References in periodicals archive ?
The significant problem of NILM models based on FHMM is that the performance is decreased as increasing the number of appliances.
The first reason is that the time complexity was getting higher as the number of appliances was increased in the FHMM variants.
Require: P = [p.sub.1], [p.sub.2], ..., [p.sub.T], Reflection rate [alpha] Ensure: VP = [vb.sub.1], [vb.sub.2], ..., [vp.sub.T] (1) VP [left arrow] 0 (2) for i = 2 to T do (3) d [left arrow] [p.sub.i] - [vp.sub.i-1] (4) [vp.sub.i] [left arrow] [vp.sub.i-1] + [alpha] x d (5) end for (6) return VP The second reason is that the long-term pattern cannot be learned in the FHMM variants (see Figure 7), because they are based on the first-order Markov chain.
Also, we take the same experiment by FHMM and compare the result.
Firstly, we train all houses of UK-DALE and the result will be compared with FHMM. The parameters for UK-DALE experiments are in Table 15.
We compare the performance with FHMM by implementing it.
We compare our model with three existing models such as FHMM, Additive FHMM, and HieFHMM [28, 32].
Acronyms browser ?
Full browser ?