Event Detection of Air Condition for NILM Using Piecewise Linear Representation Optimized by Dynamic Programming

Air-conditioning systems represent a crucial controllable resource within virtual power plants (VPPs), and their operational status is of significant importance to the development of VPPs. Nonintrusive load monitoring (NILM) is an effective method for monitoring air-conditioning systems and other VPP resources, with event detection serving as its foundation. However, event detection for air-conditioning loads faces numerous challenges. Air conditioners exhibit prolonged start-up times, and their electrical parameters, such as current and power, fluctuate significantly during operation. Additionally, events often occur in close temporal proximity. These characteristics render existing fixed-parameter event detection methods incapable of achieving high accuracy. To address this challenge, this article proposes a robust piecewise linear representation (PLR)-dynamic programming (DP)-Bayesian information criterion (BIC)-based event detection method, leveraging PLR, DP, and the BIC. The proposed method first employs PLR to segment the collected load electrical signals, thereby simplifying the complex fluctuations in a long time series into a series of smoother linear segments. This approach effectively captures the state changes of air conditioner loads and mitigates the impact of signal noise. The globally optimal segmentation results are subsequently determined using DP, and the optimal number of segments is identified via the BIC, with the corresponding segment boundaries recorded to pinpoint the exact timing of events. The effectiveness of the algorithm is validated through the collection of actual operational data from commercial and industrial central air-conditioning systems.