US 12,462,670 B2
Long duration alarm sequence predictions using Bi-LSTM and operator sequence recommendations thereof
Yogesh Angad Tambe, Pune (IN); and Trinath Gaduparthi, Pune (IN)
Assigned to TATA CONSULTANCY SERVICES LIMITED, Mumbai (IN)
Filed by Tata Consultancy Services Limited, Mumbai (IN)
Filed on Feb. 23, 2024, as Appl. No. 18/586,270.
Claims priority of application No. 202321042213 (IN), filed on Jun. 23, 2023.
Prior Publication US 2024/0428678 A1, Dec. 26, 2024
Int. Cl. G08B 31/00 (2006.01); G06N 3/0442 (2023.01); G08B 29/18 (2006.01)
CPC G08B 31/00 (2013.01) [G06N 3/0442 (2023.01); G08B 29/186 (2013.01)] 12 Claims
OG exemplary drawing
 
1. A processor implemented method for alarm sequence predictions, the method comprising:
generating, via one or more hardware processors, an alarm dataset for a facility from historical alarm data further comprising a timestamp of a triggered alarm, an alarm identity (ID), an alarm description and related equipment in the facility, wherein the facility is monitored via a plurality of sensors to capture the historical alarm data, each datapoint in the alarm dataset comprising a past input alarm sequence and a future output alarm sequence comprising single or multiple occurrences of unique alarms each identified with the alarm ID;
preprocessing, via the one or more hardware processors, the alarm dataset to generate a training dataset, wherein the preprocessing further comprising chattering removal, unimportant alarms removal, resizing the past input alarm sequence to an optimal input sequence length and the future output alarm sequence to an optimal output sequence length;
training, via the one or more hardware processors, a Bidirectional Long short-term memory (BiLSTM) on a training dataset for predicting an alarm sequence for the facility for a time duration of a plurality of hours, wherein a total number of the unique alarms in the historical alarm data represents a plurality of features for the BiLSTM, the optimal input sequence length is an input to the Bi-LSTM to generate the optimal output sequence length of the future output alarm sequence to be predicted post training by the BiLSTM;
generating, via the one or more hardware processors, a mapping table using the preprocessed historical alarm data of the facility to map a plurality of unique alarm sequences with associated plurality of operator action sequences;
predicting, during inferencing stage, by the trained BiLSTM executed by the one or more hardware processor, a future output alarm sequence for a next plurality of hours based on an immediate input alarm sequence recorded in recent past for the facility; and
recommending, via the one or more hardware processors, an operator action sequence from among the plurality of operator action sequences for the predicted future alarm output sequence, wherein the recommended operator action sequence is associated with a unique alarm sequence from among the plurality of unique alarm sequences that has a highest Matching Sequence Score (MSS) with the predicted future output alarm sequence.