0

I would like to know, a mathematical model, which suits in predicting future values at different time based on past values .

In my problem, a list nodes have different speeds at different time; I want to estimate speed of all nodes in future using any prediction model.

Say, node-1 has the speed at time 10 seconds, 20s, 30s, as 43 m/s, 23 m/s, 29 m/s, then how to predict its future speed at time 40s, 50s ? similarly, for node-2,3,4?

TZakrevskiy
  • 22,980
Prasanna venkatesan T
  • 3
  • I think you'll need to state your problem much more explicitly. – littleO Nov 27 '17 at 10:01
  • Say, node-1 has the speed at time 10 seconds, 20s, 30s, as 43 m/s, 23 m/s, 29 m/s, then how to predict its future speed at time 40s, 50s ? similarly, for node-2,3,4? – Prasanna venkatesan T Nov 27 '17 at 10:16
  • You have to think about what variables will affect the measured speeds and perhaps how these might change daily, weekly and annually. If these speeds change depending on human activity/habits in any way then you might want to read up about how the use of electricity is modelled for example. – James Arathoon Nov 27 '17 at 10:36
  • @JamesArathoon, The max speed and time is set to 100 m/s, 500 seconds respectively + some random factors – Prasanna venkatesan T Nov 27 '17 at 10:56

1 Answers1

0

Your model will depend on the assumptions that you make about how future values are related to past values.

If you think future values are completely determined by past values - for example, if you are looking for a numerical solution to a system of differential equations - then you will build a deterministic model.

If you think future values are affected by other factors that are not captured in past values then you will build a stochastic model in which you include some random factors. Then you will have to make some assumptions about the degree to which future values are correlated with past values, and the probability distribution of the random factors.

The Monte Carlo method is a method of generating aggregate results from a stochastic model by means of random sampling. It is not a type of model.

gandalf61
  • 15,326
  • future values will depend both past and some random factors; max speed is set to 100 m/s. which probability distribution suits with these scenario? – Prasanna venkatesan T Nov 27 '17 at 10:50
  • From the information you have given (3 data points) it is not possible to say. To build a credible model you need either (a) some understanding of the underlying physical psocesses which suggests an appropriate mathematical model or (b) a much larger number of data points (say 100 or more) so that you can use some data points to construct a possible model and the remainder to validate your model. – gandalf61 Nov 27 '17 at 11:12
  • not only 3 data points, each node has larger data over different time (more than 100). the my problem has 100 nodes with 500 seconds and max speed 100 m/s of varies combination of data values. – Prasanna venkatesan T Nov 27 '17 at 11:31
  • If you have 100+ data points then you could use, say, 50 data points to build and parameterise a model and the remainder to validate your model. The exact form of your model will depend on the values in your data points. If they lie close to a straight line then you could build a linear model using linear regression to determine its parameters. If the values lie close to some other curve then you aill need to build a non-linear model. – gandalf61 Nov 27 '17 at 12:08
  • Which model or tool suits with my problem? – Prasanna venkatesan T Nov 27 '17 at 12:41