Towards the use of LSTM-based Neural Network for Industrial Alarm Systems

Abstract

With the coming of the complex industrial plants, the majority of control rooms sharing a common scenario involve operators faced amount of alarms during an overload. Besides, were often unable to determine which were important in an alarm widespread flooding scenario is an increasingly frequent situation. To solve this issue, proper handling of industrial processes requires designing, implementing and updating Process Malfunction Prediction (PMP) systemsable to give advice to the operators being a recommendation to make necessary adjustments in operating variables. From the wide range of models, we apply Recurrent Neural Networks (RNNs) using Long Short-Term Memory (LSTM) units, built was a regressor trained to predict the behavior of failures in an industrial process. The activity values of LSTM units can give recommendations for the monitoring of malfunctions. To this end, data mining and machinelearning techniques are used, which allow the implementation of a regression. The distinctive feature of PMP is that dynamically provides information using the data process. Further, proposed approach was evaluated in a simulated industrial process case study scenario. Lastly, the evaluation of the experimental results demonstrate the contribution of this work.