System and method employing a self-organizing map load feature database to identify electric load types of different electric loads

1. A method of identifying electric load types of a plurality of different electric loads, said method comprising:

providing a self-organizing map load feature database of a plurality of different electric load types and a plurality of neurons, each of said different electric load types corresponding to a number of said neurons;

employing a weight vector for each of said neurons;

sensing a voltage signal and a current signal for each of said different electric loads;

determining a load feature vector comprising at least four different load features from said sensed voltage signal and said sensed current signal for a corresponding one of said different electric loads;

identifying by a processor one of said different electric load types by relating the load feature vector to the neurons of said self-organizing map load feature database by identifying the weight vector of one of said neurons corresponding to said one of said different electric load types that is a minimal distance to the load feature vector;

employing with said identifying the weight vector an average squared Euclidean distance to a plurality of neurons in a class corresponding to said one of said plurality of different electric load types;

employing i as an index;

employing ω i as said class;

employing x as said load feature vector;

for each of said plurality of different electric load types, employing a group of values of said self-organizing map load feature database having a mean y i and a square covariance matrix Σ i ;

employing with said identifying the weight vector a point-to-cluster function of average squared Euclidean distance from said load feature vector to every point in said class, ω i , corresponding to said one of said plurality of different electric load types;

employing Tr( ) as a trace of the square covariance matrix Σ i ; and

determining the average squared Euclidean distance from:

d E ( x,ω i ) 2 =( x−y i ) T ( x−y i )+ Tr (Σ i ).

2. A method of identifying electric load types of a plurality of different electric loads, said method comprising:

providing a self-organizing map load feature database of a plurality of different electric load types and a plurality of neurons, each of said different electric load types corresponding to a number of said neurons;

employing a weight vector for each of said neurons;

sensing a voltage signal and a current signal for each of said different electric loads;

determining a load feature vector comprising at least four different load features from said sensed voltage signal and said sensed current signal for a corresponding one of said different electric loads;

identifying by a processor one of said different electric load types by relating the load feature vector to the neurons of said self-organizing map load feature database by identifying the weight vector of one of said neurons corresponding to said one of said different electric load types that is a minimal distance to the load feature vector;

training said self-organizing map load feature database with a statistical distance measure;

employing i as an index;

training said self-organizing map load feature database with x as an input load feature vector;

for each class of said plurality of different electric load types, employing a group of values of said self-organizing map load feature database having a mean y i and a square covariance matrix Σ i ;

employing as said statistical distance measure a Mahalanobis distance from said input load feature vector to the weight vector, y, of each of said neurons of said self-organizing map load feature database;

employing an average of a within-class square covariance matrix Σ i for the class that the weight vector of a corresponding one said neurons belongs to as the square covariance matrix of both said input load feature vector and the last said weight vector;

determining the Mahalanobis distance from:

d M ( x,y ) 2 =( x−y ) T Σ −1 ( x−y );

providing a plurality of samples of extracted load features from said sensed voltage signal and said sensed current signal for one of said different electric loads;

employing k as an index;

employing ω i as said class;

quantizing said determined load feature vector as x′[k] based on a predetermined resolution;

identifying said quantified determined load feature vector x′ [k] appearing in said samples a plurality, T k , of times;

identifying said one of said different electric load types as a best match to said ω i a plurality, T ki , of times; and

determining likelihood of said identifying by the processor of said one of said different electric load types from:

Pr

⁡

(

ω

i

❘

x

′

⁡

[

k

]

)

≈

T

ki

T

k

.