A Review of Optimal Energy Management Strategies Using Machine Learning Techniques for Hybrid Electric Vehicles

송창희,김기영,성동환,김경현,양현준,이희윤,조구영,차석원 한국자동차공학회 2021 International journal of automotive technology Vol.22 No.5

A hybrid electric vehicle (HEV) is defined as a vehicle that has two or more power sources, the hybrid electric vehicle is a representative eco-friendly vehicle because it can operate efficiently with each power source and requires only a small sized electric power source. However, it is not possible to develop high efficiency HEVs without an effective energy management system (EMS), a well-designed EMS is vital in HEVs because they need to manage two power sources. Motivated by this, there are continuing efforts being made to research and establish suitable energy management strategies in order to develop high efficiency HEVs. In the past, many energy management strategies for HEVs were developed based on optimal control theory. Recently, various kinds of machine learning technologies have been applied to HEV EMS development based on breakthroughs in the fields of machine learning and artificial intelligence (AI). Machine learning is a field of research that allows computers to perform arbitrary tasks guided by data rather than explicit programming. Machine learning can be classified into supervised learning, reinforcement learning (semi-supervised learning), and unsupervised learning depending on how the training data is structured. In this study, we look at cases and studies in which machine learning techniques from each category were used to develop HEV energy management strategies.

기계학습의 미디어 산업 적용 :콘텐츠 평가 및 제작 자원을 중심으로

권신혜,박경우,장병철,장병희 한국콘텐츠학회 2019 한국콘텐츠학회논문지 Vol.19 No.7

This study researched the effect of application systems for media industry by using machine learning method focusing on industrial organization theory. First, for applying the system successfully, formation of sympathy about needs is required. The introduction of machine learning can bring change in each stage of value chain especially, decision making process of investment and production process. In investment side, objective performance prediction data can enhance efficiency, and content diversity can decrease with concentrated investment phenomenon to secured content by the system. In production side, if the system support to make creators decrease simple repeat works, production efficiency will increase. 이 연구는 기계학습의 도입이 미디어 산업구조에 어떠한 영향을 미칠 것인가에 대해 산업조직론적 관점에서 살펴보았다. 먼저 기계학습 기법이 미디어 산업에 성공적으로 도입되기 위해서는 각 산업 단계의 조직구성원 사이에서 기계학습 기반 시스템의 필요성에 대한 공감대 형성이 선행되어야 할 것으로 분석된다. 기계학습의 도입은 기존 방송 및 영화산업의 투자 의사결정과정과 제작 과정에 유의미한 변화를 가져올 것이며, 투자 측면에서는 객관적 데이터의 제공으로 인해 효율성이 증대될 것으로 보인다. 또한, 성과가 담보된 장르 및 형식의 콘텐츠에 투자가 집중됨에 따라 다양성이 감소할 가능성이 있다. 제작 측면에서는 창작자의 반복적 행위를 기계학습 시스템이 담당하는 역할을 한다면 생산효율성이 증대될 수 있다.

Prediction of the Emotion Responses to Poster Designs based on Graphical Features: A Machine learning-driven Approach

Juhee Kim,Hyeon-Jeong Suk 한국디자인학회 2020 디자인학연구 Vol.33 No.2

Word2Vec을 활용한 한국어 도덕기반사전의 작성과 사용: ‘갑질’에 대한 적용을 중심으로

김현섭,전성재 한국윤리학회(8A3209) 2022 윤리학 Vol.11 No.2

According to the Moral Foundations Theory, there are five main reasons why an action is wrong: harm, unfairness, betrayal, subversion, and degradation. The Moral Foundations Theory is supported by self-report surveys, implicit measures, physiological evidence, and text analyses. The traditional method of text analysis is to count words: collect words for each moral foundation on the basis of linguistic intuition and count how many times those words appear in the text. Recently, a new method of text analysis has been developed: represent each word as a dense vector in a high dimensional space and use the distributed representations of words in compiling a dictionary and analyzing short texts. We used this method of distributed representation and analyzed words in terms of the Moral Foundations Theory. We produced distributed representations of words by training a Word2Vec model on a newspaper corpus of approximately 100 million Korean words. Then we used the distributed representations of Korean words to improve on an intuition-based Korean Moral Foundations Dictionary. We produced the vectors representing the moral foundations by averaging the word vectors of the improved Korean Moral Foundations Dictionary, and analyzed individual words correctly by calculating cosine similarities between the vector of words to be analyzed and the vectors that represent the moral foundations. It turns out that the Korean word ‘gapjil’ is strongly correlated with the foundations of Fairness/Cheating and Authority/Subversion, a result that agrees with our moral intuition. 도덕기반이론에 의하면 어떤 행위가 도덕적으로 그른 이유를 피해를 입히므로, 불공정하므로, 배신이므로, 전복적이므로, 타락했기 때문으로 대별할 수 있다. 도덕기반이론은 자기보고 설문, 암묵적 측정, 안면근육의 움직임이나 두뇌용적과 같은 생리학적 측정, 텍스트 분석 등의 경험적 증거에 의해 뒷받침되었다. 텍스트 분석은 주로 각 기반에 해당하는 단어들을 직관적으로 수집하여 사전을 편찬하고 그 단어들이 대상 텍스트에서 얼마나 자주 나타나는가를 세는 방법으로 수행되었다. 최근에는 각 단어를 다차원의 밀집 실수벡터에 대응시키고, 그 벡터 즉 단어의 분산표현을 활용하여 사전 편찬을 개선하고 소량의 텍스트를 분석하는 기법이 개발되었는데, 본 연구에서는 이러한 분산표현 기법을 써 텍스트를 도덕기반이론적으로 분석하였다. 한국어 신문기사 말뭉치를 Word2Vec 알고리즘으로 학습하여 사용된 단어들을 밀집 벡터에 대응시켜 분산표현을 얻었고, 이를 활용하여 직관적으로 영어 도덕기반사전을 참조하여 만든 한국어 도덕기반사전을 수정한 한국어 도덕기반사전을 작성하였다. 이렇게 작성한 사전상 단어의 벡터들을 평균하여 각 도덕기반을 표상하는 벡터를 얻었고, 그 도덕기반 벡터와의 코사인 유사도를 계산하여 도덕기반사전에 포함되지 않은 단어도 직관적 의미에 맞게 분류할 수 있음을 보였다. 그 사례로 ‘갑질’이라는 단어는 불공정/공평, 전복/권위 기반 차원에서 부정적 평가가 강하게 나타나, 우월한 지위에 있는 사람이 세력을 남용하여 공정하지 않게 이익을 추구하는 행위라는 직관적 의미에 부합하였다.

Prelude to Machine Learning-Based IRT Research: Bayesian Item Parameter Recovery

Taeyoung Kim,Seungbae Choi,Hae-Gyung Yoon 한국자료분석학회 2021 Journal of the Korean Data Analysis Society Vol.23 No.4

Using the complimentary software R and WinBugs, this study examined the item parameter recovery via Markov chain Monte Carlo (MCMC) and its convergence diagnostic measures. Ten sets of dichotomous response data were generated under the condition of 1000 examinees and 21 items with the simplest item response theory (IRT) model (that is, one parameter logistic model (1-PLM)). This study followed four steps: (1) generating 10 replication data sets using R, (2) calibrating 1-PL IRT model to those artificial data sets with WinBugs, (3) checking convergence measures via the specialized diagnostic tool, R package ‘boa’, and (4) evaluating parameter recovery performance by examining a sort of summary statistics such as RMSE. The present study shed light on the Bayesian IRT framework and the performance of statistical software in terms of item parameter recovery. Plus, by employing a series of education big data, this paper serves as a prelude to subsequent machine learning-based IRT research.

빅데이터 분석을 통한 유명인 모델의 광고효과 예측 모형 개발

김유나,한상필 한국융합학회 2020 한국융합학회논문지 Vol.11 No.8

The purpose of this study is to find out whether image similarity between celebrities and brands on social network service be a determinant to predict advertising effectiveness. To this end, an advertising effect prediction model for celebrity endorsed advertising was created and its validity was verified through a machine learning method which is a big data analysis technique. Firstly, the celebrity-brand image similarity, which was used as an independent variable, was quantified by the association network theory with social big data, and secondly a multiple regression model which used data representing advertising effects as a dependent variable was repeatedly conducted to generate an advertising effect prediction model. The accuracy of the prediction model was decided by comparing the prediction results with the survey outcomes. As for a result, it was proved that the validity of the predictive modeling of advertising effects was secured since the classification accuracy of 75%, which is a criterion for judging validity, was shown. This study suggested a new methodological alternative and direction for big data-based modeling research through celebrity-brand image similarity structure based on social network theory, and effect prediction modeling by machine learning. 본 연구는 소셜 빅데이터에 기반을 둔 유명인과 브랜드의 이미지 유사도가 광고효과를 예측할 수 있는 결정변수가 될 수 있는지를 파악하기 위해, 광고효과 예측모형을 생성하고 빅데이터 분석기법인 기계학습 방법을 통해 그 타당도를 검증하는 것을 목적으로 하였다. 이를 위해 SNS상의 키워드 네트워크 구조에 기반하여 유명인-브랜드 이미지 유사도를 정량화하고, 학습 데이터를 통해 이미지 유사도를 독립변수로, 광고효과 데이터를 종속변수로 하는 다중회귀모형을 반복 실시하여 광고효과 예측모형을 생성하였다. 이렇게 생성된 예측모형의 정확도를 판단하기 위해 예측 데이터에서 얻은 광고효과 예측값과 비교 기준으로서의 서베이값을 비교한 결과, 타당도를 판단하는 기준치인 75%의 분류 정확도를 보였으므로 본 광고효과 예측 모델링의 타당성은 확보된 것으로 입증되었다. 본 연구는 유명인-브랜드 이미지 유사성 구조를 소셜 네트워크 구조로 설명하고 그 효과를 기계학습을 통한 예측 모델링으로 검증하여 빅데이터 기반 모델링 연구에 새로운 방법론적 대안과 방향을 제시하였다.

Intelligent optimization of axial-flow pump using physics-considering machine learning

KANKANAM GAMAGE PIYUMIKA MADUSHANI,Zhou Jie,Feng Jiangang,XUHUI ZHOU,Yuan Zheng,Chen Huixiang,Chen Jinbo 한국CDE학회 2024 Journal of computational design and engineering Vol.11 No.1

To address the significant energy waste generated by axial flow pumps, this paper proposes an intelligent optimization method based on physics-considering machine learning. First, a highly parameterized geometric design theory is constructed using six featured variables to achieve a complete three-dimensional modeling of the blade geometry. Four hundred preliminary cases are studied using the computational fluid dynamics method with various combinations of these featured variables to obtain a preliminary solution. The best preliminary design has an efficiency of 83.33%, and a head of 5.495 m. To further improve this performance, this paper also presents a high-precision prediction model for the energy performance of axial flow pump based on back-propagation neural network and the encoding layers of random sampling and local feature aggregator network created. Afterwards, a multi-population genetic algorithm is used to quickly find the optimal solution within the prediction mode range. The algorithm achieved a highest efficiency of 86.373% and was validated by numerical simulation with a value of 86.057% and a prediction error of 0.316%. Compared with the preliminary solution, the efficiency of the optimized axial flow pump is increased by 1.615%, with a wider high-efficiency range and an optimal operating point closer to the design conditions. Overall, this intelligent optimization method has the potential to significantly reduce the design time of axial pumps and increase their performance.

Big Data and Doing Research in the Management Discipline

정의교 한국무역연구원 2018 무역연구 Vol.14 No.5

We argue that big data should be understood as an indispensable element in a wider context of big data science that also includes machine learning and results interpretations. By addressing this wider context, we examine the differences between big data science and modern sciences in general and management discipline in particular. While the former adopts data-driven approach to enhance predictive accuracy, the latter adopts theory-driven approach to produce causal explanation. Data-driven approach in conjunction with machine learning strives to enhance the predictive accuracy by allowing big data to choose a set of parameters on its own under rather loose assumptions and learning processes. In contrast, management discipline emphasizes the role of theories in deriving testable hypotheses and encourages scholars to present compelling arguments without explicitly referring to data to be used for estimation at a later stage. This implies that management discipline may not benefit much from big data science in doing academic research. But we believe that big data may prove helpful for the management discipline if we carefully identify small but meaningful patterns that are not easily detected in small data. We also argue that sampling is still an important issue in using big data for academic research.

그래프 기반 준지도 학습을 이용한 속성값 전파 결측치 추정

신유경(Yukyung Shin),신현정(Hyunjung Shin) 한국정보과학회 2019 정보과학회 컴퓨팅의 실제 논문지 Vol.25 No.10

데이터의 레코드들 중에 하나 이상의 속성값이 없는 경우는 비일비재하다. 많은 경우에 있어서 데이터의 수 대비 결측치가 없는 완전레코드의 수의 비율이 적다. 이에 대하여 평균값, 최빈값, 그리고 중앙값 등으로 대체하는 통계적 방법이 가장 보편적으로 쓰이고 있다. 또한 기계학습에서도 k-최근접 이웃탐색이나 의사결정나무 등을 활용한 결측치 추정방법들이 자주 활용된다. 전자는 각 속성의 대표하는 값으로 대체하는 전역적 방법인데 반해 후자는 해당 레코드와 유사한 레코드들의 속성값으로 대체하는 지역적 방법이라 할 수 있다. 그러나 한 속성의 값이 대부분 결측된 경우라면 두 방법 모두 활용하기 어렵다. 이러한 한계를 극복하기 위하여, 본 연구에서는 결측치의 속성과 상관성이 큰 이웃 속성들로부터 값을 추정하는 방법을 제안한다. 속성 간 상관성을 기반으로 하여 한 속성의 대부분의 값이 결측이 되더라도 활용할 수 있다. 제안 방법론으로는 속성들 간의 상관계수로 이루어진 상관 그래프를 만들고, 그래프 기반 준지도 학습을 적용한다. 결측치는 다른 속성값들로부터 상관계수에 비례하여 전파되어 추정된다. 본 논문에서 제안한 결측치 대체 추정 방법과 기존에 결측치 대체에 많이 사용하는 통계적 방법과 기계학습을 비교하여 실험을 진행하였다. The number of data records without one or more attributes is very large. In many cases, few complete records are available without missing the data values. Statistical methods that replace the missing values with mean, mode and median are commonly used. In machine learning algorithms such as K-nearest neighborhood or decision tree, the missing values are replaced by estimation methods. The statistical method is a global method that replaces each attribute with a representative value, whereas the machine learning algorithm is a local method that replaces the attribute values similar to the records. However, it is difficult to use both methods for records that contain almost all the missing values. In order to overcome these limitations, in this paper, we propose a method to estimate values from neighborhood properties associated with large correlation with the missing attribute. It is based on correlation between attributes, and can be used even if the attributes carry almost missing values. In this proposed method, a correlation graph representing correlation coefficients related to attribute values was constructed based on graph-based semi-supervised learning. Missing values were estimated in proportion to the correlation coefficient derived from related attributes. In this paper, the proposed method compared the statistical method and machine learning algorithm, which are generally used for missing value imputation.

A Stochastic Optimality Theoretic Study of the Adaptation of Korean Syllable Final Consonants to Chinese by Korean Wave Fans in China

Seoyoung HAN(?瑞英) 한국중어중문학회 2021 한국중어중문학회 우수논문집 Vol.- No.-

This study investigates the phonetic adaptation of the syllable final [p], [t], [k], [m], [l] from Korean to Chinese based on stochastic Optimality Theory. Since [p], [t], [k], [m], [l] are prohibited at the coda position in Chinese, they were variably repaired to satisfy native phonotactics. After describing the variable adaptation patterns of K-pop lyrics and basic Korean expressions shared on Baidu Zhidao, specific weights of regarding constraints were calculated based on machine learning using Maximum Entropy Modeling. Theoretical implications on the non-typical quality of epenthetic vowels, the preferences for consonant deletion, and the irrelevance between preceding vowels and adaptation typology were discussed. To conclude, phonotactics on the Chinese syllable final position conveys both non-categorical and categorical characteristics at the same time.