Method for training neural networks
The invention relates to a system and a method for training a number of neural networks, by determining a first training data record, wherein the training data have a particular accuracy, generating a number of second data training records by perturbing the first training data record with a random variable, and training each of the neural networks with one of the training data records. A prognosis and an estimation of the prognosis error can be carried out by means of such a system.
1. A method for training a plurality of neural networks for process optimization, comprising the steps of:
determining a first training data record, wherein each training datum has a measurement accuracy and wherein the training data are obtained by perturbing the data within an error range of the measurement accuracy;
generating a plurality of second data training records by perturbing the first training data record with a random variable;
training each of the plurality of neural networks with one of the training data records; and applying the trained neural networks to process optimization.
2. The method according to claim 1 , wherein the second training data is assigned an accuracy.
3. The method according to claim 1 , wherein the first and second training data involve measurement quantities, each measurement quantity having a measurement accuracy.
4. The method according to claim 1 , wherein the random variable is an equidistributed random variable.
5. The method according to claim 4 , wherein the random variable has an expectation of zero and a result of an addition of the random variable to the training datum lies within an accuracy range of the training datum.
6. The method according to claim 1 , wherein the random variable is a normally distributed random variable.
7. The method according to claim 6 , wherein the random variable has an expectation of zero and a result of an addition of the random variable to the respective datum lies with a predetermined probability within the accuracy range of the training datum.
8. The method according to claim 7 , wherein the predetermined probability is about greater than or equal to 95%.
9. The method according to claim 8 , wherein the predetermined probability is about 99%.
10. The method according to claim 1 , wherein the random variable is taken from a rigidly predetermined level.
11. The method according to claim 1 , wherein the random variable is taken from an interval dependent on the accuracy.
12. The method according to claim 11 , wherein the prognosis value is calculated by averaging the prognoses output by the plurality of neural networks.
13. The method according to claim 12 , wherein the prognosis value is compared with a threshold and a signal is outputted when the standard deviation lies above the threshold value.
14. The method according to claim 11 , wherein a standard deviation of the prognoses output by the plurality of neural networks is determined as a measure of the reliability of the prognosis value.
15. A method for providing a prognosis by utilizing a plurality of neural networks, comprising the steps of:
determining a first training data record, each training datum has a measurement accuracy and wherein the training data are obtained by perturbing the data in an error range of the measurement accuracy;
generating a plurality of second data training records by perturbing the first training data record with a random variable;
training each of the plurality of neural networks with one of the training data records; and
determining a prognosis value based on an evaluation of the prognoses output by the plurality of neural networks.
16. The method according to claim 15 , wherein the standard deviation of the prognoses output by the plurality of neural networks is used as a measure of the prognosis error.
17. A computer system for determining at least one of prognosis value, comprising
a plurality of neural networks trained by training data records, each training datum has a measurement accuracy and wherein the training data are obtained by perturbing the data in an error range of the measurement accuracy;
an input means for entering input data into the neural networks,
an evaluating means for evaluating the prognoses output by the neural networks for determining a prognosis value.
18. The computer system according to claim 17 , wherein the evaluation means are designed to calculate the mean of the prognoses.
19. The computer system according to claim 17 , wherein the evaluation means are designed to calculate a standard deviation of the prognoses as a measure of the reliability of the prognosis value.
20. The computer system according to claim 19 , further comprising means for comparing the standard deviation with a predetermined threshold value, and for outputting a signal when the standard deviation lies above the threshold value.
21. The computer system according to claim 17 , further comprising means for outputting a prognosis error, the prognosis error being determined on the basis of the standard deviation.