Browsing by Author "Nyirenda, Juwa"

Now showing 1 - 7 of 7
  • Thumbnail Image
    Item
    Open Access
    Decision support for the production and distribution of electricity under load shedding
    (2016) Rakotonirainy, Rosephine Georgina; Durbach, Ian; Nyirenda, Juwa
    Every day national power system networks provide thousands of MW of electric power from generating units to consumers. This process requires different operations and planning to ensure the security of the entire system. Part of the daily or weekly operation system is the so called Unit Commitment problem which consists of scheduling the available resources in order to meet the system demand. But the continuous growth in electricity demand might put pressure on the ability of the generation system to sufficiently provide supply. In such case load shedding (a controlled, enforced reduction in electricity supply) is necessary to prevent the risk to system collapse. In South Africa at the present time, a systematic lack of supply has meant that regular load shedding has taken place, with substantial economic and social costs. In this research project we study two optimization problems related to load shedding. The first is how load shedding can be integrated into the unit commitment problem. The second is how load shedding can be fairly and efficiently allocated across areas. We develop deterministic and stochastic linear and goal programming models for these purposes. Several case studies are conducted to explore the possible solutions that the proposed models can offer.
  • Thumbnail Image
    Item
    Open Access
    Exploring the application of Natural Language Processing to scientific medical cannabis publications
    (2022) de Beer, James Charles; Nyirenda, Juwa
    Cannabis has become recognised internationally as a powerful medicinal plant. The explosion of clinical research on cannabis has made it difficult for researchers and medical professionals to keep up to date with new findings. Analyzing the large quantities of available text data using natural language processing and machine learning algorithms could improve the speed and accuracy at which cannabis research is processed, as well as expose hitherto unknown connections between cannabis compounds and the treatment of healtth conditions. In turn, this would help direct future research and clinical trials. This thesis aims to develop an appropriate method to extract the key connections between cannabis compounds, human physiology and disease from the existing medical literature. First, natural language processing techniques (such as document clustering and topic modelling, global vector word embeddings and supervised document classifiers) are used to group 500 journal articles from the general literature on cannabis according to broad research topics; analyse the interaction between cannabis compounds, human physiology and diseases; and train a classifier to classify unseen documents. Second, the connections generated through this quantitative process are assessed qualitatively against those in a manual dataset of research findings from more than 500 studies collated over a number of years and provided by a medical company specialising in cannabis research. The results indicate that the methods developed were able to effectively and accurately demonstrate conenction between cannabis plant compounds and diseases. Hence, the working code accurately reproduced the results of manual analysis. This was shown by the close similarity of ranked key word to diseases. The unsupervised methods were able to effectively cluster and model topic distributions between the data to group documents by topic, while the supervised learning methods were able to accurately train models based on these suggestions, thereby solving a real-world practical problem in data management and analysis.
  • Thumbnail Image
    Item
    Open Access
    Investigating automated bird detection from webcams using machine learning
    (2022) Mirugwe, Alex; Nyirenda, Juwa; Dufourq, Emmanuel
    One of the most challenging problems faced by ecologists and other biological researchers today is to analyze the massive amounts of data being collected by advanced monitoring systems such as camera traps, wireless sensor networks, high-frequency radio trackers, global positioning systems, and satellite tracking systems being used today. It has become expensive, laborious, and time-consuming to analyze the large datasets using manual and traditional statistical techniques. Recent developments in the field of deep learning are showing promising results towards automating the analysis of these extremely large datasets. The primary objective of this study is to test the capabilities of the state-of-the-art deep learning architectures to detect birds in the webcam captured images. A total of 10592 images were collected for this study from the Cornell Lab of Ornithology live stream feeds situated in six unique locations in United States, Ecuador, New Zealand, and Panama. To achieve the main objective of the study, two convolutional neural network object detection meta-architectures, single-shot detector (SSD) and Faster R-CNN in combination with MobileNet-V2, ResNet50, ResNet101, ResNet152, and Inception ResNet-V2 feature extractors were studied and evaluated. Through the use of transfer learning, all the models were initialized using weights pre-trained on the MS COCO (Microsoft Common Objects in Context) dataset provided by the TensorFlow 2 object detection API. The Faster R-CNN model coupled with ResNet152 outperformed all other models with a mean average precision of 92.3%. However, the SSD model with the MobileNet-V2 feature extraction network achieved the lowest inference time (110ms) and the smallest memory capacity (30.5MB) compared to its counterparts. The outstanding results achieved in this study confirm that deep learning-based algorithms are capable of detecting birds of different sizes in different environments and the best model could potentially help ecologists in monitoring and identifying birds from other species present in the environment.
  • Thumbnail Image
    Item
    Open Access
    Natural Language Financial Forecasting: The South African Context
    (2021) Katende, Simon; Er, Sebnem; Nyirenda, Juwa; Rajaratnam, Kanshukan
    The stock market plays a fundamental role in any country's economy as it efficiently directs the flow of savings and investments of an economy in ways that advances the accumulation of capital and the production of goods and services. Factors that affect the price movement of stocks include company news and performance, macroeconomic factors, market sentiment as well as unforeseeable events. The conventional prediction approach is based on historical numerical data such as price trends and trading volumes to name a few. This thesis reviews the literature of Natural Language Financial Forecasting (NLFF) and proposes novel implementation techniques with the use of Stock Exchange News Service (SENS) announcements to predict stock price trends with machine learning methods. Deep Learning has recently sparked interest in the data science communities, but the literature on the application of deep learning in stock prediction, especially in emerging markets like South Africa, is still limited. In this thesis, the process of labelling announcements, the use of a more statistically relevent technique called the event study was used. Classical textual preprocessing and representation techniques were replaced with state-of-the-art sentence embeddings. Deep learning models (Deep Neural Network (DNN)) were then compared to Classical Models (Logistic Regression (LR)). These models were trained, optimized and deployed using the Tensorflow Machine Learning (ML) framework on Google Cloud AI Platform. The comparison between the performance results of the models shows that both DNN and LR have potential operational capabilites to use information dissemination as a means to assist market participants with their trading decisions.
  • Thumbnail Image
    Item
    Open Access
    Predicting social unrest events in South Africa using LSTM neural networks
    (2021) Zambezi, Samantha; Nyirenda, Juwa
    This thesis demonstrates an approach to predict the count of social unrest events in South Africa. A comparison is made between traditional forecasting approaches and neural networks; the traditional forecast method selected being the Autoregressive Integrated Moving Average (ARIMA model). The type of neural network implemented was the Long Short-Term Memory (LSTM) neural network. The basic theoretical concepts of ARIMA and LSTM neural networks are explained and subsequently, the patterns of the social unrest time series were analysed using time series exploratory techniques. The social unrest time series contained a significant number of irregular fluctuations with a non-linear trend. The structure of the social unrest time series suggested that traditional linear approaches would fail to model the non-linear behaviour of the time series. This thesis confirms this finding. Twelve experiments were conducted, and in these experiments, features, scaling procedures and model configurations are varied (i.e. univariate and multivariate models). Multivariate LSTM achieved the lowest forecast errors and performance improved as more explanatory features were introduced. The ARIMA model's performance deteriorated with added complexity and the univariate ARIMA produced lower forecast errors compared to the multivariate ARIMA. In conclusion, it can be claimed that multivariate LSTM neural networks are useful for predicting social unrest events.
  • No Thumbnail Available
    Item
    Open Access
    Real-time video sentiment analysis through the use of snapshots
    (2025) Ramma, Sudiptee; Nyirenda, Juwa
    There are many types of emotions that one can experience and they usually have a direct impact on a person's behaviour. Emotions can be conveyed in several ways such as gestures/body movement, words or facial expressions and this dissertation we aim to distinguish the emotional state of a person based on their facial expressions. Several approaches have been devised in this regard by various past researchers within the computer vision field but unfortunately, despite the similarities in the adopted techniques for the facial and emotion detection processes, there still exist some discrepancies regarding their performances when applied to different images or video streams. As such, the goal of this dissertation is to develop a program that can analyse a real-time video stream and take in each of the frames as an image snapshot which can be in turn processed to efficiently identify faces and recognise a person's emotion based on their facial expressions. Two scenarios, namely Frontal only, and Profile and Frontal, each with their datasets were accounted for in this research. The first dataset (Frontal) consists only of users who are facing forward and the second one (Profile and Frontal) consists of users who are facing forward as well as sideways. Convolutional Neural Network (CNN) models were constructed for each of the given datasets on both the augmented and non-augmented versions of these datasets to obtain the best possible model for each scenario before applying such model to a real-time video stream. In both scenarios, the augmented models outperformed the non-augmented models when tested on unseen static image data and when such a model was applied to a real-time video stream with the help of the OpenCV library and the relevant Haar Cascade classifiers, required for the face detection process (depending on which scenario), fairly accurate results were obtained when each frame within the video stream were converted into an image snapshot before classification. The code for this dissertation can be found here: https://github.com/Drish19/Facial-Emotion-Recognition.
  • Thumbnail Image
    Item
    Open Access
    Toward a sustainable energy future: Peak load shaving in commercial properties to reduce cost of energy
    (2022) Woodley, Tiffany Deanne; Nyirenda, Juwa
    Transitioning from fossil fuel-based energy systems to renewable sources is a is a global environmental imperative. South Africa has a coal-based energy sector, and consumers could be incentivised to pursue renewable energy alternatives if these solutions were financially advantageous. In South Africa, commercial properties are billed per kWh and can incur an additional demand charge that often accounts for a substantial portion of the energy bill, depending on the load factor. This thesis investigates peak load shaving as a solution for commercial properties to reduce their cost of electricity while supporting the transition to a greener energy future. Of the methods proposed for peak load shaving, reinforcement learning holds the greatest promise. However, its application in practice has been limited due to the “curse of dimensionality”. To make reinforcement learning a feasible option for peak load shaving, this thesis introduces a novel approach that employs clustering the energy demand profile shapes and training separate learning agents to target specific demand shapes, thereby reducing the complexity of the problem presented to the individual agents. The reinforcement learning model was trained on historical data from a commercial shopping centre in Cape Town using a hypothetical battery. Two scenarios were considered; the first assumed the absence of solar in the energy system while the second assumed its presence. Once trained, the learning agents were tested on unfamiliar energy data from the same shopping centre, and they achieved practical peak load shaving results. In Scenario 1 when using only a battery, monthly demand was reduced by 91 kW on average. Introducing a solar system in Scenario 2 increases uncertainty in the problem. The results, only demonstrated on one cluster, show the battery most often achieved a 50 kW reduction per day. In both scenarios, a learning agent trained on particular clusters of demand profiles was able to reduce peak energy demand for all unfamiliar days. Furthermore, in Scenario 2, the agent's learning progression indicated that the agent was learning to increase the battery output during the predominant peak. This suggests that our method's efficacy would improve with increased training time. If implemented, this approach could provide a practical peak shaving solution for the commercial shopping centre in Cape Town, effectively lowering their energy demand charges. This thesis has shown that clustering techniques used in conjunction with reinforcement learning is a promising approach when considering the peak shaving problem.