In recent years, machine learning applications have gained great attention in the wind power industry. Among these, artificial neural networks have been utilized to predict the fatigue loads of wind turbine components such as rotor blades. However, the limited number of contributions and differences in the used databases give rise to several questions which this study has aimed to answer. Therefore, in this study, 5-min SCADA data from the Lillgrund wind farm has been used to train two feed-forward neural networks to predict the fatigue loads at the blade root in flapwise and edgewise directions in the shape of damage equivalent loads.The contribution of different features to the model’s performance is evaluated. In the absence of met mast measurements, mesoscale NEWA data are utilized to present the free flow condition. Also, the effect of wake condition on the model’s accuracy is examined. Besides, the generalization ability of the model trained on data points from one or multiple turbines on other turbines within the farm is investigated. The results show that the best accuracy was achieved for a model with 34 features, 5 hidden layers with 100 neurons in each hidden layer for the flapwise direction. For the edgewise direction, the best model has 54 features, 6 hidden layers, and 125 neurons in each hidden layer.For a model trained and tested on the same turbine, mean absolute percentage errors (MAPE) of 0.78% and 9.31% are achieved for the flapwise and edgewise directions, respectively. The seen difference is argued to be a result of not having enough data points throughout the range of edgewise moments. The use of NEWA data has been shown to improve the model’s accuracy by 10% for MAPE values, relatively. Training the model under different wake conditions did not improve the model showing that the wake effects are captured through the input features to some extent. Generalization of the model trained on data points from one turbine resulted in poor results in the flapwise direction. It was shown that using data points from multiple turbines can improve the model’s accuracy to predict loading on other turbines.

Sustainable management of finite resources is vital for ensuring livable conditions for both current and future generations. Measuring the total water consumption of residential households at high temporal resolutions and automatically disaggregating the sole signal into classified end usages (e.g. shower, sink) allows for identification of behavioural patterns that could be improved to minimise wasteful water consumption. Such disaggregation is not trivial, as water consuming patterns vary greatly depending on consumer behaviour, and further since at any given time, an unknown amount of fixtures may be used simultaneously. In this work, we approach the disaggregation problem by evaluating the performance of a set of recurrent and convolutional neural network structures provided approximately one year of high resolution water consumption data from a single apartment in Sweden. Unlike previous approaches to the problem, we let the models process the full, uninterrupted flow traces (as opposed to extracted segments of water consuming activity) in order to allow for temporal dependencies within and between water consuming activities to be learned. Out of four networks applied to the task, we find that a deeper temporal convolutional network structure yields the best overall results on the test data, with prediction accuracy of 85% and F1-score above 0.8 averaged over all end-use categories - a performance exceeding that of commercial analysis tools, and comparable to components of current state-of-the-art approaches. However, significant decreases in performance are observed for all of the networks, particularly for toilet and washing machine activity, when evaluating the models on unseen and augmented data from the apartment, indicating the results can not be fully generalised for usage in other households. / Hållbar användning av ändliga resurser är avgörande för att försäkra god livskvalitet för både nutida och framtida generationer. I Sverige är vatten för många en självklarhet, vilket öppnar upp för slösaktigt användande. En metod för att utbilda användare och identifiera icke hållbara beteenden är att kvantifiera vattenförbrukningen i hushåll baserat på syfte (t.ex. tvätta händerna, diska) eller källa (t.ex. dusch, handfat) av slutanvändningen. För att göra en sådan sammanställning mäts den totala åtkomsten av vatten i hög upplösning från hushåll, och signalen delas sedan upp i respektive kategori av slutanvändning. En sådan disaggregering är inte trivial, och försvåras av skillnader i beteendemönster hos användare samt faktumet att vi inte vid någon tidpunkt vet hur många vattenarmaturer som används samtidigt. I syftet att förbättra nuvarande tekniker för disaggregeringsproblemet implementerar och utvärderar vi alternativa lösningar baserade på rekurrenta och konvolutionerande neurala nätverk, på flödesdata insamlad med hög upplösning från en lägenhet i Sverige under en period av cirka ett år. Till skillnad från tidigare förhållningssätt till problemet låter vi våra modeller bearbeta den fullständiga, oavbrutna, flödesdatan (i motsats till extraherade segment av vattenförbrukande aktiviteter) för att möjliggöra lärandet av tidsmässiga beroenden inom och mellan vattenförbrukande aktiviteter. Utav fyra testade nätverk finner vi att ett djupt konvolutionerande nätverk ger den bästa klassificeringen överlag, givet testdata, med genomsnittlig igenkänningsnogrannhet på 85%. Signifikant försämrade resultat observerades för samtliga modeller i kategorierna toalett och tvättmaskin när nätverken testades på augmenterad data från hushållet, vilket indikerar att resultaten inte kan generaliseras för användning i andra lägenheter.

Developing and improving upon a good empirical model for an engine can be time-consuming and costly. The goal of this thesis has been to evaluate data-driven modelling, specifically neural networks, to see how well it can handle training for some static models like the mass flow of air into the cylinder, mean effective pressure and pump mean effective pressure but also for transient modelling, specifically the exhaust gas temperature. These models are evaluated against the classical empirical models to see if neural networks are a viable modelling option. This is done with five different types of neural networks which are trained. These are the feed-forward neural network, Nonlinear autoregressive exogenous model network, layer recurrent network, long short term memory network and gated recurrent network.The inputs were determined by looking at more simple physical models but also looking at the covariance to determine the usefulness of the input. If the calculation time is small for the specific network, the neural network structure is tested and optimized by training many networks and finding the median/mean result for that specific test.The result has shown that the static models are handled very well by the most simple feed-forward network. For the exhaust temperature, both NARX and Layer recurrent network could predict and handle it well giving results very close to the empirical models and could be a viable option for transient modelling, on the other hand, Long short term memory, gated recurrent network and the feed-forward network had trouble predicting the exhaust gas temperature and returned bad results while training.

Att manuellt välja en eller flera meningar ur en filmrecension att använda som citat kan vara en tidskrävande uppgift. Denna rapport utvärderar övervakade maskininlärningsmodeller för att skapa en prototyp som automatiskt kan välja lämpliga citat ur recensioner. Utifrån resultatet av en litteraturstudie valdes två modeller att implementera och utvärdera på data bestående av filmrecensioner och tillhörande manuellt valda citat. Av arbetets två implementerade modeller, BERT med Inter-Sentence Transformer och BERT med ett artificiellt neuronnät, visade den sistnämnda marginellt bättre resultat. Modellerna utvärderades med ROUGE och jämfördes med tidigare studiers toppresultat inom automatisk textsummering. Slutsatsen är att de modeller som utvärderades inte presterar tillräckligt väl inom problemområdet för att motivera en driftsättning utan ytterligare utvecklingsarbete. Dock visar resultaten att det finns potential i att de utvärderade tillvägagångssätten delvis kan ersätta manuella val av citat i framtiden. / To choose a number of sentences from a movie review to use as a quote can be time consuming if done manually. This thesis evaluates supervised machine learning models to create a prototype that automatically can choose such quotes. The thesis chose, based on a literature study, two models to implement and evaluate on data consisting of movie reviews and their respective corresponding manually chosen quotes. Out of the thesis two implemented models, BERT with Inter-Sentence Transformer and BERT with an artificial neural network, the latter showed marginally better results. The models were evaluated with ROUGE and was compared with state-of-the-art models regarding automatic text summarization. The conclusion is that the models that were evaluated do not perform well enough for the problem to motivate full deployment without further development efforts. However, the results show that there is potential that the evaluated methods can partially replace manual labour when choosing quotes.

Permanent magnet synchronous machines have various advantages and have showed the most superiorperformance for Electric Vehicles. However, modeling them is difficult because of their nonlinearity. In orderto deal with the complexity, the artificial neural network and machine learning models including k-nearest neighbors, decision tree, random forest, and multiple linear regression with a quadratic model are developed to predict electrical parameters and losses as new prediction approaches for the performance of Volvo Cars’ electric vehicles and evaluate their performance. The test operation data of the Volvo Cars Corporation was used to extract and calculate the input and output data for each prediction model. In order to smooth the effects of each input variable, the input data was normalized. In addition, correlation matrices of normalized inputs were produced, which showed a high correlation between rotor temperature and winding resistance in the electrical parameter prediction dataset. They also demonstrated a strong correlation between the winding temperature and the rotor temperature in the loss prediction dataset.Grid search with 5-fold cross validation was implemented to optimize hyperparameters of artificial neuralnetwork and machine learning models. The artificial neural network models performed the best in MSE and R-squared scores for all the electrical parameters and loss prediction. The results indicate that artificial neural networks are more successful at handling complicated nonlinear relationships like those seen in electrical systems compared with other machine learning algorithms. Compared to other machine learning algorithms like decision trees, k-nearest neighbors, and multiple linear regression with a quadratic model, random forest produced superior results. With the exception of q-axis voltage, the decision tree model outperformed the knearestneighbors model in terms of parameter prediction, as measured by MSE and R-squared score. Multiple linear regression with a quadratic model produced the worst results for the electric parameters prediction because the relationship between the input and output was too complex for a multiple quadratic equation to deal with. Random forest models performed better than decision tree models because random forest ensemblehundreds of subset of decision trees and averaging the results. The k-nearest neighbors performed worse for almost all electrical parameters anticipation than the decision tree because it simply chooses the closest points and uses the average as the projected outputs so it was challenging to forecast complex nonlinear relationships. However, it is helpful for handling simple relationships and for understanding relationships in data. In terms of loss prediction, the k-nearest neighbors and decision tree produced similar results in MSE and R-squared score for the electric machine loss and the inverter loss. Their prediction results were worse than the multiple linear regression with a quadratic model, but they performed better than the multiple linear regression with a quadratic model, for forecasting the power difference between electromagnetic power and mechanical power.

蓄電池は携帯機器，電気自動車をはじめ，自然エネルギー有効利用に至るまで広範囲に利用され，その重要性はますます高まっている。これら機器の使用時間や特性は蓄電池の特性に大きく依存することから，電池自体の特性改善に加え，劣化を診断してより効率的に電池を運用することが求められている。本論文は，非線形情報処理を得意とする人工知能を用いた2次電池の劣化診断法を開発し，エネルギーの有効利用に資する技術を確立した。機器動作時の電池電圧・電流波形と電池劣化特性との関連性を，人工知能を用い学習することにより，機器稼働時に電池の劣化を診断することができる。なお，この関連性は非線形で複雑であるが，非線形分析を得意とする人工知能は劣化診断に適している。学習には時間を要するものの，診断は短時間になし得ることから，提案法は稼働時劣化診断に適している。本論文では，この特徴を生かし，電池の等価回路（ECM）を導出し，充電率（SOC），容量維持率（SOH）を推定している。また，本論文では現在産業応用分野で用いられている，リチウムイオン電池，ニッケル水素電池，鉛蓄電池を対象とし，提案法はあらゆる電池使用機器に応用可能である。また，提案法を電池状態監視装置（BMU）や，マイコンなどを用いた組み込みシステムに応用可能とし，実証している。以上のことから，本論文は，新たな蓄電池の劣化診断法の確立し，その有効性を確認している。 / The importance of rechargeable batteries nowadays is increasing from the portable electronic devices and solar energy industry up to the development of new EV models. The rechargeable batteries have a crucial role in the storage system, mostly in mobile applications and transportation, because the period of its usage and the flexibility of the function are determined by the battery. Due to the black box approach of the ANN it is possible to connect the complex physical phenomenon with a specific physical meaning expressed with a nonlinear logic between inputs and output. Using specific input data to relate with the desired output, makes possible to create a pattern connection with input and output. This ability helps to estimate in real time the desired outputs, behaviors, phenomes and at the same time it can be used as a real time diagnosis method. / 博士(工学) / Doctor of Philosophy in Engineering / 同志社大学 / Doshisha University

Självkörande fordon kan minska trafikstockningar och minska antalet trafikrelaterade olyckor. Då det i framtiden kommer att finnas miljontals autonoma fordon krävs en bättre förståelse av omgivningen. Syftet med detta projekt är att skapa ett externt automatiskt trafikledningssystem som kan upptäcka och spåra 3D-objekt i en komplex trafiksituation för att senare skicka beteendet från dessa objekt till ett större projekt som hanterar med att 3D-modellera trafiksituationen. Projektet använder sig av Tensorflow ramverket och YOLOv3 algoritmen. Projektet använder sig även av en kamera för att spela in trafiksituationer och en dator med Linux som operativsystem. Med hjälp av metoder som vanligen används för att skapa ett automatiserat trafikledningssystem utvärderades ett målföljningssystem. De slutliga resultaten visar att systemet är relativt instabilt och ibland inte kan känna igen vissa objekt. Om fler bilder används för träningsprocessen kan ett robustare och mycket mer tillförlitligt system utvecklas med liknande metodik. / Autonomous vehicles can decrease traffic congestion and reduce the amount of traffic related accidents. As there will be millions of autonomous vehicles in the future, a better understanding of the environment will be required. This project aims to create an external automated traffic system that can detect and track 3D objects within a complex traffic situation to later send these objects’ behavior for a larger-scale project that manages to 3D model the traffic situation. The project utilizes Tensorflow framework and YOLOv3 algorithm. The project also utilizes a camera to record traffic situations and a Linux operated computer. Using methods commonly used to create an automated traffic management system was evaluated. The final results show that the system is relatively unstable and can sometimes fail to recognize certain objects. If more images are used for the training process, a more robust and much more reliable system could be developed using a similar methodology.

In this work we use a graph-theoretic representation of secondary RNA structure found in the database RAG: RNA-As-Graphs. We model the bonding of two RNA secondary structures to form a larger structure with a graph operation called merge. The resulting data from each tree merge operation is summarized and represented by a vector. We use these vectors as input values for a neural network and train the network to recognize a tree as RNA-like or not based on the merge data vector. The network correctly assigned a high probability of RNA-likeness to trees identified as RNA-like in the RAG database, and a low probability of RNA-likeness to those classified as not RNA-like in the RAG database. We then used the neural network to predict the RNA-likeness of all the trees of order 9. The use of a graph operation to theoretically describe the bonding of secondary RNA is novel.

Robotic and control industry implements different control technique to control the movement and the position of a robotic arm. PID controllers are the most used controllers in the robotics and control industry because of its simplicity and easy implementation. However, PIDs’ performance suffers under noisy environments. In this research, a controller based on Artificial Neural Networks (ANN) called the model reference controller is examined to replace traditional PID controllers to control the position of a robotic arm in a noisy environment. Simulations and implementations of both controllers were carried out in MATLAB. The training of the ANN was also done in MATLAB using the Supervised Learning (SL) model and Levenberg-Marquardt backpropagation algorithm. Results shows that the ANN implementation performs better than traditional PID controllers in noisy environments. / Robotoch kontrollindustrin implementerar olika kontrolltekniker för att styra rörelsen och placeringen av en robotarm. PID-styrenheter är de mest använda kontrollerna inom roboten och kontrollindustrin på grund av dess enkelhet och lätt implementering. PID:s prestanda lider emellertid i bullriga miljöer. I denna undersökning undersöks en styrenhet baserad på Artificiell Neuralt Nätverk (ANN) som kallas modellreferenskontrollen för att ersätta traditionella PID-kontroller för att styra en robotarm i bullriga miljöer. Simuleringar och implementeringar av båda kontrollerna utfördes i MATLAB. Utbildningen av ANN:et gjordes också i MATLAB med hjälp av Supervised Learning (SL) -modellen och LevenbergMarquardt backpropagationsalgoritmen. Resultat visar att ANN-implementeringen fungerar bättre än traditionella PID-kontroller i bullriga miljöer.

The data traffic has been growing drastic over the past few years due to digitization and new technologies that are introduced to the market, such as autonomous cars. In order to meet this demand, the MIMO-OFDM system is used in the fifth generation wireless network, 5G. Designing the optimal wireless network is currently the main research within the area of telecommunication. In order to achieve such a system, multiple factors has to be taken into account, such as the suppression of interference from other users. A traditional method called linear minimum mean square error filter is currently used to suppress the interferences. To derive such a filter, a selection of parameters has to be estimated. One of these parameters is the ideal interference plus noise covariance matrix. By gathering prior knowledge of the underlying structure of the interference channel matrices in terms of the number of interferers and their corresponding bandwidths, the estimation of the ideal covariance matrix could be facilitated. As for this thesis, machine learning algorithms were used to extract these prior knowledge. More specifically, a two or three hidden layer feedforward neural network and a support vector machine with a linear kernel was used. The empirical findings implies promising results with accuracies above 95% for each model. / Under de senaste åren har dataanvändningen ökat drastiskt på grund av digitaliseringen och allteftersom nya teknologier introduceras på marknaden, exempelvis självkörande bilar. För att bemöta denna efterfrågan används ett s.k. MIMO-OFDM system i den femte generationens trådlösa nätverk, 5G. Att designa det optimala trådlösa nätverket är för närvarande huvudforskningen inom telekommunikation och för att uppnå ett sådant system måste flera faktorer beaktas, bland annat störningar från andra användare. En traditionell metod som används för att dämpa störningarna kallas för linjära minsta medelkvadratfelsfilter. För att hitta ett sådant filter måste flera olika parametrar estimeras, en av dessa är den ideala störning samt bruskovariansmatrisen. Genom att ta reda på den underliggande strukturen i störningsmatriserna i termer av antal störningar samt deras motsvarande bandbredd, är något som underlättar uppskattningen av den ideala kovariansmatrisen. I följande avhandling har olika maskininlärningsalgoritmer applicerats för att extrahera dessa informationer. Mer specifikt, ett neuralt nätverk med två eller tre gömda lager samt stödvektormaskin med en linjär kärna har använts. De slutliga resultaten är lovande med en noggrannhet på minst 95% för respektive modell.