京都大学 / 0048 / 新制・課程博士 / 博士(工学) / 甲第22069号 / 工博第4650号 / 新制||工||1725(附属図書館) / 京都大学大学院工学研究科電気工学専攻 / (主査)教授 引原 隆士, 教授 松尾 哲司, 准教授 三谷 友彦 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

abstract: In many biological research studies, including speech analysis, clinical research, and prediction studies, the validity of the study is dependent on the effectiveness of the training data set to represent the target population. For example, in speech analysis, if one is performing emotion classification based on speech, the performance of the classifier is mainly dependent on the number and quality of the training data set. For small sample sizes and unbalanced data, classifiers developed in this context may be focusing on the differences in the training data set rather than emotion (e.g., focusing on gender, age, and dialect). This thesis evaluates several sampling methods and a non-parametric approach to sample sizes required to minimize the effect of these nuisance variables on classification performance. This work specifically focused on speech analysis applications, and hence the work was done with speech features like Mel-Frequency Cepstral Coefficients (MFCC) and Filter Bank Cepstral Coefficients (FBCC). The non-parametric divergence (D_p divergence) measure was used to study the difference between different sampling schemes (Stratified and Multistage sampling) and the changes due to the sentence types in the sampling set for the process. / Dissertation/Thesis / Masters Thesis Electrical Engineering 2020

Semi-supervised learning (SSL) is a branch of machine learning focusing on using not only labeled data samples, but also unlabeled ones, in an effort to decrease the need for labeled data and thus allow using machine learning even when labeling large amounts of data would be too costly. Despite its quick development in the recent years, there are still issues left to be solved before it can be broadly deployed in practice. One of those issues is class distribution mismatch. It arises when the unlabeled data contains samples not belonging to the classes present in the labeled data. This confuses the training and can even lead to getting a classifier performing worse than a classifier trained on the available data in purely supervised fashion. We designed a filtration method called Unfavorable Data Filtering (UDF) which extracts important features from the data and then uses a similarity-based filter to filter the irrelevant data out according to those features. The filtering happens before any of the SSL training takes places, making UDF usable with any SSL algorithm. To judge its effectiveness, we performed many experiments, mainly on the CIFAR-10 dataset. We found out that UDF is capable of significantly improving the resulting accuracy when compared to not filtering the data, identified basic guidelines...

In this research, the problem of real-time congestion management in a modern distribution system with massive active elements such as electric vehicles (EVs), distributed energy resources (DERs), and demand response (DR) is investigated. A novel hierarchical operation and management framework is proposed that can take advantage of the demand side contribution to manage the real-time congestion. There are five main steps in this framework as 1) the aggregators send their demand to the microgrid operators (MGOs), 2) the MGOs send their demand to the distribution system operator (DSO), 3) the DSO detects the congestions and calls the engaged MGOs to reduce their demand, 4) the MGOs update the electricity price to motivate the aggregators to reduce the overall demand, and 5) the DSO dispatches the system according to the finalized demand. The proposed framework is validated on two modified IEEE unbalanced test systems. The results illustrate two congestion cases at t=8:45 am and t=9:30 am in the modified IEEE 13-bus test system, which needs 363kW and 286 kW load reductions, respectively, to be fully addressed. MG#1 and MG#2 are engaged to maintain the 363 kW reduction at t=8:45, and MG#3 and MG#4 are called to reduce their demands by 386 kW at t=9:30 am. The overall interactions can relieve the congested branches. The DSO’s calculations show three congestions at t=1 pm, t=3 pm, and t=9 pm on the IEEE 123-bus test system. These congestion cases can be alleviated by reducing 809 kW, 1177 kW, and 497 kW from the corresponding MGs at t=1 pm, t=3 pm, and t=9 pm, respectively. The second part of the simulation results demonstrates that the proposed real-time data estimator (RDE) can reduce the DSO’s miss-detected congestion cases due to the uncertain data. There are two miss-detected congestions in the IEEE 13-bus test system at t=1:15 pm and t=1:30 pm that can be filtered for t=1:15 pm and minored for t=1:30 pm using the RDE. The proposed RDE can also reduce the miss-detected congestions from 18 cases to four cases in the IEEE 123-bus test system. As a result, the RDE can minimize the extra costs due to the uncertain data. The overall results validate that the proposed framework can adaptively manage real-time congestions in distribution systems.

Churn prediction is a critical task for businesses to retain their valuable customers. This paper presents a comprehensive study of churn prediction in the telecom sector using 15 approaches, including popular algorithms such as Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, and AdaBoost. The study is segmented into three sets of experiments, each focusing on a different approach to building the churn prediction model. The model is constructed using the original training set in the first set of experiments. The second set involves oversampling the training set to address the issue of imbalanced data. Lastly, the third set combines oversampling with recursive feature selection to enhance the model's performance further. The results demonstrate that the Adaptive Boost classifier, implemented with oversampling and recursive feature selection, outperforms the other 14 techniques. It achieves the highest rank in all three evaluation metrics: recall (0.841), f1-score (0.655), and roc_auc (0.793), further indicating that the proposed approach effectively predicts churn and provides valuable insights into customer behavior.

No description available.

Churn prediction is a common data mining problem that many companies face across industries. More commonly, customer churn has been studied extensively within the telecommunications industry where there is low customer retention due to high market competition. Similar to customer churn, employee churn is very costly to a company and by not deploying proper risk mitigation strategies, profits cannot be maximized, and valuable employees may leave the company. The cost to replace an employee is exponentially higher than finding a replacement, so it is in any company’s best interest to prioritize employee retention. This research combines machine learning techniques with undersampling in hopes of identifying employees at risk of churn so retention strategies can be implemented before it is too late. Four different classification algorithms are tested on a variety of undersampled datasets in order to find the most effective undersampling and classification method for predicting employee churn. Statistical analysis is conducted on the appropriate evaluation metrics to find the most significant methods. The results of this study can be used by the company to target individuals at risk of churn so that risk mitigation strategies can be effective in retaining the valuable employees. Methods and results can be tested and applied across different industries and companies.

This thesis primarily focuses on studying the impact of Distributed Generation (DG) on the electromechanical transients in the electric grid (distribution, transmission or combined transmission and distribution (TandD) systems) using a Three Phase Dynamics Analyzer (hereafter referred to as TPDA). TPDA includes dynamic models for electric machines, their controllers, and a three-phase model of the electric grid, and performs three-phase dynamic simulations without assuming a positive sequence network model. As a result, TPDA can be used for more accurate investigation of electromechanical transients in the electric grid in the presence of imbalances. At present, the Electromagnetic Transient Program (EMTP) software can be used to perform three-phase dynamic simulations. This software models the differential equations of the entire electric network along with those of the machines. This calls for solving differential equations with time constants in the order of milliseconds (representing the fast electric network) in tandem with differential equations with time constants in the order of seconds (representing the slower electromechanical machines). This results in a stiff set of differential equations, making such an analysis extremely time consuming. For the purpose of electromechanical transient analysis, TPDA exploits the difference in the order of time constants and adopts phasor analysis of the electric network, solving differential equations only for the equipment whose dynamics are much slower than those of the electric network. Power Flow equations are solved using a graph trace analysis based approach which, along with the explicit partitioned method adopted in TPDA, can eventually lead to the use of distributed computing that will further enhance the speed of TPDA and perhaps enable it to perform dynamic simulation in real time . In the work presented here, first an overview of the methodology behind TPDA is provided. A description of the object oriented implementation of TPDA in C++/C# is included. Subsequently, TPDA is shown to accurately simulate power system dynamics of balanced networks by comparing its results against those obtained using GE-PSLF®. This is followed by an analysis that demonstrates the advantages of using TPDA by highlighting the differences in results when the same problem is analyzed using a three-phase network model with unbalances and the positive sequence network model as used in GE-PSLF®. Finally, the impact of rapidly varying DG generation is analyzed, and it is shown that as the penetration level of DG increases, the current and voltage oscillations throughout the transmission network increase as well. Further, rotor speed deviations are shown to grow proportionally with increasing DG penetration. / Master of Science

This thesis involves the development and study of a finite element model of a hermetic, single-vane compressor and a single-phase alternating current induction motor assembled in a common housing. The manufacturer of this unit is experiencing a high scrap rate due to interference during operation between the stator and rotor of the motor. The rotor shaft of the motor is non-typical because of its cantilever design. The finite element model was first subjected to eigenvalue analysis. This revealed that the interference producing displacements were not the result of torque application to the rotor at a frequency close to an eigenvalue of the mechanical system. After a review of the literature and discussions with Electrical Engineering Department faculty possessing extensive motor experience, it was surmised that the physical phenomenon causing the rotor displacement was unbalanced magnetic pull. This phenomenon occurs in the air gap of rotating electric machines due to eccentricity in the air gap. The model was then subjected to simultaneous harmonic force inputs with magnitudes of unity on the rotor and stator surfaces to simulate the presence of unbalanced magnetic pull. It was found that the rotor shaft acts as a cantilever beam while the stator and housing are essentially rigid. The displacements due to these forces were examined and then scaled to develop the motor parameters necessary to produce the radial forces required for stator/rotor interference. Several recommendations were then made regarding possible solutions to the interference problem. / Master of Science

Distributed Series Reactors (DSRs) can be used to control power flow to more fully utilize the capacity of a transmission network, delaying investment in new transmission lines. In this study the IEEE 39 bus standard test system is modified to a 3-phase, unbalanced model consisting of 230 kV, 345 kV and 500 kV lines, where lines of different voltage run in parallel. This model is used to study load growth and the effect of adding DSRs to alleviate resulting overloads, and in particular to alleviate overloads on lines of different voltage running in parallel. The economic benefit of adding DSRs to the network is compared to the addition of new transmission lines in the network. In the second part of the work, the effect of unsymmetrical operation of DSRs on a single transmission line is studied and compared to the symmetrical operation of DSRs. It is found that the unsymmetrical operation of DSRs is more economical. Finally the unsymmetrical operation of DSRs to reduce voltage imbalance in the network is considered. / Master of Science