Method and apparatus for providing virtual sensor using machine learning

Abstract

The invention relates to a method for providing a virtual sensor which is designed to determine a state variable (x) which characterizes a state of a technical system (200, 300) on the basis of an operating variable curve (u) of the technical system (200, 300), in particular on the basis of measurements at the technical system (200, 300), training data (D) being provided, the training data comprises empirically (in particular on the basis of the measurement)-determined pairs of the operating variable temporal curve (u (t)) and the associated state variable curve (x (t)), a first model (h1) is determined on the basis of the training data (D), said first model characterizing the evolution of the state variable (x (t)) depending on the operating variable (u (t)) (and the state variable (x (t))), and a second model (h2) characterizing the evolution of the state variable (x (t)) depending on the operating variable (u (t)) (and the state variable (x (t))), wherein the first model (h1) comprises a machine learning system (g), and wherein a second model (h2) is determined from the first model (h1) in such a way that the machine learning system (g) is replaced with an algebraic expression (g) determined by means of a sign regression of the learning system (g), and wherein a virtual sensor is provided in such a way that the machine learning system (g) is compared with the algebraic expression (g). The operating variable (x) is determined by means of the second model (h2) from the state variable (u).

Description

Methods and devices for providing virtual sensors using machine learning Technical Field

[0001] The present invention relates to a method for providing a virtual sensor, a virtual sensor, a method for operating a virtual sensor, a computer program, a machine-readable storage medium, and a training device. Background Technology

[0002] A method for providing an aging state model is known from DE 10 2022 200 022 A1, which is used to determine the modeled aging state (SOH) of a battery cell. The method comprises the following steps: • Providing a training dataset that assigns empirically determined aging states to operating parameter change curves of one or more operating parameters at specific time points; • Providing a data-based aging model constructed using a set of neural differential equations to model the internal electrochemical state of an energy storage device, wherein the internal state can be mapped to an aging state, wherein at least one differential equation in the set of equations comprises a sum of deterministic model terms and data-based correction terms, the correction terms being formed by a data-based correction model; • Training the data-based aging model based on the training dataset to determine the model parameters of the data-based correction model. It is also known that the aging state model is then used as a virtual sensor.

[0003] Therefore, a method for determining the modeled state of aging (SOH) of a battery cell at a specific time point is also known from the aforementioned literature, comprising the following steps: • providing operating parameter variation curves for one or more operating parameters; • providing a trained data-based state of aging model based on a set of neural differential equations and trained to assign the operating parameter variation curves to the state of aging; • determining or predicting the modeled state of aging by evaluating the state of aging model using the operating parameter variation curves in step sizes.

[0004] The neural ordinary differential equations are known from Chen et al.'s Neural Ordinary Differential Equations, https: / / arxiv.org / pdf / 1806.07366.pdf.

[0005] A method for creating physical equations using symbolic regression with the help of a trained machine learning system is known from DE 10 2020 215 138 A1. Summary of the Invention

[0006] The advantages of the invention: Virtual sensors typically refer to curves that determine the change of parameters characterizing the state of a technical system based on the change curves of operating parameters (determined by measurement).

[0007] Compared to existing technologies, the present invention, having the features of independent claim 1, has the advantage of particularly powerful virtual sensor performance. In particular, the virtual sensor possesses robust extrapolation capabilities.

[0008] Other aspects of the invention are the subject of the dependent claims. Advantageous extensions are the subject of the dependent claims.

[0009] In a first aspect, the invention relates to a method for providing a virtual sensor configured to determine state parameters characterizing the state of a technical system based on, in particular, curves of change of operating parameters determined by measurements at the technical system. The method includes providing training data comprising pairs of time-varying curves of operating parameters and corresponding curves of change of state parameters, determined empirically (i.e., in particular by measurements). A first model is determined based on this training data, characterizing the state parameters depending on the evolution of operating parameters (and state parameters (x(t))). The first model includes a machine learning system. A second model is determined from the first model by replacing the machine learning system with an algebraic expression determined by symbolic regression using the machine learning system. The virtual sensor is provided such that it determines the operating parameters based on the state parameters using the second model.

[0010] The advantages of this are improved extrapolation capabilities. Furthermore, the algebraic expression has lower complexity than machine learning systems. This reduced complexity leads to more resource-efficient virtual sensors.

[0011] Some extended schemes specify that the first model also includes a physical model, and specifically as a link between the physical model and the machine learning system, such as a sum or product. This allows the introduction of prior knowledge about the desired behavior, which can lead to more robust and extrapolating virtual sensors.

[0012] In these extended schemes, it is possible to check whether the algebraic expression corresponds to the neutral element of the link, and to provide the virtual sensor based on whether the algebraic expression corresponds to the neutral element. This allows for a particularly simple check of whether prior modeling of the physical model is applicable.

[0013] It can then be stipulated that the virtual sensor is provided only if the algebraic expression corresponds to the neutral element. This ensures, in a particularly simple way, that the provided sensor corresponds to the correct prior model.

[0014] Advantageously, in these aspects, it is specified that the first and / or second models describe the evolution of the state parameters as neural differential equations. Such neural differential equations are known, for example, by Chen et al., Neural Ordinary Differential Equations, 2019, https: / / arxiv.org / pdf / 1806.07366.pdf. These neural differential equations are particularly suitable for modeling dynamic systems and therefore especially for virtual sensors due to their high model accuracy and robustness. The state parameters can then be advantageously determined by integrating the neural differential equations.

[0015] In one exemplary application, the technical system is an energy storage device, particularly a battery, and the state parameters specifically characterize the aging state of the energy storage device.

[0016] In another exemplary application, the technical system is an electric motor, and the state parameters are in particular the temperature of the motor's rotor.

[0017] In another aspect, the present invention relates to a virtual sensor configured to determine state parameters characterizing the state of a technical system based on, in particular, on a change curve of operating parameters determined by measurements at the technical system, the virtual sensor being provided using the aforementioned method.

[0018] In another aspect, the present invention relates to a method for operating a virtual sensor, wherein the operating parameter variation curves of the technical system are determined, in particular, measured, and state parameters characterizing the state of the technical system are determined. Attached Figure Description

[0019] Embodiments of the present invention will now be explained in more detail with reference to the accompanying drawings. In the drawings: FIG1 schematically illustrates an exemplary first application of a virtual sensor; FIG2 schematically illustrates an exemplary second application of a virtual sensor; FIG3 illustrates, in flowchart form, one embodiment of a method for providing a virtual sensor, which is then optionally operated; FIG4 schematically illustrates an apparatus for performing the method of providing a virtual sensor. Detailed Implementation

[0020] Figure 1 illustrates a motor vehicle (100) having a sensor (110) that, in embodiments, measures the voltage applied to an electrical energy storage device (200), particularly a battery (which may include at least one battery cell) or a fuel cell. The electrical energy storage device (200) can, for example, be used to supply energy to a drive unit (130) of the motor vehicle (100). The measurements from the sensor (110) are transmitted to a computer (120), which determines a parameter from the virtual sensor provided using the method shown in Figure 3. This parameter, in some embodiments, determines the State of Health (SOH) of the electrical energy storage device (200). Based on this determined SOH, the drive unit (130) can be operated accordingly. In some embodiments, the computer (120) may be located within the motor vehicle (100). In alternative embodiments, the computer (120) may be located outside the motor vehicle (100).

[0021] For example, the drive unit (130) can be manipulated with an optimal operating strategy.

[0022] Alternatively or additionally, maintenance measures can be predictively displayed at the energy storage device (200) based on the predicted aging condition.

[0023] In other embodiments, the energy storage device (200) is composed of a battery, which can optimally control the charging strategy of the battery.

[0024] Figure 2 illustrates another embodiment for the application of a virtual sensor. The motor vehicle (100) has a sensor (111) which, in this embodiment, measures the voltage applied to a motor (300), particularly a generator or electric motor, especially the voltage applied to the stator windings or the rotor (310) windings of the motor (300). The motor (300) may, for example, be part of a drive unit (130) shown in Figure 1, and receives energy fed from an electrical energy storage device (200), also shown in Figure 1. The measurements from the sensor (111) are transmitted to a computer (120), which uses the virtual sensor provided by the method shown in Figure 3 to determine a parameter characterizing the temperature in the rotor (310) of the motor (300). Based on this determined temperature, the actuators (135) of the motor vehicle (100) can be manipulated accordingly. In some embodiments, the computer (120) may be located inside the motor vehicle (100). In alternative embodiments, the computer (120) may be located outside the motor vehicle (100).

[0025] In some implementations, the cooling strategy of the motor (300) can be adjusted according to the determined temperature.

[0026] Figure 3 illustrates the flow of an embodiment for providing and subsequently operating virtual sensors using a flowchart.

[0027] First, a virtual sensor is provided. For this purpose, (1000) training data is provided. .

[0028] The training data includes time-varying curves of the operating parameters (u) of the technical system, and the corresponding state parameters (x) characterizing the state of the technical system. In the embodiments shown in Figures 1 and 2, the operating parameters (u) correspond to parameters determined by sensors (110, 111). These related values ​​from the changing curves of the operating parameters (u) and state parameters (x) can, for example, be determined at a test bench.

[0029] Then, a first model (h1) is provided (1100), which in some embodiments has the following form: .

[0030] The first model (h1) includes, in some implementations, a physical model (f), such as a 1-node model or a 2-node model (e.g., a rotor temperature model) or an electrochemical cell model. This physical model (f) determines an approximation of the state vector x based on an operating parameter (u) (in the case of the cell shown in Figure 1, this could be, for example, a curve showing the variation of the operating parameter (u) over a period of time with respect to the current intensity required during battery charging or discharging). This physical model is parameterized with a parameter θ.

[0031] The first model (h1) also includes a machine learning system (g), such as an artificial neural network or a Gaussian process. The machine learning system (g) describes the data-based part of the modeling, which is used to improve the pure physics modeling. This machine learning system is parameterized with parameter φ.

[0032] The physical model (f) and the machine learning system (g) are thus linked by addition.

[0033] In some implementations, the physical model (f) and the machine learning system (g) are linked by multiplication, and the first model (h1) has the following form: .

[0034] In some implementations, the physical model (f) and the machine learning system (g) are linked exponentially, and the first model (h1) has the following form: .

[0035] In some implementations, the first model (h1) is entirely provided by the machine learning system: .

[0036] Now, in some implementations, the first model (h1) is trained in a known manner using training data (D) as a system of neural differential equations, i.e., according to the equations .

[0037] The state parameter (x) describes the state vector in some implementations and includes physically measurable or interpretable parameters, such as the temperature of the rotor (310) of the motor (300), or characterizing the thickness of the SEI or the amount of cyclic lithium in the battery (200).

[0038] Now, perform symbolic regression (1200) on the trained machine learning system (g) and obtain the algebraic expression (g) from it. a The algebraic expression is determined from a pre-defined number of pre-defined functional blocks with pre-defined links to reproduce the input-output behavior of the machine learning system (g) as well as possible. The data required to perform symbolic regression can be generated very easily, for example, by providing data such as training data D and feeding it to the machine learning system (g), and then determining the relevant output parameters of the machine learning system (g) with low input.

[0039] In some implementations, the first model (h1) is given by the link between the physical model (f) and the machine learning system (g) (see equations (A), (B) and (C) above). Now examine the algebraic expression (g) of (1250). a ) whether it corresponds to the neutral element of the link; that is, "0" in the case of addition, "1" in the case of multiplication, "1" in the case of exponentiation, etc. If not, stop (1251) the provision of the virtual sensor, otherwise proceed to step (1300).

[0040] In this step, a second model is provided. It is derived from the first model (h1) by replacing the machine learning system (g) with an algebraic expression (g). a In some implementations, the resulting expression can be simplified, for example, by omitting the algebraic expression (g) if it is equal to or approximately equal to the neutral element within a pre-given tolerance. a (link).

[0041] Now, a virtual sensor (1400) including the second model (h2) is provided. The virtual sensor is configured to transmit the change curve of the operating parameter (u) supplied to it to the second model, and to determine the change curve of the state parameter (x) by integrating equation (2) (in some embodiments, numerical integration), and to provide it at the output of the virtual sensor.

[0042] This concludes the methods for providing virtual sensors.

[0043] Subsequently, in some embodiments, it may be specified that the virtual sensor is now running (1500), as shown in the embodiments in Figures 1 and 2.

[0044] Figure 4 illustrates a training device (400) configured to perform the method shown in Figure 3 (except for optional step 1500). In some embodiments, the method is implemented as a computer program stored on a computer-readable storage medium (420) and contains instructions that, when the computer (410) runs the computer program, cause the computer (410) of the training device (400) to execute the method.

Claims

1. A method for providing a virtual sensor, the virtual sensor being configured to determine a state parameter (x) characterizing the state of a technical system (200, 300) based on determined operating parameter variation curves (u), wherein training data (D) is provided, the training data comprising pairs of operating parameter time variation curves (u(t)) and their corresponding state parameter variation curves (x(t)) determined empirically (i.e., especially based on measurements), wherein a first model (h1) is determined based on this training data (D), the first model characterizing the state parameter (x(t)) depending on the evolution of the operating parameter (u(t)) (and the state parameter (x(t))), wherein the first model (h1) includes a machine learning system (g), wherein a second model (h2) is determined from the first model (h1) by replacing the machine learning system (g) with an algebraic expression (g) determined by symbolic regression of the machine learning system (g). a And in which a virtual sensor is provided, such that it determines the operating parameter (x) based on the state parameter (u) using the second model (h2).

2. The method of claim 1, wherein the first model (h1) further includes a physical model (f), and in particular serves as a link, for example, and / or product, between the physical model (f) and the machine learning system (g).

3. The method according to claim 2, wherein the algebraic expression (g) is checked. a Whether ) corresponds to the neutral element of the link, and according to the algebraic expression (g) a Whether it corresponds to the neutral element to provide the virtual sensor.

4. The method of claim 3, wherein the algebraic expression (g) is true only if the algebraic expression (g) is true. a The virtual sensor is provided only when it corresponds to the neutral element.

5. The method according to any one of the preceding claims, wherein the first model (h1) and / or the second model (h2) describe the evolution of the state parameter (x(t)) as a neural differential equation.

6. The method according to any one of the preceding claims, wherein the technical system is an energy storage device (200), in particular a battery, and the state parameter (x(t)) in particular characterizes the aging state (SOH) of the energy storage device (200).

7. The method according to any one of claims 1 to 6, wherein the technical system is an electric motor (300), and the state parameter (x(t)) characterizes the temperature of the rotor (310) of the electric motor (300).

8. A virtual sensor configured to determine a state parameter (x) characterizing the state of the technical system (200, 300) based on a change curve (u) of the operating parameters of the technical system, the virtual sensor being provided using the method according to any one of claims 1 to 6.

9. A method for operating a virtual sensor according to claim 8, wherein a change curve (u) of the operating parameters of the technical system (200, 300) is determined, in particular, measured, and a state parameter (x) characterizing the state of the technical system (200, 300) is determined.

10. A computer program configured to contain instructions that, when executed by a computer (410), cause the computer to perform the method according to any one of claims 1 to 7 or 9.

11. A machine-readable storage medium (420) having a computer program as claimed in claim 10 stored thereon.

12. A training device (400) comprising a computer (410), said training device being configured to perform the method according to any one of claims 1 to 7.

Images