Power Safety Warning Method Based on GM-FHMM Load Decomposition Algorithm

With the promotion and installation of large-scale intelligent electric energy metering devices being carried out all over the world, non-intrusive equipment load decomposition has become a research focus. Non-intrusive load decomposition refers to the process of decomposing bus data acquired by a port/node into branch data with feature identification. Decomposing the main node's load can find the non-technical attacks in the downstream nodes. In this paper, the Gaussian Factorial Hidden Markov Model (GuassFHMM) is used to construct and experiment the load decomposition model, safety warning requirements are realized by improving the accuracy of load decomposition and local optimality. A dynamic programming algorithm-the extended Viterbi algorithm is given as the core of the model to decompose the total Markov chain more precisely. The Baum-Welch algorithm is introduced to study the model parameters, and add the simulated annealing algorithm to solve the local optimal problem. Case studies based on the REDD load data set are conducted to prove the effectiveness and practicality of the algorithm.