how to decrease the error autocorrelation of a neural network

Hi all, I'm building neural networks, whose task is the prediction of time series of reactive power knowing the active power.

For the pre-processing of the data I carried out the statistical normalization of the data.

The problem is that in outputs the autocorrelation of the error is always outside the confidence limits. How can I reduce it? do i have to treat the incoming data differently? Someone help me please.

Answer 1

[Aditya Patil]

As per my understanding, you want to reduce the autocorrelation of a signal.

Decorrelation algorithms depend on the nature of the data. For example, if the signal follows a sine wave, it might not be sensible to change the nature of the signal to reduce autocorrelation.

Currently, there are no decorrelation algorithm implementations available for power, however, you should be able to implement any algorithm using Simulink/MATLAB.