Evolutionary optimisation for Volt-VAR power quality control

Boyi Bukata, Bala (2013) Evolutionary optimisation for Volt-VAR power quality control. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 2012bukataphd.pdf] PDF
Download (3MB)


With the more environmentally friendly smart grid initiatives during the past few years, intelligent operation and optimisation of the electricity distribution system have received an increasing attention in power system research worldwide. Power flow from the distribution substation to the customer can be optimised at Volt-Ampere-Reactive (VAR) level by reducing the reactive power. Distributed Generation (DG) and Renewable Energy Sources (RES) represent both the broadest potentials and the broadest challenges for intelligent distribution systems and smart grid control. In general, the flexibility envisaged by integrating RES during smart grid transformation is often surrounded by nonlinearities such as wave-form deformations caused by harmonic currents or voltages, which impliedly increase control system complexity. Therefore,
conventional controllers presently implemented need to be re-engineered in order to solve power quality (PQ) problems therein.

This work aims to improve the controllability of Distribution Static Compensators (DSTATCOMs) through the development of improved control systems using evolu-
tionary computation enabled design automation and optimisation. The resultant Volt-VAR Control (VVC) optimises PQ in the presence of nonlinearities and uncertainties. It also aims at increasing overall system’s sensitivity to unconsidered parameters in the
design stage like measurement noise, unmodelled dynamics and disturbances. This is otherwise known as the robustness of the system offering it with valuable potential for
future smart grids control, which are anticipated to present more nonlinearities due to virtual power plant (VPP) configuration. According to European Project FENIX, a
Virtual Power Plant (VPP) aggregates the capacity of many diverse Distributed Energy Resources (DER), it creates a single operating profile from a composite of the
parameters characterizing each DER and can incorporate the impact of the network on aggregate DER output.

To particularly solve PQ problems, two objectives are realised in this thesis. First, a non-deterministic evolutionary algorithm (EA) is adopted to generate optimum fuzzy logic controllers for DSTATCOMs. This design methodology extends the traditional computer-aided-design (CAD) to computer-automated-design (CAutoD), which provides a unified solution to diverse PQ problems automatically and efficiently. While realizing this objective, the prediction ability of the derivative term in a proportional and derivative (PD) controller is improved by placing a rerouted derivative filter in the feedback path to tame ensuing oscillations. This method is then replicated in a
fuzzy PD scheme and is automated through the capability of a “generational” tuning using evolutionary algorithm.

Fuzzy logic controllers (FLCs) are rule-based systems which are designed around a fuzzy rule base (RB) related through an inference engine by means of fuzzy implication and
compositional procedures. RBs are normally formulated in linguistic terms, in the form of if ...then rules which can be driven through various techniques. Fundamentally, the
correct choice of the membership functions of the linguistic set defines the performance of an FLC. In this context, a three rule-base fuzzy mapping using Macvicar-Whelan matrix has been incorporated in this scheme to reduce the computational cost, and to avoid firing of redundant rules. The EA-Fuzzy strategy is proven to overcome the limitation of conventional optimisation which may be trapped in local minima, as the optimisation problem is often multi-modal.

The second objective of the thesis is the development of a novel advanced model-free predictive control (MFPC) system for DSTATCOMs through a deterministic non-gradient algorithm. The new method uses its “look-ahead” feature to predict and propose solutions to anticipated power quality problems before they occur. A describing function augmented DSTATCOM regime is so arranged in a closed-loop fashion to locate limit cycles for settling the systems nonlinearities in a model-free zone. Predictive control is performed upon the online generated input-output data-set through the power of a non-gradient simplex algorithm. The strategy is to boycott the usage of a system model which is often based on gradient information and may thus be trapped in a local optimum or hindered by noisy data.

As a model-free technique, the resultant system offers the advantage of reduction in system modelling or identification, which is often inaccurate, and also in computational load, since it operates directly on raw data from a direct online procession while at the same time dealing with a partially known system normally encountered in a practical industrial problem. Steady-state and dynamic simulations of both control and simulation models in Matlab/Simulink environment demonstrate the superiority
of the new model-free approach over the traditional trial-and-error based methods. The method has been varified to offer faster response speed and shorter settling time
at zero overshoot when compared to existing methods.

A SimPowerSystems software simulation model is also developed to check experimental validity of the designs. Where specific PQ problems such as harmonics distortion,
voltage swells, voltage sags and flicker are solved. A noticeable record level of THD reduction to 0.04% and 0.05% has respectively been achieved. It is therefore safe to
recommend to the industry the implementation of this model-free predictive control scheme at the distribution level. As the distribution system metamorphoses into decen-
tralised smart grid featuring connectivity of virtual power plants mostly through power electronic converters, e.g., DSTATCOM, it stands to benefit from the full Volt-VAR
automated controllability of the MFPCs low control rate.

Based on CAutoD, the practical implementation of this technique is made possible through digital prototyping within the real-time workshop to automatically generate
C or C++ codes from Simulink, which executes continuous and discrete time models directly on a vast range of computer applications. Its overall wired closed-loop structure with the DSTATCOM would offer reliable and competitive advantages over its PID and SVC (CAD-based) counterparts currently being implemented through physical prototyping, in terms of; quick product-to-market pace, reduced hardwire size, small footprint, maintenance free as it is model-free (and automated), where pickling the controller timers and model contingencies are unnecessary as would be with the
conventional controllers. More importantly, the scheme performs the aforementioned control functions robustly at a high speed in the range of 0.005 → 0.01 seconds. High
enough to capture and deal with any ensuing PQ problem emanating from changes in customer’s load and system disturbances in an environmentally friendly, but less
grid-friendly renewable generators.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Keywords: CAutoD, Distribution network, DSTATCOM, Evolutionary computation, Fuzzy logic control, Model-free predictive control, PID, Power quality, Self-healing, Simplex algorithm,
Subjects: A General Works > AS Academies and learned societies (General)
T Technology > TK Electrical engineering. Electronics Nuclear engineering
Colleges/Schools: College of Science and Engineering > School of Engineering
Supervisor's Name: Li, Professor Yun and Acha, Professor Enrique
Date of Award: 2013
Depositing User: Mr Bala Boyi Bukata
Unique ID: glathesis:2013-3882
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 23 Jan 2013 16:48
Last Modified: 23 Jan 2013 16:51
URI: https://theses.gla.ac.uk/id/eprint/3882

Actions (login required)

View Item View Item


Downloads per month over past year