point;
• - Providing a database with profiles of real operating feature points from a plurality of energy stores for consecutive evaluation periods;
• - Training or updating the state model depending on at least one training data set,

wherein the at least one training data set is generated with the following steps:

• ◯ Determination of a measure of improvement for each of the current operating characteristic points of the energy storage devices determined for the current evaluation period, with the measure of improvement indicating how state uncertainties improve at all real operating characteristic points when a state variable is determined for the respective current operating characteristic point;
• ◯ selecting at least one operational attribute point with the best measure of improvement;
• ◯ Determination of a corresponding state variable at the at least one selected operating characteristic point as a label;
• ◯ Generate the at least one training data set from the at least one selected operating characteristic point and the determined corresponding state variable.

Furthermore, the degree of improvement for each of the current operating feature points can be approximately determined by a simulation by updating the provided state model with a simulation training data set that is created for the respective current operating feature point, in order to obtain a simulation state model, the simulation state -Uncertainties for each of the operating characteristic points of the progression of operating characteristic points of real energy storage devices are determined using the simulation state model, with the degree of improvement being determined depending on the simulation state uncertainties.

In particular, the measure of improvement can be determined as a sum of the uncertainties of the simulation state, in particular at least one of the uncertainties of the simulation state being reduced using at least one predetermined rule based on domain knowledge before the determination of the measure of improvement.

Furthermore, the state model can correspond to an aging state model for providing an aging state depending on the operating characteristics or a state of charge model for providing a state of charge depending on the operating characteristics of the energy store.

It can be provided that the data-based state model includes a data-based machine learning model that is designed to specify a state uncertainty for a modeled state variable, the state model, in particular a Gaussian process model as a supervised learning model with quantified uncertainty calculation . Alternatively, the supervised learning model can be implemented in particular as an ensembler method or as a Bayesian neural network.

The operating characteristics can be derived from time series of continuously recorded operating variables, with the operating characteristics being determined for consecutive evaluation periods, with a number of operating characteristics defining an operating characteristic point for a specific evaluation period of a specific energy store.

For electrical energy stores in higher-level technical devices such as motor vehicles, it is necessary to determine states in order to be able to operate the higher-level system in an improved or optimal manner. These are specified as state variables. For example, the determination of an aging state of the electrical energy store and a charge state of a rechargeable battery are important state variables that are essential for the operation of the technical devices operated with them, such as an electrically operated motor vehicle.

Such state variables cannot usually be modeled with great precision using physically based models. Furthermore, in this case, the inaccuracy of the calculation of the state variables can hardly or only with difficulty be quantified online. Therefore, the use of data-based or hybrid state models is increasingly being considered for modeling these state variables. In particular when a large number of training data is available, such state models can enable a very precise state calculation or state prediction.

The training of such data-based state models requires highly accurate training data for supervised learning techniques, which usually have to be measured or evaluated in advance. The process of assembling training data is commonly referred to as label generation.

After the training, data-based state models can be implemented in advance in devices operated with them by transferring corresponding model parameters. It is also possible for devices that have a regular connection to a central unit (cloud) to regularly receive the model results, such as the state variables or updated model parameters of the data-based state model, from the central unit. In particular, if operating variables that have been measured in a large number of devices equipped with the same electrical energy storage devices are evaluated in the central unit, the data-based status model can be continuously retrained or updated in the central unit and the corresponding model parameters of the updated or retrained status model can be sent to the devices of the same type be transmitted back in order to improve a state prediction there. The status calculation or status prediction preferably takes place in the central unit (cloud) and the results are made available to the devices as statuses.

To provide training data for training the data-based status model, a status variable can be derived for one of the devices from the operating variables obtained in a complex manner, usually by means of time-consuming diagnostic measurements, which is determined as a label, in particular in the central unit, in order to create a training data set there, based on Operating variables or operating characteristics derived therefrom and associated state variable (Label) to provide. For example, to determine an aging state model in a central unit, operating variables for the operation of a vehicle battery of an electrically driven motor vehicle can be continuously transmitted to the central unit and evaluated there by evaluating certain operating profiles, e.g. B. triggered by diagnostic measurements, such as a complete charging cycle under defined load and environmental conditions, an indication of aging can be derived, so that there a training data set is made available.

In an initial phase of the operation of a technical system with a central unit and a large number of similar devices that are in communication with the central unit and are based on a data-based state model for estimating a state variable, the data-based state model is often not yet sufficiently accurate in all areas of an input data space at the start of operation trained. In order to continuously improve the performance of the data-based status model during its use, the data-based status model can be updated regularly in the central unit based on training data sets that can be determined using the operating variables recorded by the devices. Appropriate access variables cannot be determined in a simple manner by measuring real devices for all areas of the input variable space, which is defined by the operating variable points or operating characteristic points, so that the data-based status model remains very imprecise in these areas and may not be able to be used.

The above method now provides for predictively evaluating the data-based status model, in particular with the aid of a simulation, in order to identify at least one operating feature point. The identified operating feature point can be the basis for determining a training data set, with which the status model can be updated in an optimized manner in order to improve the status uncertainties at the real operating feature points. Fleet data can be used for this purpose, which indicate the modeled state variable progressions via respective progressions of real operating characteristic points derived from operating variables. This is preferably done with the help of the simulation, in which a simulation training data set is assumed, which assumes a possible state at a current operating characteristic point and determines an overall improvement in the reliability/confidence of the state predictions. The operating characteristic point at which the best possible reduction in the overall uncertainty is achieved corresponds to the identified operating characteristic point. This represents the starting point for generating a new training data set.

The determination of the operating characteristic point for the training data record to be generated is thus carried out with the aid of an improvement measure, which is obtained through simulations for each energy storage device or for each of the current operating characteristic points of the energy storage device determined for the current evaluation period. The measure of improvement indicates how the status uncertainties improve at all real operating feature points in the database if a simulation training data set with a status variable is determined for the respective current operating feature point and the status model is updated accordingly. The degree of improvement is approximately determined by the simulations, in which (only for the simulation) the status model is evaluated on the basis of available information on training data with regard to the added value for fleet relevance and the associated global model uncertainty reduction in order to retrain an aging status model.

The simulation state model can now be used in each case to determine the measure of improvement as a sum of the simulation state uncertainties. The individual simulation state uncertainties can be reduced according to rules based on domain knowledge, and the correspondingly reduced simulation state uncertainties can be taken into account for determining the measure of improvement.

Since the current operating characteristic points are known for the large number of energy stores, such an assessment can be carried out based on a reduction in the overall model uncertainty for the totality of the curves of the real operating characteristic points of the energy stores. According to the above method, this reduction can be quantified using a respective improvement measure for each of the current operating feature points, with the improvement measure indicating how the status uncertainties at all real operating feature points, i. H. for the entirety of the operating feature points that have occurred in real operation if a state variable for the respective current operating feature point is determined by label generation. In this case, the uncertain, modeled state variable of the state model is preferably accepted as the label state variable.

This enables targeted label generation based on that of the energy storage device (with regard to its current operating characteristic point) for which the specific improvement measure requires the highest possible reduction in the total mo indicates the uncertainty of the fleet, which consequently quantifies the label relevance.

A method for statistical label generation can thus be used, with the determination of the corresponding state variable being carried out, in particular in connection with rules based on domain knowledge for reducing state uncertainties.

In particular, there are system-related dependencies for state variables with regard to their time profiles, which are known and are available or can be used as domain knowledge. This makes it possible to reduce large confidence intervals or high state uncertainties of state variable predictions (modeled state variables) resulting from the modeling with the state model, in that physically impossible value ranges of the relevant state variable are eliminated by setting e.g. B. the upper and lower limit of the confidence interval can be excluded.

Furthermore, the reduction of the status uncertainties at the selected operating feature points can be done by at least one rule, the rule depending on a temporal development of the operating feature points and status uncertainties of operating feature points at evaluation periods other than the evaluation periods of the selected operating feature points.

In particular, the state uncertainties at the selected operating feature points can be reduced by limiting the state uncertainty at each of the selected operating feature points to an upper or lower limit of a confidence interval that is defined by the state uncertainty at an evaluation period preceding the relevant operating feature point is determined, and/or that the state uncertainty at each of the selected operating characteristic points is limited to a lower or upper limit of a confidence interval, which is determined by the state uncertainty at an evaluation period following the relevant operating characteristic point. This can be traced back to domain knowledge that a slowly time-varying state variable such as the aging state of an energy storage device only changes sluggishly.

Alternatively or additionally, the state uncertainties at the selected operating feature points can be reduced by limiting the state uncertainty at each of the selected operating feature points with an upper and a lower limit of a corresponding confidence interval, which is obtained by interpolating upper limits of confidence intervals from is determined for the relevant operating characteristic point preceding and subsequent evaluation periods.

Alternatively, the corresponding state variable can be determined by asking the technical system assigned to the at least one selected operating feature point to carry out a measurement of the state variable, with the measured state variable being used as the label target variable to be measured after the measurement in order to update the training data set to create the state model.

In the above method, a data-based status model, which can also be designed as a hybrid status model, is made available for a large number of similar technical devices in a central unit. The data-based state model is updated and improved by further training data sets being created by evaluating available training data sets and evaluating the data-based state model in real operation of the large number of technical devices, which are suitable for retraining or updating the data-based state model. As a result, an initially provided data-based state model can be retrained in areas of the input variable space in which there is a high degree of state uncertainty.

Provision can be made for the operating variables to be transmitted from a large number of technical systems to a central unit, with a database containing operating characteristic points from real energy stores in the large number of technical systems being provided, with the method being executed in the central unit, with the selection of the operating characteristic points from the database is done.

• - Bereitstellen eines datenbasieren Zustandsmodells, das einem Betriebsmerkmalspunkt eine modellierte Zustandsgröße, insbesondere einen Alterungszustand des Energiespeichers, zuordnet;
• - Bereitstellen einer Datenbank mit Verläufen von realen Betriebsmerkmalspunkten von einer Vielzahl von Energiespeichern für aufeinanderfolgende Auswertungszeiträume;
• - Trainieren oder Aktualisieren des Zustandsmodells abhängig von mindestens einem Trainingsdatensatz,

wobei der mindestens eine Trainingsdatensatz mit folgenden Schritten generiert wird:

• ◯ Ermitteln eines Verbesserungsmaßes für jeden der für den aktuellen Auswertungszeitraum ermittelten aktuellen Betriebsmerkmalspunkt der Energiespeicher, wobei das Verbesserungsmaß angibt, wie sich Zustands-Unsicherheiten an allen realen Betriebsmerkmalspunkten verbessern, wenn eine Zustandsgröße für den jeweiligen aktuellen Betriebsmerkmalspunkt bestimmt wird;
• ◯ Auswählen mindestens eines Betriebsmerkmalspunkts mit dem besten Verbesserungsmaß;
• o Ermitteln einer entsprechenden Zustandsgröße an dem mindestens einen ausgewählten Betriebsmerkmalspunkt als Label;
• ◯ Generieren des mindestens einen Trainingsdatensatzes aus dem mindestens einen ausgewählten Betriebsmerkmalspunkt und der ermittelten entsprechenden Zustandsgröße.

According to a further aspect, a device, in particular a control unit in a central unit which is in communication with a large number of devices with energy stores, is provided for training a state model for determining a state of an electrical energy store, the device being designed for:

• - Providing a data-based status model, which assigns a modeled status variable, in particular an aging status of the energy store, to an operating feature point;
• - Providing a database with histories of real operating characteristic points from a Multiple energy stores for consecutive evaluation periods;
• - Training or updating the state model depending on at least one training data set,

wherein the at least one training data set is generated with the following steps:

• ◯ Determination of a measure of improvement for each of the current operating characteristic points of the energy storage devices determined for the current evaluation period, with the measure of improvement indicating how state uncertainties improve at all real operating characteristic points when a state variable is determined for the respective current operating characteristic point;
• ◯ selecting at least one operational attribute point with the best measure of improvement;
• o determining a corresponding state variable at the at least one selected operating feature point as a label;
• ◯ Generate the at least one training data set from the at least one selected operating characteristic point and the determined corresponding state variable.

character list

• 1 eine schematische Darstellung eines Systems für eine Fahrzeugflotte mit einer Vielzahl von Kraftfahrzeugen und einer Zentraleinheit zum Bereitstellen eines datenbasierten Zustandsmodells;
• 2 ein Flussdiagramm zur Veranschaulichung eines Verfahrens zum Trainieren bzw. Aktualisieren eines datenbasierten Zustandsmodells, insbesondere in Form eines Alterungszustandsmodells für eine Fahrzeugbatterie der Kraftfahrzeuge;
• 3 eine Veranschaulichung der Reduzierung des Gesamtmaßes für alle ermittelten Zustands-Unsicherheiten an den Betriebsmerkmalspunkten durch Label-Generierung an einem Betriebsmerkmalspunkt.

Embodiments are explained in more detail below with reference to the accompanying drawings. Show it:

• 1 a schematic representation of a system for a vehicle fleet with a large number of motor vehicles and a central unit for providing a data-based status model;
• 2 a flowchart to illustrate a method for training or updating a data-based status model, in particular in the form of an aging status model for a vehicle battery of motor vehicles;
• 3 an illustration of the reduction of the overall measure for all determined state uncertainties at the operating feature points by label generation at an operating feature point.

Description of Embodiments

The method according to the invention is described below using vehicle batteries as electrical energy stores in a large number of motor vehicles as devices of the same type. A data-based state of health model for the respective vehicle battery can be implemented in a control unit in motor vehicles. The aging state model is an example of a state model that characterizes an internal state of the energy store that cannot be measured directly. The state of health model can be continuously updated or retrained in a central unit based on operating variables of the vehicle batteries from the vehicle fleet.

The above example is representative of a large number of stationary or mobile devices with a network-independent energy supply, such as vehicles, systems, loT devices, energy supply systems, autonomous robots and the like, which have a corresponding communication connection (e.g. LAN, Internet) with a Central unit (cloud) are connected. The state variables represent variables that cannot be determined in a simple manner with high precision and at the same time based on models in the similar devices, but can only be determined by complex calculations, internal, in particular destructive measurements or according to predefined operating cycles of the device.

1 shows a system 1 for providing fleet data from motor vehicles 4 of a vehicle fleet 3 in a central unit 2. In the central unit 2, based on the fleet data, a prediction of a course of an aging state of a vehicle battery of a respective motor vehicle 4 of the vehicle fleet 3 is to be made.

One of the motor vehicles 4 is in 1 shown in more detail. The motor vehicles 4 each have a vehicle battery 41 as a rechargeable electrical energy store, an electric drive motor 42 and a control unit 43 . The control unit 43 is connected to a communication module 44 which is suitable for transmitting data between the respective motor vehicle 4 and a central unit (cloud). The control unit 43 is connected to a sensor unit 45 which has one or more sensors in order to continuously record operating variables.

The central unit 2 has a data processing unit 21 in which the method described below can be executed, and a database 22 for storing operating variables and aging states of vehicle batteries that have been determined in a large number of vehicles 4 of the vehicle fleet 3 .

The motor vehicles 4 send the operating variables F to the central unit 2, which indicate at least variables on which the aging state of the vehicle battery depends. The company sizes F In the case of a vehicle battery, they can indicate an instantaneous battery current, an instantaneous battery voltage, an instantaneous battery temperature and an instantaneous state of charge (SOC: State of Charge). The operating variables F are recorded in a rapid time frame between 2 Hz and 100 Hz, with their curves being regularly transmitted to the central unit 2 in uncompressed and/or compressed form.

From the operating variables F, operating characteristics can be generated in the central unit 2 or, in other embodiments, in the respective motor vehicles 4, which relate to an evaluation period. The evaluation period for determining the aging state can be a few hours (e.g. 6 hours) to several weeks (e.g. one month). A usual value for the evaluation period is one week.

The operating characteristics can include, for example, characteristics relating to the evaluation period and/or accumulated characteristics and/or statistical values ascertained over the entire service life to date. In particular, the operating features can include, for example: Histogram data on the course of the state of charge, the temperature, the battery voltage, the battery current, in particular histogram data with regard to 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, accumulated total charge (Ah), an average increase in capacity during a charge (especially for charges where the increase in charge is above a threshold proportion (e.g. 20%) of the total battery capacity), a maximum of the differential capacity (dQ/dU: change in charge divided by change in battery voltage) and Further.

Additional information can be derived from the operating characteristics: a load pattern over time, such as charging and driving cycles, determined by usage patterns (e.g. fast charging at high currents or strong acceleration or braking processes with recuperation), a usage time of the vehicle battery, a charge accumulated over the running time and a discharge accumulated over the term, a maximum charging current, a maximum discharging current, a charging frequency, an average charging current, an average discharging current, a power throughput during charging and discharging, a (especially average) charging temperature, a (especially average) spread of the state of charge and the like.

The state of health (SOH: State of Health) is the key variable for indicating a remaining battery capacity or remaining battery charge. The aging state is a measure of the aging of the vehicle battery or a battery module or a battery cell and can be specified as a capacity retention rate (SOH-C) or as an increase in internal resistance (SOH-R). The capacity retention rate SOH-C is given as a ratio of the measured instantaneous capacity to an initial capacity of the fully charged battery. The relative change in internal resistance SOH-R increases as the battery ages.

In 2 the method for training or updating the data-based state of health model is described in more detail using a flow chart. The method is executed in the data processing unit 21 of the central unit and can be implemented there as software and/or hardware.

In step S1, a data-based state of health model is provided, which is pre-trained to output a modeled state of health and a modeling inaccuracy depending on operating feature points. Furthermore, a database is provided that provides time series of operating characteristic points for electrical energy stores of the vehicles in the vehicle fleet for each evaluation period for a specific vehicle battery.

Operating variables F are continuously received from the vehicles 4 of the vehicle fleet 3 in step S2. The operating variables F can include curves of the current battery current, the current battery voltage, the current battery temperature, the current state of charge and the like.

In step S3, operating characteristics for successive evaluation periods are determined from the curves of the operating variables F. In the case of a determination of the aging state of the vehicle battery, these evaluation periods can be one day to one month, preferably one week. As a rule, for the model-based determination of the state of health of a specific vehicle battery, an operating characteristic point, which is determined by the individual operating characteristics for a specific vehicle battery within an evaluation period, is assigned to a state of health using the data-based state of health model. This data-based model can also be designed as a hybrid model, in particular a combination of physical and data-driven models, preferably with a supervised learning component.

In step S4, a measure of improvement is now determined on the basis of all current operating characteristic points for vehicle batteries 41, ie all vehicle batteries 41, with regard to their current operating state. The measure of improvement indicates how the status uncertainties would improve at all real operating feature points if a state variable is assumed for the respective current operating feature point. It is thus possible to determine the vehicle battery that results in a measure of improvement for the assigned current operating feature point.

The measures of improvement are determined in each case using a simulation state model which, starting from the state model, is created by post-training with a simulation training data set. The simulation training data set assumes a training data point at the current operating point, to which a state label is assigned, which corresponds, for example, to the model value of the aging state at the current operating point based on the aging state model.

Based on the simulation state model, the state uncertainties are now determined for all operating characteristic points. That is to say, for each evaluation period, aging states and their associated state uncertainty are determined for all energy stores that are taken into account in the central unit.

The state uncertainties result from querying the aging state model. However, individual state uncertainties at individual operational feature points can be reduced using domain knowledge. For example, the mean uncertainty of the previous training process can be used as the potential for how much the uncertainty at the relevant feature point can be reduced. Consequently, the domain knowledge for the training process is incorporated into the simulation model. Furthermore, for each operating characteristic point under consideration, a vehicle-specific reduction of the corresponding confidence interval (indicated by the state uncertainty) can be undertaken using domain knowledge. For this purpose, the course of the aging state over time (always for the individually identified vehicle battery) can be analyzed by the associated evaluation period, which is assigned to the respective operating characteristic point of the identified vehicle battery. Each operating feature point is assigned to an evaluation period, with further operating feature points for the relevant vehicle battery generally being present before and after the evaluation period.

3 shows an example for the reduction of an overall measure of the state uncertainties at further operating feature points in a label generation at a specific operating feature point. 3 shows the modeled m aging condition with a corresponding confidence interval for two different vehicles F1, F2 on the basis of operating characteristic points recorded one after the other in different temperature ranges. The aging state model used has a higher model uncertainty at high temperatures, so that the confidence interval becomes correspondingly larger. This may be due to the fact that high temperatures of 70°C are not yet adequately represented in the training database.

If there is no label for the vehicle F1 at a specific point in time t under a specific condition that was neither generated by fleet vehicles nor determined from diagnostic measurements, then this must be determined during operation. For vehicle F1, the modeled aging condition has a low uncertainty for 23°C and a high uncertainty for 40°C. Because of the high uncertainty at time t-1, there is now a great deal of interest in a label at time t to minimize confidence in the model at t-1.

In order to evaluate the effect of a possible label generation of vehicle 1 at time t with regard to the relevance and added value for the entire vehicle fleet 3, a possible chain reaction across fleet vehicles (swarm intelligence) is simulated, as shown for example with regard to vehicle F2. This exploits the fact that a slowly time-varying state variable such as the state of aging cannot change significantly within a short period of time. An effect of label generation in vehicle F1 at time t is simulated. The effect of this label generation is directly visible in vehicle F1 (1st chain reaction, represented by arrow P1), since in this case the uncertainty at time t-1 at 40°C can also be minimized, resulting in a new model value at t -1, especially if several vehicles can be analyzed in this state.

Under the premise that this label generation is now also successful at 40°C, the reduced uncertainty of the model at 40°C and other model uncertainties are minimized, e.g. at T = 70°C. The reduction in state uncertainties is represented by arrows P2 and P3.

In this example, by generating a label for vehicle F1 at time t, a certain feature (e.g. temperature 50°C), not only the label can be determined precisely for this operating characteristic point, but also the parameter gaps at 40°C (vehicle F1) and 70°C (vehicle F2) can be closed.

With a global fleet analysis, the effect of a possible label generation of each vehicle (current operating characteristic point) and thus each available parameter space at the current time can be quantified and evaluated in a forward-looking manner with regard to the added value for the vehicle fleet 3 . The advantageous embodiment of this invention allows the total degree of uncertainty to be expected to be maximized in the form of an optimization problem.

In particular, the confidence interval can be restricted to aging states that are no greater than an upper limit of the confidence interval of a previously determined aging state and no lower than a lower limit of an aging state determined in a later evaluation period. Such a restriction can be made on the basis of the domain knowledge that the temporal progression of aging states can only develop in a strictly monotonically decreasing manner. Thus, the confidence limits of the aging state defined as uncertain are shifted in the direction of the modeling value of the aging state in such a way that a smaller confidence interval results.

Other rules that can be derived from domain knowledge for limiting the confidence ranges can also be defined. Thus, as a possible further rule, a continuous and sluggish, or piecewise almost linear course of an aging state within a limited period of a few, for example between 3 and 10, evaluation periods can be assumed. Such a restriction can be made if e.g. B. it can be seen from the fleet load profile that the load profile for the vehicle under consideration was constant in terms of energy throughput, temperature conditions, charging behavior, driving behavior and the like in the period under consideration between the previous evaluation period and the following evaluation period. An upper and lower limit of a limited confidence interval for the aging state in question can result from piecewise interpolation of the upper confidence limits of a preceding and a subsequent evaluation period and the lower confidence limits of the preceding and the subsequent evaluation period.

In this way, the confidence intervals for the aging states of the identified vehicle battery can be limited.

The measure of improvement can be taken as a total measure for all determined state uncertainties. For example, the measure of improvement is determined as the sum of all state uncertainties determined with one of the simulation state models.

In step S5, one or more of the energy stores (current operating feature points) are selected for which there is a maximum degree of improvement.

In this way, the one or those operating feature points can be identified that bring about the best possible effect for reducing the state uncertainty for other operating feature points as well. Thus, the above procedure uses the determination of the operating characteristic points for which a label is to be generated or from which a training data set is to be generated, also to reduce the state uncertainty for other operating variable points, so that the label relevance for other fleet participants is maximized

In step S6, a method for label generation, specifically for determining the aging state, is carried out for the one or more selected energy stores. One or more training datasets can thus be generated in order to retrain the state of health model.

In a subsequent step S7, the data-based aging state model is retrained or updated based on the new training data set. A plurality of training data sets determined in this way can also initially be determined and/or training data sets can be collected during a specific period of time before an update takes place.

The corresponding state variable can be determined using a statistical method, in particular in connection with rules based on domain knowledge for reducing state uncertainties and using a clustering method and joint consideration of operational characteristic points that are similar to the identified operational characteristic point.

From the vehicle batteries 41, the clustering method determines those with similar operating characteristic points, in particular operating characteristic points with a predefined Euclidean distance from a centroid point that is not greater than a predefined threshold value. The centroid point corresponds to an artificial operating feature point that is not assigned to a specific energy store. The goal of the Fleet Clus The primary analysis is to identify vehicle batteries that are in a comparable operational characteristic space (parameter space). K-means++ or competitive learning can be used as possible unsupervised clustering methods.

In a subsequent step S8, the model parameters of the updated aging state model or the states determined by the model can be transmitted back to the vehicles in the vehicle fleet 3, so that the data-based aging state model in the motor vehicle 4 can be used to determine the aging state SOH of the vehicle battery 41. This step is optional. The aging state model can also be operated in the central unit 2 and executed there.

Alternatively, the corresponding state variable can be determined by requesting the device assigned to the at least one selected operating feature point to measure the state variable, with the measured state variable being used after the measurement to create the training data set.

A training data set can also be determined by measuring a vehicle 4 using a diagnostic measurement in order to determine a corresponding training data set. For this purpose, active learning can be used, which selects from all possible energy stores those for carrying out a diagnostic measurement that best possible closes the desired effect on reducing the overall model uncertainty in the operating characteristics space, especially after step S5 if several highly relevant energy stores have been selected. The final selection of vehicles from the vehicle fleet with regard to their maximized label relevance for the fleet can be carried out using pool-based sampling, uncertainty sampling, variance reduction or expected error reduction methods.

The above method allows for the training and improvement of a state model that improves the modeling uncertainties at a variety of operational feature points on the fly.

Claims (16)

Computer-implemented method for training a state model for determining a state of an electrical energy store of a device operated therewith, with the following steps: - Providing a data-based status model, which assigns a modeled status variable, in particular an aging status of the energy store, and a status uncertainty measure to an operating characteristic point; - Providing a database with profiles of real operating feature points from a plurality of energy stores for consecutive evaluation periods; - Training or updating the state model depending on at least one training data set, wherein the at least one training data set is generated with the following steps: ◯ Determination of a measure of improvement for each of the current operating characteristic points of the energy storage devices determined for the current evaluation period, with the measure of improvement indicating how state uncertainties improve at all real operating characteristic points when a state variable is determined for the respective current operating characteristic point; ◯ selecting at least one operational attribute point with the best measure of improvement; ◯ Determination of a corresponding state variable at the at least one selected operating characteristic point as a label; ◯ Generate the at least one training data set from the at least one selected operating characteristic point and the determined corresponding state variable. procedure after claim 1 , wherein the measure of improvement for each of the current operating feature points is determined by a simulation by updating the provided status model with a simulation training data set that is created for the respective current operating feature point, in order to obtain a simulation status model, with simulation status Uncertainties for each of the operating feature points of the progression of operating feature points of real energy storage devices are determined using the simulation state model, with the degree of improvement being determined as a function of the simulation state uncertainties. procedure after claim 2 , wherein the measure of improvement is determined as a sum of the uncertainties of the simulation state, wherein in particular at least one of the uncertainties of the simulation state is reduced using at least one predetermined rule based on domain knowledge before the measure of improvement is determined. procedure after claim 3 , wherein the reduction of the simulation state uncertainties at the selected operating characteristic points is effected by at least one rule which depends on a temporal development of the operating characteristic points of the respective energy store and simulation State uncertainties of operating feature points of the respective energy store to other evaluation periods than the evaluation periods of the selected operating feature points. procedure after claim 3 , wherein the simulation state uncertainties at the selected operating feature points are reduced by limiting the simulation state uncertainty at each of the selected operating feature points to an upper or lower limit of a confidence interval, which is reduced by the state uncertainty to a dem relevant operating characteristic point is determined in the preceding evaluation period, and/or that the simulation state uncertainty at each of the selected operating characteristic points is limited to a lower or upper limit of a confidence interval, which is determined by the state uncertainty at an evaluation period following the relevant operating characteristic point. Procedure according to one of claims 3 until 5 , wherein the simulation state uncertainties at the selected operating feature points are reduced in that the simulation state uncertainty at each of the selected operating feature points is limited with an upper and a lower limit of a corresponding confidence interval, which is obtained by interpolating upper limits of Confidence intervals of the operating characteristic point in question preceding and subsequent evaluation periods is determined. Procedure according to one of Claims 1 until 6 , wherein the determination of the corresponding state variable is carried out by a statistical method, in particular in connection with rules based on domain knowledge to reduce state uncertainties. Procedure according to one of Claims 1 until 6 , wherein the corresponding state variable is determined by requesting the device assigned to the at least one selected operating feature point to measure the state variable, with the measured state variable being used after the measurement to create the training data set. Procedure according to one of Claims 1 until 8th , wherein the state model corresponds to a state of health model for providing a state of health depending on the operating characteristics or a state of charge model for providing a state of charge depending on the operating characteristics of the energy store. Procedure according to one of Claims 1 until 9 , the operating characteristics being derived from time series of continuously recorded operating variables, the operating characteristics being determined for successive evaluation periods, a plurality of operating characteristics defining an operating characteristic point for a specific evaluation period of a specific energy store. procedure after claim 10 , operating variables being transmitted from a large number of technical devices to a central unit which is in communication with them, the operating characteristics being determined from the operating variables for each evaluation period, the method being carried out in a central unit. Procedure according to one of Claims 1 until 11 , wherein model parameters of the state model are transmitted to the plurality of devices with the energy stores, the devices being operated using the state model. Procedure according to one of Claims 1 until 12 , wherein the energy store (41) is used to operate a device such as a motor vehicle, a pedelec, an aircraft, in particular a drone, a machine tool, a consumer electronics device such as a mobile phone, an autonomous robot and/or a household appliance. Device for training a status model for determining a status of an electrical energy storage device, the device being designed to: - provide a data-based status model which assigns a modeled status variable, in particular an aging status of the energy storage device, to an operating characteristic point; - Providing a database with profiles of real operating feature points from a plurality of energy stores for consecutive evaluation periods; - Training or updating the status model depending on at least one training data set, wherein the at least one training data set is generated with the following steps: ◯ Determination of an improvement measure for each of the current operating characteristic points of the energy storage devices determined for the current evaluation period, the improvement measure indicating how status uncertainties change improve at all real operating feature points when determining a state variable for the respective current operating feature point; ◯ Select at least one operational characteristic spunkts with the best measure of improvement; ◯ Determination of a corresponding state variable at the at least one selected operating characteristic point as a label; ◯ Generate the at least one training data set from the at least one selected operating characteristic point and the determined corresponding state variable. Computer program product comprising instructions which, when the program is executed, cause at least one data processing device to carry out the steps of the method according to one of the Claims 1 until 13 to execute. Machine-readable storage medium, comprising instructions which, when executed by at least one data processing device, cause the latter to carry out the steps of the method according to one of Claims 1 until 13 to execute.

Images