Online assessment of failure probability for smart meters based on SARIMA-LTFRLS

Abstract

• An adaptive time-varying forgetting-factor based on the fuzzy control principle is proposed for the first time, which can balance the stability and convergence speed of identification, thus ensuring the real-time prediction of SARIMA. • The limited memory principle is introduced into the identification process to keep the amount of BE data used for recursion fixed, and then a new LTFRLS which can well track the dynamic change of parameters in SARIMA is presented. • A new FP online assessment framework for SMs is given by combining SARIMA and LTFRLS, which combines the failure thresholds of BE and its future predictions to evaluate the running state of instruments. Online assessment of the failure probability (FP) for smart meters (SMs) is crucial for accurately measuring the electrical energy of instruments and judging their operation state. Basic error (BE) is the main performance parameter of SMs, and its trend and seasonality can be well described by the seasonal autoregressive integrated moving average (SARIMA) model. However, when the BE data increase online, the static SARIMA can't quickly and accurately reflect the dynamic performance of SMs. To this end, a limited-memory time-varying forgetting-factor recursive least squares (LTFRLS) algorithm for real-time updating of parameters in SARIMA is proposed for the first time, and a new FP online prediction framework is constructed by combining it with SARIMA. Firstly, an adaptive time-varying forgetting-factor (TF) based on the fuzzy control theory is presented to dynamically adjust the forgetting factor in LTFRLS. Next, the limited memory principle is introduced to keep the amount of BE data involved in identification iteration unchanged, thereby overcoming data saturation. Actual datasets from three companies show that LTFRLS has good recognition accuracy, fast convergence speed, and versatility. Compared with several others, SARIMA-LTFRLS can realize online prediction of BE, which is more effective for the state prediction of SMs.