• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 1
  • 1
  • Tagged with
  • 2
  • 2
  • 2
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

Elektrisk lastprognostisering för byggnader / Electrical load prediction for buildings

Bojestig, John January 2019 (has links)
Om världen ska kunna ställa om till förnyelsebara energikällor krävs det nya och bättre tekniklösningar. En liten del av lösningen på balanseringsproblematiken på elnätet som icke-reglerbara energikällor som sol- och vindkraft står för kan vara att sköta en del av balanseringen lokalt i byggnader med hjälp av batterilager. För att kunna styra den balanseringen på ett optimalt sätt behöver styrningen ha prognoser för hur stor den elektriska lasten i byggnaden kommer vara framöver. Syftet med denna studie har varit att utföra en elektrisk lastprognostisering för en byggnad över ett dygn. Modellen som utförde elektrisk lastprognostisering för en byggnad har baserats på neurala nätverk. Istället för att ha ett neuralt nätverk som prognostiserar över hela dygnet har 24 olika neurala nätverk prognostiserat varsin timma. Varje neuralt nätverk har valts efter tester mellan ett flertal neurala nätverk med variationer i parametrar som har tagits fram med hjälp av en klusteralgoritm. Resultatet visade att modellen som tagits fram i studien prognostiserade den elektriska lasten i en byggnad över ett dygn med en felmarginal enligt mean average percentage error på 5.67%. Det gick även att se fördelar med att dela upp prognostiseringen i mindre delar och testa olika parametrar för varje timma som skulle prognostiseras. Med avseende på jämförelser med andra studier och att bostadshus är ett välkänt svårt prognostiseringsproblem bör resultatet anses som godkänt. Det mesta tyder på att prognostiseringsmodellen är tillräckligt bra för att kunna assistera en smart styrning av ett batteri i en byggnad med användbar information / If the world should be able to convert to renewable energy sources, new and better technical solutions is required. A small part of the solution to the balancing problem on the electricity grid, as non-controllable energy sources such as solar and wind power is highly responsible for, can be to handle part of the balancing locally in buildings using battery storage. In order to be able to control this balancing in the optimal way, the control system needs to have forecasts of how large the electric load in the building will be in the future. The aim of this study has been to carry out electrical load prediction for a building over one day. The model that carried out electrical load forecasting for a building has been based on neural networks. Instead of having one neural network that predicts the whole day, 24 different neural networks have been forecasting each hour. Each neural network has been selected after testing between several neural networks with variations in parameters that have been selected using a cluster algorithm. The result showed that the model developed in the study predicted the electric load in a building over one day with a mean average percentage error of 5.67%. It was also possible to see the advantages of dividing the prediction into smaller parts and testing different parameters for each hour that would be forecast. With regard to comparisons with other studies and that residential buildings are a well-known difficult forecasting problem, the result should be considered as acceptable. Most indications show that the forecasting model is good enough to be able to assist a smart control of a battery in a building with useful information.
2

Lost in Transcription : Evaluating Clustering and Few-Shot learningfor transcription of historical ciphers

Magnifico, Giacomo January 2021 (has links)
Where there has been a steady development of Optical Character Recognition (OCR) techniques for printed documents, the instruments that provide good quality for hand-written manuscripts by Hand-written Text Recognition  methods (HTR) and transcriptions are still some steps behind. With the main focus on historical ciphers (i.e. encrypted documents from the past with various types of symbol sets), this thesis examines the performance of two machine learning architectures developed within the DECRYPT project framework, a clustering based unsupervised algorithm and a semi-supervised few-shot deep-learning model. Both models are tested on seen and unseen scribes to evaluate the difference in performance and the shortcomings of the two architectures, with the secondary goal of determining the influences of the datasets on the performance. An in-depth analysis of the transcription results is performed with particular focus on the Alchemic and Zodiac symbol sets, with analysis of the model performance relative to character shape and size. The results show the promising performance of Few-Shot architectures when compared to Clustering algorithm, with a respective SER average of 0.336 (0.15 and 0.104 on seen data / 0.754 on unseen data) and 0.596 (0.638 and 0.350 on seen data / 0.8 on unseen data).

Page generated in 0.0831 seconds