Algorithm 1. Algorithm for the FCM clustering/piloting RBF-ANN model.

Applying the FCM clustering on an on-site collected training dataset $\Im ={\left\{X_k{,}^{}{d}_k\right\}}_{k=1}^Q$ to coarsely determine the center and spread parameters of the q (= c in the FCM) Gaussian basis functions of the RBF-ANN model of Equation (5).   $\looparrowright$ Specify values: the number of clusters c, the degree of fuzziness m > 1, and a tolerance to be set. $\looparrowright$ Generate the c cluster centers randomly.   l ← 0 $\looparrowright$ Repeat  Compute the c cluster centers, using Equation (3).  Update the membership matrix, U, using Equation (4).   l ← l + 1,  until  the tolerance, ||U(l+1) − U(l)||, is approximately met.   $\looparrowright$ Return the resulting c cluster centers found with the membership matrix U.   $\looparrowright$ Center the Gaussian basis functions of Equation (5) at the resulting c cluster centers found through the FCM clustering. $\looparrowright$ Compute the spread parameters of the Gaussian basis functions of Equation (5), using Equation (10). 1

Use the SVD technique, with the on-site collected training dataset, to train the RBF-ANN model of Equation (5). Its Gaussian basis functions are heuristically initialized by the FCM clustering.   $\looparrowright$ Train Equation (5), by Equation (9), in cloud analytics, to get W. $\looparrowright$ Deploy the well-trained AI model on-site on the presented smart Arduino MCU-based power meter prototype for on-line load monitoring in DSM. (1 The spread parameter of each of the Gaussian basis functions can be computed with e−1 from clustered data with their cluster mean/the membership matrix.)