A Scenario-Based Stochastic Optimization Approach for Non-Intrusive Appliance Load Monitoring

Abstract

In this paper an energy disaggregation problem, commonly studied as “non-intrusive load monitoring”, is presented. Non-intrusive load monitoring is a procedure for estimating the energy profiles of individual electric home appliances from the aggregated power measurement. We present this problem as an optimization-based problem in which the least-square errors are minimized to find the set of active appliances, using the instantaneous aggregated power. The convex regularization terms are also added, which exploit the information of the probability of individual appliance usage and assume the behavior of the appliances’ power profiles to be constant as piecewise signal over time. The problem is formulated using a scenario-based stochastic optimization approach to consider the underlying uncertainties in power measurements. The expected value of the problem’s objective function is computed by using the sample average approximation method, where normally distributed samples of a random variable are introduced into the problem using the Monte Carlo simulation. Moreover, the appliances’ limits on the power modes are formulated as a chance constraint in the optimization problem. The training and testing of the proposed algorithm are done by using the benchmark data: AMPds and REFIT. The simulation results show that the proposed scenario-based stochastic optimization approach successfully estimates the energy profiles of individual appliances with multiple power modes.