# Forecatsing through time series analysis

I have values of the current consumption of the motor from a system that automatically opens and closes a door. The measurements are carried out on average every 3 months, with all opening and closing operations within one day being recorded for each recording period. If the standard deviation is formed from the measured values of one day, it is possible to derive statistical values from this to determine the condition of the system (very roughly expressed).

I would now like to forecast the future behaviour or rather the future condition based on already existing, asset-specific data, which I have roughly had since 2008. Hence my question: Is it possible to calculate such forecasts relatively easily in R, or can someone perhaps give me a tip on how I can implement this task in R. For a better visualization, here is a figure showing how the individual values of the standard deviation behave.

Each improvement represents a maintenance carried out. A linear course between 2 maintenance operations can be assumed. My goal would be to predict the future course of the standard deviation in order to determine the time of the next necessary maintenance.

Hi Mike,

You can do linear regresion model with lm function or maybe a simple time series forecast with "Forecast" package, using ARIMA mode if your model presents seasonality. How many observations or rows, do you have? Could you show a summary of your database?

Hello Rafael.F

Thank you for your answer. I have 35 observations for this facility. The Data:

2005,88 2,07
2006,36 2,29
2006,87 1,7
2007,27 2,09
2007,88 2,4
2008,34 2,57
2009,35 2,56
2009,64 2,71
2009,88 2,9
2010,7 3,1
2010,9 2,89
2011,3 3,05
2011,9 3,2
2012,3 3,3
2012,7 2,74
2012,9 2,93
2013,7 3,5
2013,9 3,58
2014,2 3,8
2014,4 2,65
2014,9 3,05
2015,2 3,22
2015,4 3,28
2015,9 2,43
2016,2 2,8
2016,4 2,89
2016,7 2,5
2016,9 2,9
2017,2 2,25
2017,4 2,27
2017,6 2,51
2017,9 1,82
2018,2 2,15
2018,3 2,33
2018,6 2,45

The first row represents the measurement date (year,month). The second row represents the measurement value. Additional, the executed maintenane actions:

2006,71
2009,04
2010,8
2012,5
2014,3
2015,6
2016,5
2017,05
2017,8

Once again: year,month

I think your idea with ARIMA is quite good and I also think that a linear regression would be very good. However, the "red lines" in the graph above represent maintenance tasks. How could one predict these maintenance actions?

Thanks, Mike

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.