EventClustering for improved real time input variable selection and data modelling


This paper proposes a unique and novel approach for real time input variable selection (IVS) sensitivity analysis (SA) applicable to large scale complex systems. Borrowed from the EventTracker [1] principle of interrelation of causal events, it deploys the Rank Order Clustering (ROC) method to automatically group every relevant system input (e.g. sensor and actuation) within the known boundaries of the system to parameters that define the state of the systems (e.g. Performance Indicators or status). The proposed event modelling technique removes all the logical boundaries of isolation that exits in complex systems with the principle that every acquirable knowledge or data (input) affects the output unless proven otherwise. In addition to being able to filter unwanted data, it is capable of including information that was thought irrelevant at the outset. This feature is unique and novel. The underpinning logic of the proposed event clustering (EC) technique is building an event cause-effect relationship between the inputs and outputs of the system the technique is not only capable of group inputs with relevant corresponding output, but also in short spans of time (relative real-time) measure the weight of each input variable on the output variables. The proposed method will become the foundation for control and stability operations in large and complex systems. Our motivation is that components of current complex and organised systems are capable of generating and sharing information within their known domain (network of interrelated devices and systems). Normally monitoring and control systems are equipped with sensors and actuators that allow for the monitoring and control of isolable systems. The purpose of isolating control system into smaller components is to simplify functionality. The isolation allows for mathematical solutions to work. However, modern interrelated complex systems do not necessarily lend themselves to the classical control engineering solutions. The knowledge of systems has improved, thanks to sensor and actuation, communication and overall computer and electronic engineering. Such systems combined with mechanical parts require better models. The authors believe that the proposed event clustering and sensitivity analysis technique allows monitoring and control systems to become more flexible and responsive in dealing with real-time events. By removing the boundaries of the systems a more accurate representation of the cause-effect relationship is thus generated. This improvement in the quality and at times the quantity of input data may lead to improved higher level mathematical formalism. One may hope that better models will result into better control and decision making. In this paper, an experiment in Cement Kiln operation case demonstrates the suitability and applicability of Event Clustering modelling method in industrial applications. We use the existing Supervisory Control and Data Acquisition (SCADA) in the plant that monitors the operations of the Kilns in a Cement factory. The data collected from the sensors and actuators of the production process corresponds to the input data that measures the kiln's production rate. The EventCluster algorithm resides within the control centre of the SCADA system to assess the contribution of each input to the overall key performance indicators (kiln output) of the process. The data produced by the event modellers is used for generating the fuzzy parameters/inference rules of the fuzzy controller of the plant.

DOI: 10.1109/CCA.2014.6981574

6 Figures and Tables

Cite this paper

@article{Danishvar2014EventClusteringFI, title={EventClustering for improved real time input variable selection and data modelling}, author={Morad Danishvar and Ali Mousavi and Pedro Angelo Morais de Sousa}, journal={2014 IEEE Conference on Control Applications (CCA)}, year={2014}, pages={1801-1806} }