Stochastic modelling framework for synthetic time series simulation

Research output: Contribution to conferenceOther

Abstract

Quite often, observed records (time-series) are not long enough to extract reliable statistics to understand the variability and uncertainty in the patterns and the occurrences of all significant rare events. One possible solution to such issues is to develop robust data-centric modelling techniques for artificially generating realistically possible synthetic sequences based on the historical dataset and within the acceptable statistical errors. The Hidden Markov Model (HMM) is popular stochastic modeling approach and has been applied successfully to model a range of complex processes, such as bio-informatics, speech, molecular evolution, the stock market, natural languages, human and animal behaviour.

This presentation is aimed to present the systematic organisation of an HMM-based modeling framework (referred in this presentation and relevant publications as HMM-GP) that couples STL: a Seasonal-Trend decomposition procedure based on Loess and the extreme values distribution (such as Generalised Pareto Distribution), to develop a robust modeling schematic for artificially synthesizing univariate time series. Considering the environmental statisticians as the key audience, the presentation will be organized to demonstrate the step-by-step development of the HMM-GP modeling schematic, including a thorough investigation to examine its suitability for simulating highly stochastic time series of streamflow at a much finer temporal resolution of 15 minutes. A robust validation of the proposed HMM-GP schematic would be presented by conducting an extensive comparison of various statistical characteristics of the observed records with the synthetically simulated flow time series across four hydrologically distinct case-studies River in the UK, namely Don, Nith, Dee, and Tweed. Further, for the benefit of the general audience interested in potential of data-centric modelling techniques and to illustrate the potential of HMM-GP for a wider applicability and transferability across different themes, some key highlights covering the application of HMM-GP in the area of synthetic energy demand synthesis (one/five minutely highly stochastic annual energy demand time series) will also be presented.

Fingerprint

time series
modeling
simulation
stock market
bioinformatics
human behavior
loess
streamflow
decomposition
river
energy demand
distribution

Cite this

Patidar, S. (2018). Stochastic modelling framework for synthetic time series simulation. RSS 2018 International Conference , Cardiff, Wales, United Kingdom.
Patidar, Sandhya. / Stochastic modelling framework for synthetic time series simulation. RSS 2018 International Conference , Cardiff, Wales, United Kingdom.
@conference{9707e205a34f4158b8d9a8f6a6cc764d,
title = "Stochastic modelling framework for synthetic time series simulation",
abstract = "Quite often, observed records (time-series) are not long enough to extract reliable statistics to understand the variability and uncertainty in the patterns and the occurrences of all significant rare events. One possible solution to such issues is to develop robust data-centric modelling techniques for artificially generating realistically possible synthetic sequences based on the historical dataset and within the acceptable statistical errors. The Hidden Markov Model (HMM) is popular stochastic modeling approach and has been applied successfully to model a range of complex processes, such as bio-informatics, speech, molecular evolution, the stock market, natural languages, human and animal behaviour. This presentation is aimed to present the systematic organisation of an HMM-based modeling framework (referred in this presentation and relevant publications as HMM-GP) that couples STL: a Seasonal-Trend decomposition procedure based on Loess and the extreme values distribution (such as Generalised Pareto Distribution), to develop a robust modeling schematic for artificially synthesizing univariate time series. Considering the environmental statisticians as the key audience, the presentation will be organized to demonstrate the step-by-step development of the HMM-GP modeling schematic, including a thorough investigation to examine its suitability for simulating highly stochastic time series of streamflow at a much finer temporal resolution of 15 minutes. A robust validation of the proposed HMM-GP schematic would be presented by conducting an extensive comparison of various statistical characteristics of the observed records with the synthetically simulated flow time series across four hydrologically distinct case-studies River in the UK, namely Don, Nith, Dee, and Tweed. Further, for the benefit of the general audience interested in potential of data-centric modelling techniques and to illustrate the potential of HMM-GP for a wider applicability and transferability across different themes, some key highlights covering the application of HMM-GP in the area of synthetic energy demand synthesis (one/five minutely highly stochastic annual energy demand time series) will also be presented.",
author = "Sandhya Patidar",
year = "2018",
month = "9",
day = "5",
language = "English",
note = "RSS 2018 International Conference : The conference for all statisticians and data scientists ; Conference date: 03-09-2018 Through 06-09-2018",
url = "https://events.rss.org.uk/rss/frontend/reg/thome.csp?pageID=57555&ef_sel_menu=1152&eventID=194",

}

Patidar, S 2018, 'Stochastic modelling framework for synthetic time series simulation' RSS 2018 International Conference , Cardiff, Wales, United Kingdom, 3/09/18 - 6/09/18, .

Stochastic modelling framework for synthetic time series simulation. / Patidar, Sandhya.

2018. RSS 2018 International Conference , Cardiff, Wales, United Kingdom.

Research output: Contribution to conferenceOther

TY - CONF

T1 - Stochastic modelling framework for synthetic time series simulation

AU - Patidar, Sandhya

PY - 2018/9/5

Y1 - 2018/9/5

N2 - Quite often, observed records (time-series) are not long enough to extract reliable statistics to understand the variability and uncertainty in the patterns and the occurrences of all significant rare events. One possible solution to such issues is to develop robust data-centric modelling techniques for artificially generating realistically possible synthetic sequences based on the historical dataset and within the acceptable statistical errors. The Hidden Markov Model (HMM) is popular stochastic modeling approach and has been applied successfully to model a range of complex processes, such as bio-informatics, speech, molecular evolution, the stock market, natural languages, human and animal behaviour. This presentation is aimed to present the systematic organisation of an HMM-based modeling framework (referred in this presentation and relevant publications as HMM-GP) that couples STL: a Seasonal-Trend decomposition procedure based on Loess and the extreme values distribution (such as Generalised Pareto Distribution), to develop a robust modeling schematic for artificially synthesizing univariate time series. Considering the environmental statisticians as the key audience, the presentation will be organized to demonstrate the step-by-step development of the HMM-GP modeling schematic, including a thorough investigation to examine its suitability for simulating highly stochastic time series of streamflow at a much finer temporal resolution of 15 minutes. A robust validation of the proposed HMM-GP schematic would be presented by conducting an extensive comparison of various statistical characteristics of the observed records with the synthetically simulated flow time series across four hydrologically distinct case-studies River in the UK, namely Don, Nith, Dee, and Tweed. Further, for the benefit of the general audience interested in potential of data-centric modelling techniques and to illustrate the potential of HMM-GP for a wider applicability and transferability across different themes, some key highlights covering the application of HMM-GP in the area of synthetic energy demand synthesis (one/five minutely highly stochastic annual energy demand time series) will also be presented.

AB - Quite often, observed records (time-series) are not long enough to extract reliable statistics to understand the variability and uncertainty in the patterns and the occurrences of all significant rare events. One possible solution to such issues is to develop robust data-centric modelling techniques for artificially generating realistically possible synthetic sequences based on the historical dataset and within the acceptable statistical errors. The Hidden Markov Model (HMM) is popular stochastic modeling approach and has been applied successfully to model a range of complex processes, such as bio-informatics, speech, molecular evolution, the stock market, natural languages, human and animal behaviour. This presentation is aimed to present the systematic organisation of an HMM-based modeling framework (referred in this presentation and relevant publications as HMM-GP) that couples STL: a Seasonal-Trend decomposition procedure based on Loess and the extreme values distribution (such as Generalised Pareto Distribution), to develop a robust modeling schematic for artificially synthesizing univariate time series. Considering the environmental statisticians as the key audience, the presentation will be organized to demonstrate the step-by-step development of the HMM-GP modeling schematic, including a thorough investigation to examine its suitability for simulating highly stochastic time series of streamflow at a much finer temporal resolution of 15 minutes. A robust validation of the proposed HMM-GP schematic would be presented by conducting an extensive comparison of various statistical characteristics of the observed records with the synthetically simulated flow time series across four hydrologically distinct case-studies River in the UK, namely Don, Nith, Dee, and Tweed. Further, for the benefit of the general audience interested in potential of data-centric modelling techniques and to illustrate the potential of HMM-GP for a wider applicability and transferability across different themes, some key highlights covering the application of HMM-GP in the area of synthetic energy demand synthesis (one/five minutely highly stochastic annual energy demand time series) will also be presented.

UR - https://events.rss.org.uk/rss/media/uploaded/EVRSS/event_194/2_Wednesday-PM-Sandhya-Patidar.pdf

M3 - Other

ER -

Patidar S. Stochastic modelling framework for synthetic time series simulation. 2018. RSS 2018 International Conference , Cardiff, Wales, United Kingdom.