# Method for splitting time-sampled signal into two signals

Update: the yellow area in the graph below can be ignored, it shows power produced. I'm only interested in the blue line, and how to separate power consumed by the heating system from the rest.

I have data for a signal (blue line) showing overall power consumption. The spikes come from a heating system. The image below shoes data for 24 hours.

I would like to split the signal into two separate signals, to find out how much power is consumed by the heating system.

Are there any methods to achieve this?

• What's the yellow curve / area?
– Peter K.
Commented Nov 28, 2022 at 19:57
• The yellow area can be ignored. The screenshot shoes overall power consumption (blue) and power produced by a photovoltaic collector (yellow). Question updated. Commented Nov 28, 2022 at 21:53
• What do you mean by splitting up the signal? Splitting it up into what?
– Jdip
Commented Nov 28, 2022 at 22:10
• Splitting it into one graph containing the peaks only (energy consumed by heating system) and another graph containing the "baseline" (energy consumed by all other devices). Commented Nov 28, 2022 at 22:50
• The blue line goes up during daylight hours mostly, for cooking etc., around the center of the graph. The heating system is using energy in regular intervalls, all day long, see peaks. Commented Nov 28, 2022 at 22:52

Are there any methods to achieve this?

Not without some additional information. You may be able to partially separate this by carefully studying the spikes from the heating system. Many heating systems are binary: they are either on or off and during the "on state" the power consumed is constant.

In your case it looks like the value for "heat on" is about 9.5. You could separate the graphs subtracting 9.5 for each value that's above 9.5. That would be the baseline load and the subtracted spikes would be the heating system. Of course, that only works if the constant power assumption is justified

The usual way I'd think about approaching this sort of problem is to apply a Kalman filter.

The first step in applying a Kalman filter is knowing what the model for your signal is: Can you write down how you think the power for the heating system works as one equation and the way the rest of the power usage varies as another equation?

If you can, then this approach might work.

If the equations are nonlinear (not linear, constant coefficient ordinary differential equations), then you will need to apply the Extended Kalman Filter or some other nonlinear estimator.