Implementation of Hidden Markov Model for Home Appliance Load Detection

Hello,

I am trying to implement HMM to detect electrical appliances' states (on,off or transitions). I am polling real time data (real power and reactive power) from a smart meter at 1 sample per second. Consider a simple system consisting only three simple device. These are:

A light: States- 1.On (50W real power, 10W reactive power) 2. Off (-50W real power, -10W reactive power)

A blender: States- 1. Power A (100W, 20W) 2. Power B (150W, 40W) 3. Off

A fan: States- 1. On (70W,-20W) 2. Off (-70W,+20W)

What I am getting is the temporal data sequences of real and reactive power (rise or fall).

Now my question is how to model this system to detect the states of devices (e.g. light off, blender power A) and set the parameters of HMM.

How to set the transition matrix, which is the emission matrix and initial state matrix!

[Muthu Annamalai]

What have you tried so far? MATLAB allows you to use matrices to define initial states of the HMM, and update it using state equations. You maybe missing parts of your problem definition, and algorithmic ideas, without which it is hard to get started on expressing them in MATLAB.

[Abdullah]

So far I have tried to build a model only for the blender. Here my transition matrix is [.1 .4 .5; .4 .4 .2; .4 .2 .4]

my initial state matrix is [1 0 0] and emission matrix is:

[0 100 150; -100 0 50; -150 -50 0]

then I use hmmgenerate to generate a sequence and use viterbi algorithm to find most likely state. Is it correct?

• The transition matrix is arbitrary or depends on users' usage pattern. The emission matrix is generated assuming that the change in real power is the emission.

So initially it starts from off condition and in steady state, so change is 0W. Now when I turn on power state 1 it changes 100W similarly from off to state 2 is 150W and so on.

[Ali Bou Haidar]

I have a similar task at hand. I am trying to desegregate the total power consumption into individual appliance consumption knowing the total number of appliances and assuming two states (ON or OFF). I am trying to implement the Hidden Markov Model or the Factorial Hidden Markov model. Can you provide any help with that? So far I have managed to generate the transition and observation matrices using both the 'hmm.m' and 'fhmm.m'. I am trying to use the 'hmmgenerate.m' do you know if it's applicable? If not what can I use instead?

Answer 1

[Ali Bou Haidar]

I have a similar task at hand. I am trying to desegregate the total power consumption into individual appliance consumption knowing the total number of appliances and assuming two states (ON or OFF). I am trying to implement the Hidden Markov Model or the Factorial Hidden Markov model. Can you provide any help with that? So far I have managed to generate the transition and observation matrices using both the 'hmm.m' and 'fhmm.m'. I am trying to use the 'hmmgenerate.m' do you know if it's applicable? If not what can I use instead?

Answer 2

[Wided HAMMEDI]

I am working on the home appliance load detection, I don't understand how to modelise the states of my appliance (signal with noise). Shall I make a Markov states for each appliance? I will be greatful if someone can help me. Thank you