Cycle prediction

The PredictCycle event builder predicts the magnetic field of the last cycle in the buffer passed from the Prediction buffer, and sets the CycleData.field_pred attribute.

The transformation has 2 behaviour depending on the triggering buffer. Note that the PredictCycle converter has no CMW event triggers.

Settings

The PredictCycle has the following settings:

• autoregressive Uses an internal state as the context for the next prediction, and not the measured fields present in each buffer passed. Note that the internal state has to be set at the first prediction (i.e. the first buffer must have measured fields).
• programmed_current Uses programmed current for predicting the corresponding field response, instead of measured current (that the models are trained from). Setting this to True is preferred, as the programmed current is noise-free, and is sure to correspond to the correction we are correcting with respect to.

Logic

Normal prediction

In the case of the “normal” buffer, the field of the next cycle is predicted with the past state and future current, and field_pred is set on the CycleData. Before prediction, the previous state is saved in the case of Delayed prediction, and the actual state is updated after the prediction.

Delayed prediction

In cases where the actually played cycle is not known at original prediction time (Normal prediction), the field of the actually played cycle needs to be determined for next predictions (and the internal state for the next cycle, if in Autoregressive predictions), otherwise the past context will be wrong.

Delayed predictions are used for Dynamic Economy and Full Economycycles.

For delayed predictions, the state before the last prediction (i.e. the prediction of the cycle that we want to re-predict) will be re-used, in order to use the context of the cycle for the re-prediction.

After the re-prediction the Cycle.field_pred is re-set.

Autoregressive predictions

Output

CycleData object with field_pred. There is only 1 output.