An Appliance-Agnostic Mode Identification Framework via Dynamic Programming Least Squares and Piecewise Regression for Non-Intrusive Load Monitoring

Accurate appliance-level Load Monitoring (LM), particularly through Non-intrusive LM (NILM), is increasingly important for effective Demand-Side Management (DSM) in modern power systems. However, NILM techniques often struggle to achieve high disaggregation accuracy due to challenges in identifying underlying appliance operating modes. This paper presents an analytical, appliance-agnostic framework that integrates Dynamic Programming Least Squares (DPLS), piecewise linear regression, and rule-based clustering to robustly extract and characterize the transient signatures of appliances and estimate their associated mode characteristics especially the unique consumption trend inside each of the modes. Applied to granular sub-metered data from two representative refrigerators, the proposed method constructs concise mode dictionaries that capture key consumption behaviors—including steady-state, compressor transitions, and defrost events. The resulting structured mode dictionaries achieve over 98% variance explained (R^2), with the error of as low as 0.126 p.u., which can be readily adopted by NILM algorithms, improving disaggregation accuracy and generalization across appliance types.