Supervised machine learning of a computer-implemented method for performing a technical process, computer program and a computer-readable storage medium having a computer program for implementing the method
A supervised machine learning of a computer-implemented method performs a technical process in which a data pool is created, containing data sets with input data and output data that describes a correct process result. The data sets are divided into training and validation data sets. The method is trained in a training phase, wherein process parameters of the method are varied during repeated performances of the method. A trained method is checked in a validation phase by comparing the output data calculated with the trained method with the input data, with the output data describing the correct process result and calculating an actual probability of failure for the method as a result of the comparison. An empirical probability of failure is defined for the training phase, which is at most as large as a specified target probability of failure.
1 . A supervised machine learning of a computer-implemented method for performing a technical process in which output variables of the technical process resulting from input variables can be ascertained, the method comprises the following steps of:
a) creating a data pool, containing data sets with input data for the technical process and output data that is assigned to the input data and describes a correct process result;
b) dividing the data sets in the data pool into training data sets and validation data sets;
c) training the computer-implemented method using the training data sets in a training phase, wherein process parameters of the computer-implemented method are varied during repeated performances of the computer-implemented method;
d) checking a trained computer-implemented method using the validation data sets in a validation phase by comparing the output data calculated with the trained computer-implemented method with the input data, with the output data describing the correct process result and calculating an actual probability of failure for the computer-implemented method as a result of a comparison;
e) defining a safety factor related to a target probability of failure for the training phase, with which an empirical probability of failure is calculated, which is lower than the target probability of failure, wherein the empirical probability of failure is reduced by a confidence level compared to the target probability of failure; and
f) initiating the validation phase after it has been established in the training phase that the empirical probability of failure is not exceeded.
2 . The supervised machine learning according to claim 1 , wherein:
during the training phase, performances of the computer-implemented method within a predetermined observation interval are repeatedly checked by comparing the output data calculated with the input data in the performances of the computer-implemented method with the output data describing the correct process result and by calculating a current probability of failure for the computer-implemented method as a result of the comparison with respect to the predetermined observation interval; and
initiating the validation phase after it has been established that the current probability of failure of at least one said predetermined observation interval does not exceed the empirical probability of failure.
3 . The supervised machine learning according to claim 2 , which further comprises suppressing a variation of the process parameters of the computer-implemented method within the predetermined observation intervals.
4 . The supervised machine learning according to claim 1 , wherein the safety factor is defined as at least ⅓ and at most ⅔.
5 . The supervised machine learning according to claim 4 , wherein the safety factor is defined as ½.
6 . The supervised machine learning according to claim 1 , which further comprises estimating the confidence level for a given number of the validation data sets as a product of a critical factor and a factor consisting of a square root of a quotient of the empirical probability of failure in a numerator and a number of the validation data sets in a denominator.
7 . The supervised machine learning according to claim 6 , which further comprises calculating an estimated probability of failure for the given number of the validation data sets as a sum of the confidence level and the empirical probability of failure.
8 . The supervised machine learning according to claim 7 , which further comprises increasing the number of the validation data sets if a calculated estimated probability of failure is above the target probability of failure.
9 . The supervised machine learning according to claim 1 , wherein the training data sets are released for multiple use for method runs in the training phase.
10 . The supervised machine learning according to claim 1 , which further comprises taking into account a statistical significance of a result of the validation phase by defining a corrected probability of failure, which is lower than the target probability of failure, in the validation phase for comparison with an actual probability of failure.
11 . The supervised machine learning according to claim 10 , which further comprises calculating the corrected probability of failure by multiplying the target probability of failure by a factor resulting from subtracting the statistical significance from 1 and adding the statistical significance multiplied by a correction factor which is less than 1.
12 . A technical system for performing a technical process, the technical system comprising:
a system component having a computer, said system component of the technical system is configured to execute the computer-implemented method according to claim 1 .
13 . The technical system according to claim 12 , wherein the technical system is a railroad technical installation being a railroad crossing, wherein the computer-implemented method is configured at said railroad crossing to perform obstacle detection in a danger zone of said railroad crossing.
14 . A non-transitory computer-readable storage medium having computer-executable instructions for performing the computer-implemented method according to claim 1 .