A neural network approach to control performance assessment

Abstract

Purpose -The purpose of this paper is to present a neural network approach to control performance assessment. Design/methodology/approach -The performance index under study is based on the minimum variance control benchmark, a radial basis function network (RBFN) is used as the pre-whitening filter to estimate the white noise sequence, and a stable filtering and correlation analysis method is adopted to calculate the performance index by estimating innovations sequence using the RBFN pre-whitening filter. The new approach is compared with the auto-regressive moving average model and the Laguerre model methods, for both linear and nonlinear cases. Findings -Simulation results show that the RBFN approach works satisfactorily for both linear and nonlinear examples. In particular, the proposed scheme shows merits in assessing controller performance for nonlinear systems and surpasses the Laguerre model method in parameter selection. Originality/value -A RBFN approach is proposed for control performance assessment. This new approach, in comparison with some well-known methods, provides satisfactory performance and potentials for both linear and nonlinear cases.

Performance Assessment of Predictive Control—A Survey

“…• minimum variance and normalized Harris index [75], Control Performance Index [76], and other variance benchmarking methods [77]; • all types of the model-based measures [78], derived from close loop identification, such as aggressive/oscillatory and sluggishness indexes [79]; • frequency methods starting from classical Bode, Nyquist and Nichols charts with phase and gain margins [69] followed by deeper investigations, such as with the use of Fourier transform [80], sensitivity function [81], reference to disturbance ratio index [82], and singular spectrum analysis [83]; and • alternative indexes using neural networks [84] or support vector machines [85].…”

Control Rehabilitation Impact on Production Efficiency of Ammonia Synthesis Installation

“…Within such a long history and wide research scope, different domain groups of methods were evaluated: time domain indexes based on step response: undershoot, overshoot, response times, area index, output index, R-index, idle index, etc. ; time-series-based indexes: mean square error (MSE), integral of absolute error (IAE), amplitude index (AMP); statistical factors utilizing different probabilistic distribution function (standard deviation, variance, skewness, kurtosis, scale or shape, ...); minimum variance and model-based measures; novel indexes using wavelets, Fourier transform, orthonormal functions, singular spectrum analysis, neural networks, Hurst exponent, persistence measures, entropy, ...; and business KPIs expressed in monetary terms. …”

Control Quality Assessment for Processes With Asymmetric Properties and its Application to pH Reactor

“…Data-driven approaches use only process time series. Developed methods use different domains such as: integral time indexes [27], correlation methods [28], statistical distributions [29], frequency domain approach [30], neural networks [31], Hurst exponent [32], persistence measures [33], entropy [34], specific business KPIs [35], etc.…”