Integrating artificial intelligent techniques and continuous time simulation modelling. Practical predictive analytics for energy efficiency and failure detection

Energy efficiency and reliability needs are growing in many economic sectors, where predictive analytics are becoming essential tools for these key variables forecasting. When predicting these variables, in many occasions, the problem to simplify the prediction model format when dealing with similar systems, which are placed in different functional locations, is a very complex problem due to model unavoidable dependency on changing operating conditions (per time and location). So effort is placed in this paper to develop tools that can easily adapt prediction models' structure to existing operating conditions, for a given time period and place where the asset is located. Furthermore, these tools may allow the model to be easily trained and tested for automated implementation within the plant's remote surveillance system. To this end, Artificial Intelligence (AI) techniques, and in particular artificial neural network (ANN) models, have been selected in this paper as prediction models, since their structure can be adapted to improve predictions accuracyand they can also learn from dynamic changes in environmental conditions. To demonstrate the adaptability for prediction accuracy and self-learning capabilities of the model, we have implemented an ANN with a backpropagation algorithm as a continuous time simulation model, which is then implemented using Vensim simulation environment, to benefit of the outstanding software optimization features for fast training. Using this model we provide predictions of asset degradation and operational risk under existing real time internal and locational variables. We can also dynamically release preventive maintenance activities. This prediction model is exemplified in an industrial case for failures in cryogenic pumps of LNG tanks. (C) 2019 Elsevier B.V. All rights reserved.