ADAPTIVE PREDICTION SYSTEM

摘要

Methods, computer program products, and systems are presented. The methods include, for instance: modeling for a subject process by machine learning with adaptive inputs. In one embodiment, the modeling may include: generating a model by use of machine learning with training data from measurements of successive components of a process to be modeled in order to predict measurements of a succeeding component within a statistically meaningful prediction range; adjusting the generated model by use of machine learning with less-deviation inducing measurements from a preceding component in case the measurement of the succeeding component is out of the prediction range; and presenting the adjusted model as a prediction model for the process.

主权项

1. A method for modeling for a subject process by machine learning with adaptive inputs, comprising:
obtaining training data for machine learning, the training data comprising a first input and an output, wherein the first input is at least one measurement from corresponding points of a first subprocess of the subject process and the output is at least one measurement from corresponding points of a second subprocess of the subject process, wherein the first subprocess precedes the second subprocess such that the measurements of the first input respectively influence the measurements of the output within the subject process, and wherein the first input and the output are stored as respective digital data in a memory coupled to a computer; recording, on at least one location in the memory, a first model generated by use of machine learning with the training data; storing, in the memory, a prediction range calculated by use of values of the first input; in response to ascertaining that a present measurement in the output falls out of the prediction range, adjusting the first model into a second model based on a second input that induces less deviation in the output than the first output such that the second model is more likely to have the output within the prediction range than the first model; and producing, on at least one output device coupled to the computer, the second model as a prediction model for the subject process.