CN-113298257-B - Method and device for creating a model of a technical system from measurements

Abstract

Methods and apparatus for creating a model of a technical system from measurements. The invention relates to a method for creating a model of a technical system from measured sensor data of the technical system, comprising the step of initializing a symbolic regression problem (English: symbolic regression problem). A list of mathematical functions is determined, said list comprising at least one linear and/or non-linear function and/or at least one-dimensional parametrizable characteristic. The at least one-dimensional characteristic is implemented by a smooth mesh regression (SGR) model. The problem of symbolic regression is solved by means of genetic algorithms.

Inventors

• A. Junginger
• H. Ulmer
• J. S. Buchner
• P. ENGEL
• S. Bob Lester

Assignees

• 罗伯特·博世有限公司

Dates

Publication Date

20260505

Application Date

20210223

Priority Date

20200224

Claims (13)

1. 1. A computer-implemented method for creating a model of a technical system from measured sensor data of the technical system, the method comprising the steps of: providing the measured sensor data, which is detected by means of different sensors; the model generator is initialized and the model is set up to the model, Wherein a list of mathematical functions is provided, said list comprising at least one linear and/or non-linear function and at least one-dimensional parametrizable characteristic, Wherein at least one-dimensional characteristic curve is implemented by a smooth mesh regression model SGR; Wherein the genetic algorithm combines a plurality of models with mathematical functions from the list, Wherein the combined models are each optimized such that the models can be based on further measured sensor data Calculating measured sensor parameters One of them is that, Wherein the genetic algorithm evaluates the trained model according to an fitness function, Wherein the fitness function is related to the efficiency of the model to be evaluated and to the complexity of the model, and The best model in terms of fitness function is output according to its complexity.
2. 2. The method of claim 1, wherein the technical system is an electric motor or an internal combustion engine.
3. 3. The method according to claim 1 or 2, wherein a second order local optimization method is used for optimizing the combined model in terms of its efficiency.
4. 4. The method according to any one of claims 1 to 2, wherein a model is selected from the outputted models by statistical analysis of data residuals.
5. 5. The method according to any of the preceding claims 1 to 2, wherein a user selects one of the outputted models according to the fitness function and the complexity.
6. 6. Method according to any of the preceding claims 1 to 2, wherein the range of input values of the combined model is adapted at optimization time.
7. 7. The method of claim 6, wherein for each iteration step, the range of input values of the combined model is adapted at the time of optimization.
8. 8. The method according to any of the preceding claims 1 to 2, wherein a part of the mathematical functions of the list are pre-defined by a user.
9. 9. The method according to claim 8, wherein a part of the mathematical functions of the list is predefined by a user based on his a priori knowledge about the technical system and/or the measured sensor parameters.
10. 10. The method according to claim 8, wherein one of the models output is stored on a control device, wherein the control device determines a control variable for the technical system using the stored model.
11. 11. A computer program product having a computer program comprising instructions which, when executed by a computer, cause the computer to perform the method according to any of the preceding claims 1 to 10.
12. 12. A machine readable storage element having stored thereon a computer program comprising instructions which, when executed by a computer, cause the computer to perform the method according to any of the preceding claims 1 to 10.
13. 13. An apparatus for creating a model of a technical system from measurements, the apparatus being set up to perform the method according to any one of claims 1 to 10.

Description

Method and device for creating a model of a technical system from measurements Technical Field The present invention relates to a method and a device for creating a model of a technical system from measurements, a computer program and a machine-readable storage medium. Background In order to develop tuning strategies for components of a technical or physical system, mathematical models describing these components are required. However, the application of these regulation strategies in industry standards by using them in embedded control units has so far only been possible in a very limited way. These limitations are caused by the low interpretability of these models, low accuracy, mostly high computational performance, memory consumption, and low universality of these models to similar systems. However, popular modeling techniques, such as neural networks or gaussian process models, can only overcome some of the just-mentioned limitations. However, it is desirable to overcome all the mentioned limitations and to be able to provide a model by means of which a safe and accurate regulation strategy can then be developed and implemented. Priber, U, "Smoothed Grid regression" (Proceedings Workshop Fuzzy System, volume 13, 2003) discloses a smooth mesh regression (Smoothed Grid Regression, SGR) model. Disclosure of Invention Advantages of the invention For this reason, the inventors propose a method which automatically determines from the measured sensor data a model, in particular based on data, of the technical system itself or of a component of the technical system and which overcomes the above-mentioned limitations. The method is characterized in that it produces a particularly fast, simple and accurate model on the Pareto Front (Pareto-Front). Disclosure of Invention In a first aspect, the invention relates to a computer-implemented method for creating a model of a technical system, such as an electric motor or an internal combustion engine. The model may be a data-based or mathematical model. The method comprises the step of providing measured sensor data, which sensor data is detected by different sensors. The different sensors may be structurally identical or structurally different sensors. Furthermore, the sensors can characterize different characteristics of the technical system, such as torque, power or current consumption. This is followed by a step of initializing the model generator, in particular initializing the signed regression problem (English: symbolic regression problem). In addition, a list of mathematical functions is initialized. The list comprises at least one linear and/or nonlinear function and/or at least one-dimensional parametrizable characteristic curve (english: curve). The mathematical functions of the list are also referred to as basis functions in the following. The list may also include trigonometric and/or exponential and/or logarithmic functions. It should be noted that at least one-dimensional characteristic curves are implemented by means of a smooth mesh regression model (SGR), in particular in a model. The advantage of SGR is its direct interpretability. Additionally or alternatively, the characteristic curves in the list may already be defined by the SGR. This is followed by a step of solving the problem of symbolic regression by means of a genetic algorithm (English: genetic algorithm). In this case, a plurality of models from the list are combined with mathematical functions, in particular by genetic algorithms, and then the combined models are trained by means of an optimization method, in particular in terms of efficiency or in terms of cost functions, and the trained models are evaluated in accordance with fitness functions. The fitness function is related to the efficiency of the model to be evaluated and to the complexity of the model. It should be noted that these just mentioned steps for solving the symbolic regression problem may be performed several times in succession. This is followed by a step of outputting the best model according to its complexity in view of the fitness function. The best model is the one with the highest fitness for a given complexity compared to other models with the same complexity. A one-dimensional characteristic can be understood as a curve that maps an input parameter to an output parameter. The curve may be a (linear) interpolation consisting of a plurality of sampling points or support points. Sampling points or support points are optimized for the model. Similarly, the multidimensional characteristic is a bilinear interpolation. One-dimensional or two-dimensional characteristics may also be referred to as a one-dimensional or two-dimensional characteristic family. The input/output variables of the characteristic curve may have physical significance, for example, current strength or torque. Alternatively, these parameters may also be abstract nature. Symbolic regression can be understood as follows. Symbolic regression is a regre