Layered Integration Approach for Multi-view Analysis of Temporal Data

Abstract

In this study, we propose a novel data analysis approach that can be used for multi-view analysis and integration of heterogeneous temporal data originating from multiple sources. The proposed approach consists of several distinctive layers: (i) select a suitable set (view) of parameters in order to identify characteristic behaviour within each individual source (ii) exploit an alternative set (view) of raw parameters (or high-level features) to derive some complementary representations (e.g. related to source performance) of the results obtained in the first layer with the aim to facilitate comparison and mediation across the different sources (iii) integrate those representations in an appropriate way, allowing to trace back similar cross-source performance to certain characteristic behaviour of the individual sources. The validity and the potential of the proposed approach has been demonstrated on a real-world dataset of a fleet of wind turbines.

Advanced Discretisation and Visualisation Methods for Performance Profiling of Wind Turbines

“…Based on these labels, one can derive additional insights about the turbine operation and performance. For concrete details of the labelling approach, please refer to [6].…”

“…In : [6], we have proposed a novel approach that can be used for multi-view analysis a 279 integration :::::::::: integration :::: and :::::::: analysis of heterogeneous real-world data sets originating from multip…”

“…In : [6], we have proposed a novel approach that can be used for multi-view an integration :::::::::: integration :::: and :::::::: analysis of heterogeneous real-world data sets originating fro sources. The approach facilitates the extraction of an explicit link between the internal work ::::::: between :::: the :::::::::::: endogeneous ::::::::: operating : modes (different compositions of the endogeneous p and the expected :::::::: observed : output (e.g., active power for wind turbines) as a function of the e parameters (e.g., wind speed, wind direction and temperature) at fleet level, employing integration.…”

“…In [6], we have proposed a novel multi-view analysis approach that facilitates the integration of heterogeneous real-world data sets originating from several different sources. The main result of the application of this approach is that the multivariate time series of each turbine in the fleet can be converted into a letter code (as a DNA sequence) by assigning a unique label to each timestamp expressing a distinct operating mode.…”

“…Our main contributions in this article are as follows: (1) to explore further the letter code discretisation introduced in [6] for performance comparison and temporal profiling across the fleet; (2) to realise a circular binning discretisation technique that can produce a set of context-specific performance representations; (3) to demonstrate how visual analytics techniques can benefit from the resulting bins, allowing us to gain deeper insights into the production behaviour; (4) to exploit further the circular binning through non-negative matrix factorisation, enabling performance profiling and benchmarking across the fleet. The potential of the proposed discretisation techniques has been illustrated by discussing how they can be utilised in three use cases: (I) visual inspection for detecting trends; (II) continuous anomaly detection and (III) production forecasting.…”