Method and apparatus for providing a data-based system model and for checking a training state of the system model
A method is for providing training data for training a data-based system model for operating a technical system by defining a data point determined from input variables for determining at least one output variable depending on which the technical system is operating. The method includes providing training data that are determined with a scenario other than a real operation of the technical system, the training data are defined for data points determined from the input variables, capturing operational data points determined from the input variables in real-world operation of the technical system, and splitting the training data into training data points and validation data points. The method further includes determining a k-Nearest Neighbor tree from the training data points, and determining a first distribution of distance values of distances between each of the validation data points and a predetermined number of next training data points of the training data points.
1 . A method for training a data-based system model for operating a technical system, the data-based system model being configured to (i) receive a data point that is determined from input variables and (ii) determine at least one output variable that is used to operate the technical system, the method comprising:
providing training data that are determined with a scenario other than a real operation of the technical system, the training data being defined for data points determined from the input variables;
capturing operational data points determined from the input variables in real-world operation of the technical system;
splitting the training data into training data points and validation data points;
determining a k-Nearest Neighbor tree from the training data points;
determining a first distribution of first distance values of first distances between each of the validation data points and a predetermined number of next training data points of the training data points;
determining a second distribution of second distance values of second distances between each of the operational data points and the predetermined number of next training data points;
determining a distance dimension between the first distribution of the first distance values and the second distribution of the second distance values;
adding further training data to the training data depending on the distance dimension; and
training the data-based system model, using the training data, to operate the technical system.
2 . The method according to claim 1 , wherein the further training data is determined by augmentation, synthesis, bench measurements, or simulations.
3 . The method according to claim 1 , wherein Euclidean distances are determined as (i) the first distances between each of the validation data points and the predetermined number of next training data points, and/or (ii) the second distances between each of the operational data points and the predetermined number of next training data points.
4 . The method according to claim 1 , wherein the input variables comprise (i) at least one model-based and/or sensorially captured operational variable of the technical system and/or (ii) at least one characteristic variable determined as an aggregate variable of a time series of the at least one operational variable or as a statistical variable with respect to the at least one operational variable.
5 . The method according to claim 1 , wherein:
additional distance dimensions are determined for sub-areas of an entire data space of the data points of the training data and the operational data points,
the further training data is determined depending on the additional distance dimensions for the sub-areas, and
the training data is determined when at least one of the additional distance dimensions exceed a predetermined threshold.
6 . The method according to claim 1 , wherein:
the method is repeated in multiple replicates until the distance dimension falls below a predetermined threshold and/or a change in the distance dimension between successive repeats of the method falls below a predetermined change threshold, and
the multiple replicates include determining the second distribution, determining the distance dimension between the first distribution and the second distribution, and adding further training data to the training data that depends on the distance dimension.
7 . The method according to claim 1 , wherein the data-based system model is a non-parametric model.
8 . The method according to claim 7 , wherein the data-based system model is a neural network.
9 . A device comprising:
a processor configured to carry out the method according to claim 1 .
10 . A non-transitory machine-readable storage medium comprising instructions which, when executed by a computer, cause the computer to execute the method according to claim 1 .