Improving time series anomaly detection based on exponentially weighted moving average (EWMA) of season-trend model residuals

Advances Toward the Next Generation Fire Detection: Deep LSTM Variational Autoencoder for Improved Sensitivity and Reliability

“…(2) the CUSUM ( ) fire detection [13,14], the EWMA ( ) anomaly detection [30], and two currently used fixed temperature detectors with thresholds of 47℃ and 58℃ each ( and ) [7,8,9]. In this section, we introduce the simulation-based experiments and two of the real-world datasets.…”

“…To evaluate the effectiveness of our method, we develop computational experiments with high-fidelity large eddy simulation (LES) data, and we also use real-world fire and non-fire datasets [29]. We compare our proposed fire detection with alternative methods, including standard LSTM detection [27], CUSUM fire detection [13,14], exponentially weighted moving average (EWMA) anomaly detection [30], and fixed-temperature heat detection [7,8,9]. Our pool of alternative methods includes both point anomaly and contextual anomaly detections for a fair comparison.…”

“…We use the 300 samples before fire ignition (−3 ≤ ≤ 0 ) with this signal noise to train both the LSTM-VAE and LSTM anomaly detection and for the EWMA threshold [30].…”

“…During the exhaustion phase (206 ≤ ≤ 382 ), the temperature decreases because of the exhaustion of fuel. We use the first 300 samples in the nominal phase to train both the LSTM-VAE and LSTM anomaly detection and to calculate/set the EWMA threshold [30]. The trained LSTM-VAE and other detection methods are then used to supervise the fire event after ignition ( = 0 ) and to evaluate their performance.…”

“…The main trend of the temperature increase is the result of a heated polyurethane foam in this experiment. We use the first 300 samples to train both the LSTM-VAE and LSTM anomaly detection and to calculate/set the EWMA threshold [30]. The trained ML models are used to supervise the temperature signal and to evaluate the possibility of false alarms.…”