Method and device for generating virtual sensor system

Abstract

To provide an automated method for obtaining a state variable characterizing the state of a technical system.SOLUTION: The method includes steps of: providing a first model that obtains an estimated value of a state variable from a measured variable; providing measured pairs of measured variables and respectively assigned state variables; adapting parameters characterizing the characteristics of the first model depending on the measured pairs; providing a machine learning system, which is linked with the first model to produce an overall model and is configured to obtain an overall estimated value of the state variable from the measured variable; training the machine learning system so as to minimize an error function that includes a deviation between the state variable and the overall estimated value of the state variable; obtaining an approximation of the machine learning system from the machine learning system by means of symbolic regression; and providing a link of the first model with the symbolic regression as a generated system.SELECTED DRAWING: None

Description

[Technical field]

[0001] The present invention relates to a method, a virtual sensor system, a computer program and a machine-readable storage medium for the automated generation of a system for determining state variables characterizing the state of a technical system. [Background technology]

[0002] Prior Art Physical models can be used to model the physical interactions for virtual sensors. However, for all applications of virtual sensors, physical models with sufficient accuracy are unknown. Machine learning methods can fill this void.

[0003] Thus, from DE 10 2017 218 922 A1 a method is known for generating virtual sensors by means of an autoencoder.

[0004] However, virtual sensor systems based on this type of machine learning system are relatively new and therefore under-tested. Moreover, the visibility of this type of machine learning system is often difficult, which is an obstacle to its use in safety-critical applications.

[0005] From DE 10 2020215 138 A1 a method is known for generating physical equations using machine learning methods. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] DE 102017218922 [Patent Document 2] DE 102020215138 A1 Summary of the Invention [Problem to be solved by the invention]

[0007] Advantages of the invention In contrast, the present invention has the advantage that the generated virtual sensor system combines the high modeling quality of a virtual sensor system based on machine learning methods with good interpretability and thus the protection possibilities of a virtual sensor system based on physical models.

[0008] Further aspects of the invention are the subject of the parallel independent claims. Advantageous developments are the subject of the dependent claims. [Means for solving the problem]

[0009] Disclosure of the Invention In a first aspect, the invention relates to a method for the automated generation of a system for determining state variables characterizing a state of a technical system in dependence on measurement variables, i.e. measured variables characterizing a second state of the technical system (in other words a method for the automated generation of a virtual sensor system), said method comprising the following steps: - providing a first model, in particular a physical model, which is adapted to obtain estimates of state variables from measured variables; - providing measured pairs of measurement variables and respective assigned state variables; parameters characterizing the properties of the first model are or are adapted in dependence on the measured pairs, in particular such that the deviation between estimates determined from the measured variables by means of the first model and the state variables is minimized; a machine learning system is provided, in particular a Gaussian process or a neural network, the machine learning system being combined with the first model to generate a global model, the global model being configured to determine global estimates of the state variables from the measurement variables; a machine learning system is or is trained to minimize an error function that includes the deviation between the state variables and a global estimate of the state variables; - determining an approximation of the machine learning system from the machine learning system using symbolic regression, in particular such that the approximation of the machine learning system approximates the machine learning system as well as possible; the combination of the first model and the symbolic regression is provided as a generated system; Includes.

[0010] This provides the advantages discussed above.

[0011] In one development, for symbolic regression using a machine learning system, pairs of input data and output data belonging to the machine learning system are generated and an approximation is determined using a regression of these input data and output data.

[0012] In other words, input data is provided and the machine learning system is used to provide output data of belonging, such that pairs of input data and output data of belonging can be easily generated in large numbers, which makes symbolic regression particularly easy.

[0013] Alternatively or additionally, in the symbolic regression, regression candidates and the respective adaptation quality associated therewith can be proposed to the user, and in particular the selected regression candidates from the user can be received and taken over as an approximation of the machine learning system. This allows existing expertise to be introduced into the generated virtual sensor system in a particularly simple manner. In particular, here, it is possible to propose only regression candidates whose adaptation quality associated therewith exceeds a predefinable minimum adaptation quality.

[0014] In these methods it may be assumed that the technical system is an electric machine and / or the measurement variable is determined by means of a voltage sensor, a temperature sensor or a rotation speed sensor.

[0015] Alternatively, it may be envisaged that the technical system is an energy accumulator, in particular a battery or a fuel cell system, and / or that the measurement variable is determined by means of a voltage sensor or a temperature sensor.

[0016] Alternatively, it may be envisaged that the technical system is a braking system and / or a steering system of a motor vehicle and / or that the measurement variable is determined by means of a voltage sensor, a temperature sensor, a rotational speed sensor or a steering angle sensor.

[0017] In one development, it may be provided that, with the aid of the generated system, state variables are determined in the technical system as a function of measurement data, which characterize a state of the technical system.

[0018] In a further aspect, the invention relates to a virtual sensor system for determining state variables characterizing a state of a technical system in dependence on measurement variables characterizing a second state of said technical system, the virtual sensor system being generated by one of the methods described above for determining the state variables in dependence on the measurement variables.

[0019] In a further aspect, the present invention relates to a computer program configured to perform one of the aforementioned methods as indicated above (i.e. a computer program comprising instructions for causing a computer to perform one of these methods when the computer program is executed by the computer), and to a machine-readable storage medium on which the computer program is stored.

[0020] In the following, embodiments of the invention will be explained in more detail with reference to the accompanying drawings. [Brief description of the drawings]

[0021] [Figure 1] FIG. 1 is a schematic diagram illustrating an exemplary embodiment for creating and using a virtual sensor system. [Diagram 2] 4 is a flowchart illustrating an exemplary flow of the method. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Description of the embodiment 1 shows a schematic diagram of an exemplary embodiment for implementing the invention. A technical system (100) is shown, which in some embodiments is a fuel cell system. A sensor (101) is arranged in the technical system (100), which in some embodiments is a temperature sensor for determining the gas temperature inside the fuel cell system. Measurement variables measured by the sensor (101) are transmitted to a computer (200) in which a virtual sensor system (210) is implemented, and a system generated from the measurement variables with a second computer (300) transmits further state variables describing the state of the technical system, for example the pressure of the gas in the fuel cell system. The technical system (100) further comprises an actuator (120), for example a valve or a pump, which is actuated depending on the state variables determined with the generated system, which characterize the state of the technical system (100).

[0023] The method for generating the generated system, illustratively illustrated in FIG. 2, is implemented in an embodiment as a computer program stored in a machine-readable storage medium (301) of a second computer (300).

[0024] FIG. 2 illustrates in a flow chart the flow of an exemplary embodiment for generating a generated system.

[0025] First, a physical model is provided (1000), which describes sufficiently well the interaction between the measurement variables and the state variables of the technical system to be characterized. In other words, the physical model receives the measurement variables at its input side and provides from its output side variables which represent in a good approximation the state variables of the technical system. In other words, the output side of the physical model provides estimates of these variables.

[0026] This physical system can be parameterized by parameters in some embodiments. Then, pairs (1100) of measurement variables determined by the sensors (101) and associated state variables that can be measured by additional sensor systems in the technical system (100), for example in the development phase, are provided. Insofar as the physical system can be parameterized by parameters, these parameters are adapted (120) in such a way that the estimates of the state variables for which the physical model is determined in dependence on the measurement variables correspond as well as possible to the respectively associated state variables included in the pair.

[0027] A machine learning system then provides, in some embodiments, a neural network that is combined with the physical model and, for example, added or multiplied, and the machine learning system is trained (1300) so that the model thus combined determines as well as possible the respective state variables from the measurement variables. In other words, if the combined model is provided with measurement variables on the input side, it is provided with global estimates of the respective state variables on the output side. This training is usually performed, for example, by minimizing a cost function that evaluates the deviation between the global estimates of the state variables and the provided state variables.

[0028] Here, input data, in an embodiment the measurement data mentioned above, are fed to the machine learning system, and at its output side the associated respective output data is determined (1400).

[0029] Based on the input data and output data, a machine learning system is approximated (1500) using symbolic regression.

[0030] In some embodiments, the symbolic regression includes providing 1501 regression candidates from a preconfigurable function space, in particular regression candidates having a complexity less than a preconfigurable maximum complexity.

[0031] These regression candidates are adapted (1502) to the input data and the associated output data, i.e. the parameters of the regression candidates are adapted so that the functional course of the regression candidates corresponds as well as possible to the pairs. As a measure of how well the courses correspond to these pairs, a respective quality of adaptation is determined, which in an embodiment is the higher the value, the better the course corresponds to the pair (e.g. the known x 2 the inverse of a function).

[0032] In some embodiments, regression candidates whose matching quality is worse than a pre-configurable minimum matching quality are removed 1503 .

[0033] The regression candidates are now passed on as approximations to the machine learning system (1504). In some embodiments, this is done by selecting the regression candidate with the best fitting quality. In other preferred embodiments, the regression candidates are provided to a user along with their respective fitting qualities (1505), and a selection from the user is received and provided as the approximation (1506).

[0034] The combination of the physical model and the approximation is provided as a virtual sensor, i.e. a system for determining state variables characterizing the state of the technical system in dependence on the measurement variables (1600).

[0035] A technical system (100) as shown in FIG. 1 can then be operated (1700) using the so obtained virtual sensor (210).

[0036] This ends the method.

Claims

1. A method for the automated generation of a system for determining state variables characterizing a state of a technical system (100) in dependence on measurement variables characterizing a second state of said technical system (100), said method comprising the following steps: - providing a first model, said first model being configured to determine estimates of said state variables from said measurement variables; - providing measured pairs of said measurement variables and respective assigned state variables; - parameters characterizing the properties of said first model are adapted in dependence on the measured pairs; - providing a machine learning system, the machine learning system being combined with the first model to produce a global model, the global model being configured to determine global estimates of the state variables from the measurement variables; - the machine learning system is trained to minimize an error function that includes the deviation between the state variables and a global estimate of the state variables; - determining an approximation of the machine learning system using symbolic regression from the machine learning system; - providing the combination of said first model and said symbolic regression as a generated system; A method comprising:

2. The method of claim 1 , wherein for the symbolic regression using the machine learning system, pairs of input data and output data belonging to the machine learning system are generated, and the approximation is obtained using a regression of the input data and the output data.

3. 3. The method according to claim 1 or 2, wherein during the symbolic regression, regression candidates and their respective associated adaptation qualities are proposed, in particular selected regression candidates from a user are received and taken over as approximations for the machine learning system.

4. 4. The method according to claim 1, wherein the technical system (100) is an electric machine and / or the measured variable is determined by means of a voltage sensor, a temperature sensor or a rotation speed sensor.

5. 4. The method according to claim 1, wherein the technical system (100) is an energy accumulator, in particular a battery or a fuel cell system, and / or the measured variable is determined by means of a voltage sensor or a temperature sensor.

6. 4. The method according to claim 1, wherein the technical system (100) is a braking system and / or a steering system of a motor vehicle and / or the measurement variables are determined by means of a voltage sensor, a temperature sensor, a rotational speed sensor or a steering angle sensor.

7. 7. The method according to claim 1, further comprising determining, in the technical system (100), state variables characterizing a state of the technical system as a function of measurement data, with the aid of the generated system.

8. 7. A virtual sensor system for determining state variables characterizing a state of a technical system (100) in dependence on measurement variables characterizing a second state of said technical system (100), the virtual sensor system being generated using a method according to any one of claims 1 to 6 for determining said state variables in dependence on said measurement variables.

9. A computer program adapted to carry out the method according to any one of claims 1 to 6.

10. 10. A machine-readable storage medium having stored thereon the computer program of claim 9.

Images