148 research outputs found
A Comparative Analysis of Fuzzy Logic Control and Model Predictive Control in Photovoltaic Maximum Power Point Tracking
No full textOperating photovoltaic (PV) panels at the maximum power point (MPP) is of paramount importance for significantly enhancing the overall efficiency of the energy conversion process while simultaneously minimizing the payback period associated with the investment in such a renewable energy system. However, it is imperative to acknowledge that the voltage and current characteristics exhibited by photovoltaic panels are inherently nonlinear, displaying a strong dependence on a myriad of environmental factors, including but not limited to temperature fluctuations and variations in solar irradiance levels. This scholarly article endeavors to provide a comprehensive comparative analysis of two advanced control methodologies, namely Fuzzy Logic Control (FLC) and Model Predictive Control (MPC), specifically in the context of implementing Maximum Power Point Tracking (MPPT) techniques that are applicable to photovoltaic generation systems reliant on power converters. While a plethora of MPPT strategies have been documented extensively within the scientific literature, this investigation is particularly concentrated on FLC owing to its remarkable rapid response capabilities and its inherent robustness against fluctuations in circuit parameters that may occur during operation. For the purposes of comparison, the study also delves into the characteristics of MPC, which is well-regarded for its superior predictive capabilities and its adeptness in managing complex multi-variable control scenarios that frequently arise in practical applications. Additionally, a Proportional-Integral (PI) control strategy has been strategically employed within the framework of this study to ensure that the desired current and voltage levels are effectively maintained throughout the battery charging process. The primary objective of the aforementioned control strategies is to optimize the operational performance of the PV panel at the MPP, while concurrently supplying appropriate current and voltage levels necessary for efficient battery charging, thereby facilitating rapid charging, minimizing energy losses, and ultimately prolonging the life cycle of the battery system
Localization in Wireless Sensor Networks
No full textOne of the most fundamental aspects of wireless sensor networking based
applications is that they are either designed to monitor physical quantities,
observe various phenomena, disseminate useful information to autonomous
or semi-autonomous agents, or simply gather information in
their surrounding environment. The collected information may be used
by a cyber-physical system or transmitted via a data network to a remote
location for subsequent data processing. In both cases, the information
can become meaningless if the current location of the sending
sensor node is not known or the reported information or observation is
not accurately location stamped. In addition to this, there are certain
tracking applications, monitoring applications, and geographical routing
protocols that put a stringent demand that the location of sensor nodes
should be known a priori. This work proposes a distributed localization
algorithm that describes how a small sub-region in a sensing eld can
construct a spatial map of the locations of all the neighbouring nodes
based on inter-node distances and how each sub-region can then stitch
its own map with those of all other sub-regions in its close proximity with
the outcome that the collection of stitched maps forms a consistent coordinate
system. The proposed localization algorithm employs concepts
of range lookup, multidimensional scaling, and least-squares tting to
compute locations of static sensor nodes. The proposed algorithm can
compute relative coordinates without the use of any anchor nodes and
is also capable of converting the relative coordinates into absolute coordinates
if a certain minimum number of anchor nodes become available
at a later stage. The proposed localization scheme is only one component
of a proposed framework which aims to enhance road tra c safety
by employing static roadside sensors. In addition to the localization service,
three more components have been proposed for the road tra c
safety framework namely a road segment surveillance scheme to detect
vehicles on two-way roads, an adaptive data forwarding scheme to route
data among roadside sensors using reinforcement learning, and a reverse
forwarding scheme to deliver road condition information or warning messages
from static roadside sensors to vehicles approaching a designated
region-of-interest
Secondary Voltage Recovery Control for Multiple Heterogeneous Energy Storage Systems
No full textThis study presents a multi-agent framework incorporating two control strategies designed
to achieve secondary voltage recovery (SVR) for distributed heterogeneous energy
storage systems (HESSs) in a DC microgrid. The proposed framework addresses
the challenges of balancing State of Charge (SoC), power, and energy levels across
multiple HESS units to maintain a stable DC bus voltage despite fluctuating loads.
The first control method applies an advanced consensus-based SVR controller that
leverages the virtual resistance/capacitor droop (VRD/VCD) technique. This SVR
approach uses a consensus algorithm to evenly distribute SoC, power, and energy
among various HESS units, enhancing the system’s ability to maintain voltage stability
across changing operational conditions. By aligning the SoC levels of each HESS,
the SVR control method prevents any individual storage unit from experiencing excessive
charge or discharge, promoting reliability and longevity. The VRD/VCD droop
control further complements the SVR by fine-tuning voltage distribution, achieving a
cooperative balance among HESS units. Simulation results indicate that this method
effectively recovers and stabilizes the DC bus voltage while supporting efficient energy
distribution.
The second control approach utilizes an online deep reinforcement learning (DRL) algorithm,
specifically the deep deterministic policy gradient (DDPG), to address voltage
regulation dynamically. By directly generating control actions based on real-time
observations of the DC bus voltage, DDPG learns and adapts to varying system conditions,
providing a flexible and data-driven alternative to traditional control methods.
This approach enables continuous voltage regulation without requiring predefined consensus
rules, allowing the HESS network to respond adaptively to rapid load changes.
A comparative analysis of simulation results demonstrates the effectiveness of both
methods. The SVR control strategy achieves reliable SoC and power distribution,
ensuring steady-state voltage recovery, while the DDPG-based approach offers adaptability
and can quickly react to unpredictable fluctuations. The findings highlight the
complementary strengths of each method, with SVR providing robust performance under
steady conditions and DDPG excelling in dynamic environments where real-time
responsiveness is essential
Intelligent Control Strategies for Efficient Building Energy Management Systems (BEMS)
No full textHeating, Ventilation and Air Conditioning (HVAC) systems comprise significant proportion of energy consumption in buildings. This, therefore, warrants for developing energy-efficient control strategies for HVAC systems without compromising the occupants’ overall comfort. The design of such intelligent control strategies is critically dependent upon the accuracy of building thermal models; which in turn depends upon the proper knowledge and/or estimation of model parameters. In this research, a set of novel intelligent algorithms have been developed for the HVAC systems, with the aim of enhancing the efficiency of existing Building Energy Management Systems (BEMS).
Due to the distinct nature of the areas involved, this research is carried out in three phases. The first phase focuses on estimating the thermal parameters of a building. Estimation of building model parameters, particularly the thermal parameters, is a challenging task due to the non-linear nature of building thermal dynamics. Therefore, during this initial phase, a meta-heuristic based algorithm is proposed to estimate the thermal parameters of a building. The parameter estimation problem is formulated as a single-objective optimization problem considering the thermal model of the building, developed in EnergyPlus™, following black-box identification strategy. Effectiveness of the proposed method is demonstrated using measured energy consumption data from a real building. The results illustrate that the proposed methodology could effectively estimate the building thermal parameters. Thermal models identified using this methodology could be utilized to develop efficient Building Energy Management Systems (BEMS), by designing model based controllers for HVAC systems.
During the next phase of the research, a novel Multi-Objective Optimization (MOO) framework is proposed, using Non-dominated Sorting Genetic Algorithm (NSGA)-II, for optimizing the overall performance of forced extraction systems used in kitchen environment. The target of the proposed framework is defined to achieve a balance (trade-off) among three conflicting objectives: minimization of discomfort, minimization of energy consumed by the extraction system, minimization of the gradient in temperature and CO2 profiles within kitchen, during extraction system operation. These objectives are formulated in terms of criterion (objective) functions, using certain variables including: ambient environmental conditions and target/reference signals (set-points). The balance between the objectives is achieved by determining the optimal values of these reference signals. Performance of the proposed MOO framework is simulated by using the best optimal values of the reference signals i.e., the knee-point solution, evaluated by the framework in the Computational Fluid Dynamics (CFD) model of the kitchen environment, developed in FloVENT™. This model is validated against the experimental data collected from a real kitchen environment (test room). Comparison of the simulated results between optimal fan operation (based on our MOO framework) and the conventional fan operation show that the optimal fan operation is able to maintain an overall balance between the comfort, fan energy consumption and the gradient (in temperature and CO2 concentration) within the kitchen environment. These results, presented in the thesis, insinuate the significance of heuristic based approaches in achieving intelligent control strategies for energy efficient BEMS.
In the final phase of the research, the focus has been towards incorporating the preferences of the Decision Makers (DMs) i.e., users, in the MOO framework for having user friendly and more effective BEMS. Therefore, a MOO framework is proposed which focuses on balancing two conflicting objectives: energy consumption of Air Handling Units (AHUs) and thermal comfort, in buildings. The preferences of the DM are accommodated using two Multi-criteria Decision Making (MCDM) techniques: Conventional Weighted Aggregation (CWA) and ϵ-constraint method. Performance of the framework is examined by utilizing the real-time weather data of Auckland, New Zealand. The results of investigation clearly show that the proposed framework is able to successfully balance the AHU energy consumption with the thermal comfort, while fulfilling the DM’s preferences. This conveys the applicability of the framework in existing Demand Response (DR) based algorithms, used in the area of smart grids
Nonlinear Modelling of Converters using Computational Intelligence Technique
No full textFull Text is available to authenticated members of The University of Auckland only.Semiconductor technology has been developed rapidly during the decades which allows power electronic converters to be applied and installed in many different fields. In order to achieve a goal of deploying and controlling these kinds of converters effectively in real cases, it is necessary to analyze and develop mathematical models that can describe their dynamic and static behaviours. System parameters (e.g., inductance, capacitance) are needed among the most known models.
Besides, such kind of models does not take into account the discrete variance phenomenon in system parameters caused by external factors such as temperature changes. The project aims to take advantage of nonlinear system identification technology in order to solve these problems. This project will develop and apply particle swarm-based methods to identify nonlinear models for converters
The Event-Triggered Scheme for Networked Control Systems
No full textThe topic on Networked Control System (NCS) has received significant attention from the control community over the years. With the advent of mesh-type
communication network, the implementation and control of an industrial plant
becomes easy and cost-efficient. Yet, the main concerns evolved from NCS are
the limited network bandwidth and node energy, and the risks of facing cyberattacks and increased time delays. Therefore, real-time scheduling of control tasks
is a significant topic in the control of NCSs, and has attracted much attention
from researchers around the globe. One possible solution to such problems is
through designing effective event-triggered schemes (ETSs). In this investigation,
several event-triggered control strategies have been developed which could effectively overcome some of the limitations of the NCS, including time delays and
cyber-attacks.
During the first phase of the research, an event-triggered H∞ controller is developed for linear networked control systems which are subjected to the denialof-service (DoS) attacks. As one of the most common type of cyber-attacks, the
DoS attacks block the communication link, which may lead to fatal damage to the
control system. The proposed event-triggered scheme includes an off-time in order to reduce the consumption of network resources. This scheme guarantees a
prescribed minimum inter-triggering time and therefore avoids the Zeno problem. Sufficient conditions for the existence of an event-triggered controller, which
ensures the exponential stability of the closed-loop system with desired H∞ performance, are derived, and are solved using linear matrix inequalities (LMIs). The
effectiveness of the proposed method is examined considering a real communication network based on the ZigBee protocol.
In the next phase of the research, an integral-based event-triggered scheme
(IETS) is proposed for nonlinear NCSs. This method uses the integral of the system states, and past triggered data over a period of time. With the proposed IETS,
the integral event-triggered networked system essentially becomes a distributed
delay system. The controller for this system is designed using the Bessel-Legendre
inequalities. Sufficient conditions for the asymptotic stability of the closed-loop
system are derived, and solved using LMIs. The effectiveness of the proposed
scheme and the advantages of IETS over some other existing ETSs, have been
demonstrated considering the synchronization problem associated with delayed
neural networks (NNs). This neural network is configured to behave as a chaotic
system and is used for the application of image encryption, where a novel encryption algorithm is proposed to enhance the quality of the encryption process.
Following the development of IETS, in the last phase of the research, a novel
discrete event-triggered scheme (DETS) is developed for nonlinear NCSs. The
proposed DETS uses both the current and past samples to determine the next trigger, unlike the sampled-data ETS that uses only the current sample. The DETS
is employed in a dual setup for two network channels to significantly reduce redundant data transmission. A dynamic output feedback controller (DOFC) is designed. Stability criteria of the synchronisation error system are derived based on
the Lyapunov-Krasovskii functional method, and the co-design of the DOFC and
DETS parameters are accomplished using the Cone-complementarity linearization
(CCL) approach. The effectiveness and advantages of the proposed method are
illustrated considering an example of chaotic Chua’s circuit. A comparative investigation is carried out on the proposed DETS and the existing memory-based
event-triggered scheme (METS) in the literature.
Throughout the thesis, the performance of the proposed control strategies are
evaluated and tested within practical operating limits. Although the proposed
event-triggered mechanisms represent promising alternatives to the already established ones, there is room for improvement in the future work, in terms of both
theory and applications
The Handling Qualities Assessment of Novel Personal Air Vehicle Systems
No full textThere is a requirement for rigorous analytical tools that can be used to analyse the flying and handling qualities deficiencies of novel personal-air-vehicles (PAV), prior to experimental flight testing. In more recent years, ground-based simulation has become a critical component of aircraft-level design and a core requirement of the procuring activity. For small footprint, personal, air-vehicle systems the analysis of pilot workload has often relied, exclusively, on pilot-in-the-loop flight test and full immersion simulation experiments. In order to facilitate the preliminary stages of aircraft design there is need for an off-line, quantitative, rating method that can be brought to bear in the absence of an adequate pilot representation. The prediction of aircraft handling qualities is complicated, however, by the need to accurately assess pilot effort. This thesis provides a comprehensive analysis of the control design and handling qualities of a manually controlled, novel, PAV. The handling qualities of the vehicle are assessed, first by reference to the Rotorcraft Aeronautical Design Standard (ADS33-E-PRF), then by pilot/vehicle analysis using models of pursuit, compensatory, and regressive human pilot behaviour. In the latter case, handling qualities levels, pilot-induced oscillation rating levels, and tracking performance are predicted. Consideration is given to both linear and nonlinear pilot/vehicle behaviour. Approximate bounds on the off-nominal linear vehicle model stability derivatives are explored and the expected operational and service flights envelope, temporal and corridor constraints examined. Finally, validation of the proposed pilot modelling techniques and rating criterion are presented, and the use of traditional rotary and fixed-wing HQ rating boundaries are examined with application to a prototype, rotary-wing PAV system. The research work presented in this dissertation provides a means of (1) identifying the feasible design space of a unique aircraft based upon predicted and expected levels of handling qualities and performance, and (2) determining compliance with both civil and (where required) military requirements for manned operation of PAV systems. 2 The research presented may provide a useful template for assessing the handling qualities of more novel, personal air-vehicle concepts prior to pilot-in-the-loop simulation or flight testing
Energy-efficient communication algorithms for wireless sensor networks
No full textThis thesis is motivated to tackle the problem of decreasing sensor nodes' energy consumptions in wireless communications and extending the lifetime of wireless sensor networks (WSN) in providing satisfactory services of data sensing and transmission. To this end, studies in this thesis are focused on theoretical aspects, aiming at developing communication algorithms in the network layer to effectively organize sensor nodes and in the physical payer to directly reduce a node's energy expenditure in wireless communications, respectively. Via simulations the theoretically developed algorithms and the performances of the sensor networks based on the developed algorithms are evaluated. As for the network layer algorithm, the Slotted Waiting period Energy-Efficient Time driven (SWEET) clustering algorithm is developed. The SWEET algorithm aims at organizing sensor nodes in the form of clusters where energy-rich Cluster Head nodes are selected and distributed evenly over the network area to coordinate the communications among cluster member nodes. The SWEET algorithm uses the distribution of nodes' remaining energies to achieve its design goal. To organize densely deployed sensor nodes, the SWEET algorithm is decentralized using the distribution of the residual energies of nodes in a node's neighbourhood area. The empirical probability density function of neighbourhood node energy distribution is obtained via Hello Message Exchange (HME). The procedure of HME is carried out based on the Birthday protocol and the Carrier Sensing Mini-Slot algorithm which is modified from the solution for the initialization problem. The time and the node energy required for the procedure of HME based on the considered methods are investigated with respect to the message exchange sufficiency. By simulations, the effectiveness of the SWEET and the decentralized SWEET algorithm is confirmed. Also by simulations, the performances of the networks based on these two algorithms are evaluated. The simulation results show that the developed algorithms outperform several competing clustering algorithms in significantly improving the network lifetime and data capacity at various cluster radii and network node densities. As for the physical layer algorithm, chip-interleaving signal processing is employed to save a node's energy in transmitting data in fading channel. The Bit Error Rate (BER) expressions of two Direct Sequence Code Division Multiple Access (DS-CDMA) systems with embedded chip interleaving are derived to determine how effective the chip interleaving is in decreasing the signal power loss due to the channel fading. With the derived BER expressions, the energy savings of networks based on sensor nodes that use the direct sequence spread spectrum (DSSS) transceivers with or without embedded chip interleaving are analyzed. The considered DSSS transceivers are compliant with the physical layer specifications in the IEEE 802.15.4 standard. The randomly deployed nodes are organized based on the studied clustering algorithms, in particular the SWEET algorithm. By simulations, the correctness and accuracy of the developed BER expressions are confirmed. Simulation results also show that the lifetime of a cluster-based WSN can be extended to a great extent when the chip interleaved transceivers are used by nodes to transmit data in flat Rayleigh fading channel. In summary, the energy efficiency of a sensor network can be significantly improved by utilizing the SWEET algorithm and the chip interleaving technique, individually or in combination
Selection of base wavelet and classifier for accurate identification of power quality events
No full textThis report studies comprehensively the Power Quality (PQ) identification problem and suggests the optimum combination of base wavelet and Induction Algorithm (IA) which would give highest classification accuracy. Although this problem has been studied by various researchers in recent past, the selection of appropriate base wavelet and IA, which would give better classification accuracy, have received comparatively less attention. This report bridges this gap by investigating the classification performance of 110 wavelets and 7 well known IAs across various noise levels using over 3500 PQ events generated as per IEEE Standard 1159. The results of this investigation demonstrate that the choice of base wavelet significantly affects the classification performance. Further, it was observed that there does not exist any common wavelet which provides optimum performance with all the IAs at various noise levels. In contrast, each IA gives maximum accuracy provided it is combined with a specific wavelet. The robustness of IA against noise is studied which establishes that the simple IAs, such as Decision Tree (DT) and Naive-Bayes (NB), are more robust against noise compared to other intricate IAs. Finally, several recommendations are drawn for the selection of base wavelet and IA which yields best possible accuracy
Identification of nonlinear systems and quantised feedback control of networked systems
No full textThe research carried out in this study addresses some key issues associated with the identification of
manufacturing processes and control of networked systems and has therefore been divided into two
distinct phases.
In the first phase of this research, a machine learning based predictive model has been developed for
automated fibre placement (AFP) based composite manufacturing process. Since the manufacturing of
AFP composites is both expensive and time-consuming, the available data samples are very less. The
identification of these processes, therefore, comes under the purview of small data learning problem.
This study first solves this problem by proposing the optimum combination of virtual sample generation
(VSG) methods and various machine learning (ML) tools, which would give the highest learning
accuracy. It is established that the trend similarity assessment (TSA) method, when combined with the
back-propagation neural network (BPNN), gives the highest learning accuracy.
A direct predictive model, based on BPNN, is developed using the the virtual samples generated
from the TSA method, which accurately encodes the physics of the manufacturing process. This model
can be used to investigate how the changes in critical processing conditions alter the outputs of the manufacturing
process. Often, in manufacturing, it is of interest to know the accurate values of different
input variables which would give the desired characteristics of the output variables. This is an inverse
problem, and this study therefore develops a machine learning based, inverse predictive model to determine
the input conditions corresponding to the desired output characteristics of the manufacturing
process. The efficacy of the developed predictive models (both direct and inverse) have been established
considering varieties of experimental data of AFP based composite laminates.
The second phase of this thesis focuses on designing effective controllers for networked control systems
(NCS), where the systems are controlled by sending information (data packets) through a communication
network (often unreliable) such as internet. The present study proposes several novel quantised
control algorithms, based on delta-modulators, to effectively control both linear and nonlinear systems
under various imperfections of the communication network.
D-Modulator based quantised state-feedback controller and output-feedback controller are designed
for linear networked systems. The D-M offers many advantages, which include lower design complexity,
lower cost, and less noisy in contrast to some of the existing quantisers. The stability conditions of the
control system are derived, and the performance of these controllers are investigated experimentally by considering a real ZigBee protocol based wireless communication network.
A D-M based quantised state-feedback controller is developed for nonlinear networked systems,
where the nonlinear system is represented by T-S fuzzy model. For a prescribed quantisation error,
the gains of the state-feedback controller and the quantiser are derived. The stability conditions of the
quantised closed loop system are established using linear matrix inequalities (LMIs), both in continuous
and discrete-time domains. The performance of these controllers are experimentally validated using a
practical communication network
- …
