11 September 2014
Resolving the trade-offs between suspension travel, ride comfort, road holding, vehicle handling and power consumptions is the primary challenge in designing Active-Vehicle-Suspension-Systems (AVSS). Controller tuning with global optimization techniques is proposed to realise the best compromise between these con icting criteria. Optimization methods adapted include Controlled-Random-Search (CRS), Differential-Evolution (DE), Genetic-Algorithm (GA), Particle-Swarm-Optimization (PSO) and Pattern-Search (PS). Quarter-car and full-car nonlinear AVSS models that incorporate electrohydraulic actuator dynamics are designed. Two control schemes are proposed for this investigation. The first is the conventional Proportional-Integral-Derivative (PID) control, which is applied in a multi-loop architecture to stabilise the actuator and manipulate the primary control variables. Global optimization-based tuning achieved enhanced responses in each aspect of PID-based AVSS performance and a better resolve in con icting criteria, with DE performing the best. The full-car PID-based AVSS was analysed for DE as well as modi ed variants of the PSO and CRS. These modified methods surpassed its predecessors with a better performance index and this was anticipated as they were augmented to permit for e cient exploration of the search space with enhanced exibility in the algorithms. However, DE still maintained the best outcome in this aspect. The second method is indirect adaptive dynamic-neural-network-based-feedback-linearization (DNNFBL), where neural networks were trained with optimization algorithms and later feedback linearization control was applied to it. PSO generated the most desirable results, followed by DE. The remaining approaches exhibited signi cantly weaker results for this control method. Such outcomes were accredited to the nature of the DE and PSO algorithms and their superior search characteristics as well as the nature of the problem, which now had more variables. The adaptive nature and ability to cancel system nonlinearities saw the full-car PSO-based DNNFBL controller outperform its PID counterpart. It achieved a better resolve between performance criteria, minimal chatter, superior parameter sensitivity, and improved suspension travel, roll acceleration and control force response.
Bishop, Griffin R.
13 May 2020
Semantic segmentation methods using deep neural networks typically require huge volumes of annotated data to train properly. Due to the expense of collecting these pixel-level dataset annotations, the problem of semantic segmentation without ground-truth labels has been recently proposed. Many current approaches to unsupervised semantic segmentation frame the problem as a pixel clustering task, and in particular focus heavily on color differences between image regions. In this paper, we explore a weakness to this approach: By focusing on color, these approaches do not adequately capture relationships between similar objects across images. We present a new approach to the problem, and propose a novel architecture that captures the characteristic similarities of objects between images directly. We design a synthetic dataset to illustrate this flaw in an existing model. Experiments on this synthetic dataset show that our method can succeed where the pixel color clustering approach fails. Further, we show that plain autoencoder models can implicitly capture these cross-instance object relationships. This suggests that some generative model architectures may be viable candidates for unsupervised semantic segmentation even with no additional loss terms.
De Wet, Anton Petrus Christiaan
11 September 2014
M.Ing. (Electrical And Electronic Engineering) / This dissertation describes the development of a system of Artificial Neural Networks that enables the incremental training of feed forward neural networks using supervised training algorithms such as back propagation. It is argued that incremental learning is fundamental to the adaptive learning behavior observed in human intelligence and constitutes an imperative step towards artificial cognition. The importance of developing incremental learning as a system of ANNs is stressed before the complete system is presented. Details of the development and implementation of the system is complemented by the description of two case studies. In conclusion the role of the incremental learning system as basis for further development of fundamental elements of cognition is projected.
A new methodology for analyzing and predicting U.S. liquefied natural gas imports using neural networksBolen, Matthew Scott 01 November 2005 (has links)
Liquefied Natural Gas (LNG) is becoming an increasing factor in the U.S. natural gas market. For 30 years LNG imports into the U.S. have remained fairly flat. There are currently 18 permit applications being filed in the U.S. and another 10 permit applications being filed in Canada and Mexico for LNG import terminals. The EIA (Energy Information Agency) estimates by 2025 that LNG will make up 21% of the total U.S. Natural Gas Supply. This study developed a neural network approach to forecast LNG imports into the U.S. Various input variables were gathered, organized into groups based on similarity, and then a correlation matrix was generated to screen out redundant variables. Since a limited number of data points were available I used a restricted number of input variables. Based on this restriction, I grouped the input variables into four different scenarios and then generated a forecast for each scenario. These four different scenarios were the $/MMBTU model, natural gas energy consumption model, natural gas consumption model and the energy stack model. The standard neural network approach was also used to screen the input variables. First, a correlation matrix determined which variables had a high correlation with the output, U.S. LNG imports. The ten most correlated input variables were then put into correlation matrix to determine if there were any redundant variables. Due to the lack of data points only the five most highly correlated input variables were used in the neural network simulation. A number of interesting results were obtained from this study. The energy stack model and the consumption of natural gas forecasted a non-linear trend in U.S. LNG imports, compared to the linear trend forecasted by the EIA. The energy stack model and consumption of natural gas model predicted that in 2025 U.S. LNG imports will be about 6.5 TCF, while the other three models prediction is about three times as less. The energy stack model is the most realistic model due its non-linear trend, when the rapid increase of LNG imports is going to occur, and the quantity of U.S. LNG imports predicted in 2025.
Neural networks approach towards determining Flax-Biocomposites composition and processing parametersMondol, Joel-Ahmed Mubashshar 16 November 2009 (has links)
This research introduces neural networks (NN) as a novel approach towards aiding biocomposite materials processing. At its core, the aim of the research was to investigate NN usage as a tool for advancing the field of biocomposites. Empirical data was generated for compression-molded flax fiber and High Density Polyethylene (HDPE) matrix based biocomposite materials. In an attempt to create the NN model, tensile strength, impact strength, hardness, bending strength, and density were provided to the NN as inputs. These inputs were processed through multiple layers of the NN, and contributed to the prediction of the composition (fiber loading percentage) and operating parameter (pressure in MPa) as output. In précis, NNs use was investigated to predict composition and operational parameter for biocomposites production when the desired mechanical properties of the biocomposites were available. Flax (Linum usitatissimum) fiber biocomposite boards were manufactured using chemically pretreated flax fiber and high density polyethylene (HDPE). After extensive preprocessing (combing and size reduction to 2 mm particles) and pretreatment regimen - flax fiber was mixed with HDPE and extruded using a laboratory scale single screw extruder. Extrudates generated from the extruder were again ground to 2 mm particles. Ground extrudates from different sample sets were exposed to a compression molding unit. The mold was put under two sets of pressures, (variable operating parameters) for all individual fiber loading. These boards were used to determine the mechanical properties tensile force, impact force, hardness, bending, and density. For verification and analysis of the mechanical properties, Microsoft Office Excel and a statistical software package SAS were used. After verification five different multilayer neural networks, i.e., cascade forward neural network, feedforward backpropagation neural network, neural unit (single layer, single neuron), feedforward time delay neural network and NARX, were trained and evaluated for performance. Ultimately, the feedforward backpropagation NN (FFBPNN) was selected as the most efficient. After rigorous testing, the FFBPNN trained by the TRAINSCG algorithm (Matlab ®) was selected to generate prediction results that were the most suitable, fast and accurate. Once the selection and training of the NN architecture was complete, biocomposite materials prediction was performed. From 9 separate input sets, NNs provided overall prediction error between 2 - and 4%. This was the same amount of error that was observed in the training of the neural network. It was concluded that the neural network approach for the experimental design and operational conditions were satisfied.
Spiečiaus intelekto taikymo finansų rinkose analizė ir optimizavimas / Analysis and optimization of swarm intelligent in financial marketsVasiliauskaitė, Vilma 23 June 2014 (has links)
Prekiaujant vertybiniais popieriais, svarbiausia yra priimti teisingą sprendimą: pirkti arba parduoti. Daugelis investuotojų prieš priimdami sprendimą atkreipia dėmesį į pasirinktos akcijos kainos kitimo grafiką ir vadovaujasi juo. Tačiau ne kiekvienas investuotojas galėtų tiksliai apibūdinti savo pasirinktą grafinį modelį. Problemos aktualumas - Prognozuoti rinkas yra pakankamai sudėtinga, pastebimas žymus akcijų kursų svyravimas. Ženklūs akcijų kursų pasikeitimai skaičiuojami ne per metus ar mėnesius, o dienomis ar net valandomis. Investitoriams, finansų analitikams finansinėse rinkose sunku dirbti. Spekuliavimas akcijomis aktyviose akcijų rinkose yra labai rizikingas, bet pelningas užsiėmimas. Pasiūlius sprendimo priėmimo metodą investavimo procesas techniniu požiūriu supaprastės ir nereikalaus didelių sąnaudų, bei gilių žinių, leis platesniam ratui žmonių įeiti į akcijų rinką. Problema – Sudėtingas akcijų rinkų prognozavimas, kadangi pastebimas žymus akcijų kursų svyravimas, todėl rizikinga spekuliuoti akcijomis aktyviose akcijų rinkose. Baigiamojo darbo objektas – sprendimo priėmimo metodas finansinių rinkų prognozėms atlikti, remiantis neuroniniais tinklais ir spiečiaus algoritmu. Baigiamojo darbo tikslas – Spiečiaus intelekto taikymo finansų rinkose analizė ir optimizavimas. / One of the central problems in financial markets is to make the profitable stocks trading decisions using historical stocks' market data. This paper presents the decision-making methodology which is based on the application of neural networks and swarm intelligence technologies and is used to generate one-step ahead investment decisions. In brief, the proposed methodology draws from the analysis of historical stock prices variations. The variations are passed to neural networks and the recommendations for the next day are calculated. The stocks with the highest recommendations are considered for further experimental investigations. The core idea of this algorithm is to select three best neural networks for the future investment decisions and to adapt the weights of other networks towards the weights of the best network. The experimental results presented in the paper show that the application of our proposed methodology lets to achieve better results than the average of the market. The theme of the Master’s degree paper is “Analysis and Optimization of Swarm Intelligent in Financial Markets”. The object of the Master’s degree paper is decision making method for financial markets, re neural network and swarm intelligence.
The work described in this thesis concerns the study of the evolution of simple forms of learning behaviour in artificial agents. Our interest in the phylogeny of learning has been developed within the theoretical framework provided by the "ecological approach" to the study of learning. The latter is a recent theoretical and methodological perspective which, contrary to that suggested by the classical approaches in animal and comparative psychology, has reconsidered the importance of the evolutionary analysis of learning as a species- niche-specific adaptive process, which should be investigated by employing the conceptual apparatus originally developed by J. J. Gibson within the context of visual perception. However, it has been acknowledged in the literature that methodological difficulties are hindering the evolutionary ecological study of learning. We argue that methodological tools - i. e., artificial agent based models - recently developed within the context of biologically-oriented cognitive science can potentially represent a complementary methodology to investigate issues concerning the evolutionary history of learning without losing sight of the complexity of the ecological perspective. Thus, the experimental work presented in this thesis contributes to the discussion on the adaptive significance of learning, through the analysis of the evolution of simple forms of associative learning in artificial agents. Part of the work of the thesis focuses on the study of the nature of the selection pressures which facilitate the evolution of associative learning. The results of these simulations suggest that ecological factors might prevent the selection from operating in favour of those elements of the "learning machinery" which, given the varying nature of the environment, are of potential benefit for the agents. Other simulations highlight the properties of the agent control structure and the characteristics of particular features of the ecology of the learning scenario which facilitate the evolution of learning agents
31 July 2015
Pattern recognition is a rapidly expanding area of research, with applications ranging from character recognition and component inspection to robotic guidance and military reconnaissance. The basic principle of image recognition is that of comparing the unknown image with many known reference images or 'filters', until a match is found. By comparing the unknown image with a large data bank of filters, the diversity of the application can be extended. The work presented in this thesis details the practical development of an optical disk based memory system as applied in various optical correlators for pattern recognition purposes. The characteristics of the holographic optical disk as a storage medium are investigated in terms of information capacity and signal to noise ratio, where a fully automated opto-mechanical system has been developed for the control of the optical disk and the processing of the information recorded. A liquid crystal television has been used as a Spatial Light Modulator for inputting the image data, and as such, the device characteristics have been considered with regard to processing both amplitude and phase information. Three main configurations of optical correlator have been applied, specifically an image plane correlator, a VanderLugt correlator, and an Anamorphic correlator. Character recognition has been used to demonstrate correlator performance, where simple matched filtering has been applied, subsequent to which, an improvement in class discrimination has been demonstrated with the application of the Minimum Average Correlation Energy filter. The information processing rate obtained as a result of applying 2D parallel processing has been shown to be many orders of magnitude larger than that available with comparable serial based digital systems.
Preece, Timothy Edward
No description available.
Page generated in 0.1125 seconds