Identification of Multiple First-Order Continuous-Time Dynamic Models From Special Segments in Historical Data

Xiaotong Xing,Jiandong Wang,Poku Gyasi,Zijiang Yang

doi:10.1109/access.2020.2978555

Xiaotong Xing, Jiandong Wang + Show 2 more

Open Access

https://doi.org/10.1109/access.2020.2978555

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2020
Citations: 2	License type: CC BY 4.0

Affiliation: Shandong University of Science and Technology

Abstract

Continuous-time dynamic models are often required for controller design, process monitoring, and operation optimization. This paper proposes an approach to estimate unknown parameters of multiple first-order continuous-time dynamic models from special segments in historical data. The approach is composed by two main steps. First, special segments are defined as the ones with two ends in steady states and the middle part having significant amplitude variations in transient states; the special segments are found automatically by exploiting a piece-wise linear representation technique from a large amount of historical data samples. Second, static gains are estimated by solving a set of linear equations based on steady-state values of inputs and outputs; sums of time constants and delays are obtained by solving another set of linear equations based on integrals of model output errors from data samples in transient states; the sums are used as an optimization constraint for maximizing the fitness value between measured and simulated outputs, from which separated estimates of time constants and delays are yielded. Numerical examples are provided to illustrate the proposed approach and compare with existing ones.

Highlights

Continuous-time dynamic models are often required for controller design, process monitoring, and operation optimization in a systematic manner [1], [2]
This paper proposes an approach to estimate parameters of continuous-time dynamic systems based on special data segments extracted from historical data samples
The paper proposed an approach to estimate model parameters of multiple continuous-time dynamic systems based on special data segments extracted from historical data samples

Summary

INTRODUCTION

Continuous-time dynamic models are often required for controller design, process monitoring, and operation optimization in a systematic manner [1], [2]. This paper is inspired by a common practice that is often adopted by industrial engineers to estimate parameters of continuous-time dynamic models based on some special data segments. This paper proposes an approach to estimate parameters of continuous-time dynamic systems based on special data segments extracted from historical data samples. Doing so is able to yield accurate estimates of static gains from steady-state data samples, as well as time constants and delays from transient-state data samples; in addition, estimated model parameters can be validated in a convincing manner by looking at the measured and simulated outputs in the special data segments. Many continuous-time model identification approaches do not exploit data samples in steady and transient states separately; as a result, it is not easy to validate estimated model parameters in such a convincing manner.

PROBLEM DESCRIPTION

ESTIMATION OF MODEL PARAMETERS

ACQUISITION OF SPECIAL DATA SEGMENTS

STEPS OF THE PROPOSED APPROACH

EXAMPLES

CONCLUSION