Smart Service Systems, Operations Management, and Analytics

Inhaltsverzeichnis

1. Frontmatter

2. Cleaning and Processing on the Electric Vehicle Telematics Data

   Shuai Sun, Jun Bi, Cong Ding

   Abstract

   The development of the Internet of Vehicles (IoV) enables companies to collect an increasing amount of telematics data, which creates plenty of new business opportunities. How to improve the integrity and precision of electric vehicle telematics data to effectively support the operation and management of vehicles is one of the thorniest problems in the electric vehicle industry. With the purpose of accurately collecting and calculating the driving mileage of electric vehicles, a series of data cleaning and processing methodologies were conducted on the real-world electric vehicle telematics data. More specifically, descriptive statistics was conducted on the data, and the statistical results showed the quality of the data in general. Above all, the driving mileage data were segmented according to the rotate speed of the electric motor, and the anomaly threshold of the driving mileage data was obtained by the box-plot method. Then, the typical anomalies in the data were screened out by the threshold and analysed, respectively. Ultimately, the real-time and offline abnormal processing algorithms are designed to process real-time and offline data, respectively. After debugging and improvement, these two sets of abnormal processing algorithms we designed have been able to run on a company’s big data cloud platform. According to the feedback of the operation results of real-world massive data, the two sets of algorithms can effectively improve the statistical accuracy of driving mileage data of electric vehicle.

3. Performance Analysis of a Security-Check System with Four Types of Inspection Channels for High-Speed Rail Stations in China

   Chia-Hung Wang, Xiaojing Wu

   Abstract

   In recent years, the High-Speed Rail (HSR) in China has continued to thrive rapidly with the development of China’s booming economy, and it has become the preferred mode of transportation for many travelers. This paper investigates the stochastic process of security inspection for passengers in the high-speed rail station. A queuing model is developed for studying the proposed security-check system via computer simulation. In the numerical experiments, we illustrate the influence of varying model parameters on the average waiting time and safety level of the queuing system. The sensitivity analysis of our simulation model could contribute to improve the service level and efficiency on security check before implementing operation plans for high-speed rail stations.

4. LSTM-Based Neural Network Model for Semantic Search

   Xiaoyu Guo, Jing Ma, Xiaofeng Li

   Abstract

   To improve web search quality and serve a better search experience for users, it is important to capture semantic information from user query which contains user’s intention in web search. Long Short-Term Memory (LSTM), a significant network in deep learning has made tremendous achievements in capturing semantic information and predicting the semantic relatedness of two sentences. In this study, considering the similarity between predicting the relatedness of sentence pair task and semantic search, we provide a novel channel to process semantic search task: see semantic search as an atypical predicting the relatedness of sentence pair task. Furthermore, we propose an LSTM-Based Neural Network Model which is suitable for predicting the semantic relatedness between user query and potential documents. The proposed LSTM-Based Neural Network Model is trained by Home Depot dataset. Results show that our model outperforms than other models.

5. Research on the Evaluation of Electric Power Companies’ Safety Capabilities Based on Grey Fixed Weight Clustering

   Yijing Wang, Jie Xu, Chuanmin Mi, Zhipeng Zhou

   Abstract

   With the development of power grid and the widespread use of power, the impact of power safety is becoming more and more important. Meanwhile, Internet and Internet of Things have both provided new technical and management ideas for power safety. Combining literature research and field investigation on safety management of Jiangsu Electric Power Company, this paper focuses on solving the problem of electric power companies’ safety capabilities. First, based on the idea of data-driven management and decision-making, power safety risks are identified according to electric power companies’ actual business situation and therefore, safety capability evaluation index system is created. Second, the weight of each indicator is determined according to the proportion of safety risks in historical accidents. Finally, considering the problem of information lack in electric power safety management in China, this paper uses Grey Fixed Weight Clustering Theory including determining thresholds of each indicator and constructing whitening weight functions to make an evaluation of electric power companies’ safety capabilities which could help managers learn about their own existing safety levels and provide them a reference for future decisions.

6. Analysis of Crude Oil Price Fluctuation and Transition Characteristics at Different Timescales Based on Complex Networks

   Jiao Yan, Jing Ma

   Abstract

   Based on the theory of complex network, the crude price data for 30 years is processed by coarse-graining method and the price fluctuation network of crude oil is constructed. According to the price of crude oil at different timescales, this paper establishes the different complex network models to compare the topological properties of crude oil price fluctuation network and explores the internal characteristics of the original oil price fluctuation system. Meanwhile, using the K-core, the obtained fluctuation loops at different timescales are analyzed to achieve certain prediction effect. As is shown in the study, small-world phenomenon and scale-free characteristics exist in the crude oil price fluctuation network at different timescales. However, with the increase of timescale, the small-world phenomenon of the crude oil price fluctuation network is enhanced, while the scale-free properties are weakened. Ultimately, this paper summarizes the differences and relations between the evolution of crude oil price fluctuation under different scales, and the next research direction is put forward.

7. Understanding of Servicification Trends in China Through Analysis of Interindustry Network Structure

   Yunhan Liu, Dohoon Kim

   Abstract

   This study analyses the core changes in the entire industrial structure of China from 2002 to 2015. First, utilizing the input–output model (or interindustry relation model), we evaluate the production-inducing effects across the various manufacturing and service sectors. Employing standard centrality measures that are often used in network analysis, we identify the status of various industry sectors together with their structural roles on the input–output network and track the evolutionary paths as they change over time. Furthermore, the clustering analysis of industrial sectors also examines critical changes in the overall industrial structure. We will discuss the policy implications of these evolutionary paths and explore the trends and sources of Chinese economic development regarding manufacturing and service productions. As a result, it can be seen that the production-inducing effects in the manufacturing sectors—particularly, chemical, electric power, and primary metal products—play the core role in driving the economy. On the other hand, the distribution sector like the traditional wholesale and retail services still shows unstable connections with other industries. However, the sectors like finance and transportation, which constitute another axis of the service industry, have strengthened their connection with the manufacturing sectors since 2012. Moreover, the real estate and lease and the other business support services continue to absorb production resources, deepen their linkages with the key service industries (e.g., transportation), and the various manufacturing sectors. In conclusion, China still has manufacturing as a central driving force for economic development, but with the increasing integration with the service sectors, both manufacturing and service industries show an apparent convergence in their production activities.

8. Machine Learning Methods for Revenue Prediction in Google Merchandise Store

   Vahid Azizi, Guiping Hu

   Abstract

   Machine learning has gained increasing interests from various application domains for its ability to understand data and make predictions. In this paper, we apply machine learning techniques to predict revenue per customer for Google Merchandise Store. Exploratory Data Analysis (EDA) was conducted for the customer dataset and feature engineering was applied to the find best subset of features. Four machine learning methods, Gradient Boosting Machine (GBM), Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM) have been applied to predict revenue per customer. Results show that LightGBM outperforms other methods in terms of RMSE and running time.

9. Predicting Metropolitan Crime Rates Using Machine Learning Techniques

   Saba Moeinizade, Guiping Hu

   Abstract

   The concept of smart city has been gaining public interests with the considerations of socioeconomic development and quality of life. Smart initiatives have been proposed in multiple domains, such as health, energy, and public safety. One of the key factors that impact the quality of life is the crime rate in a metropolitan area. Predicting crime patterns is a significant task to develop more efficient strategies either to prevent crimes or to improve the investigation efforts. In this research, we use machine learning techniques to solve a multinomial classification problem where the goal is to predict the crime categories with spatiotemporal data. As a case study, we use San Francisco crime data from San Francisco Police Department (SFPD). Various classification methods such as Multinomial Logistic Regression, Random Forests, Lightgbm, and Xgboost have been adopted to predict the category of crime. Feature engineering was employed to boost the model performance. The results demonstrate that our proposed classifier outperforms other published models.

10. Optimizing Ensemble Weights for Machine Learning Models: A Case Study for Housing Price Prediction

    Mohsen Shahhosseini, Guiping Hu, Hieu Pham

    Abstract

    Designing ensemble learners has been recognized as one of the significant trends in the field of data knowledge, especially, in data science competitions. Building models that are able to outperform all individual models in terms of bias, which is the error due to the difference in the average model predictions and actual values, and variance, which is the variability of model predictions, has been the main goal of the studies in this area. An optimization model has been proposed in this paper to design ensembles that try to minimize bias and variance of predictions. Focusing on service sciences, two well-known housing datasets have been selected as case studies: Boston housing and Ames housing. The results demonstrate that our designed ensembles can be very competitive in predicting the house prices in both Boston and Ames datasets.

11. How Do Pricing Power and Service Strategy Affect the Decisions of a Dual-Channel Supply Chain?

    Houping Tian, Chaomei Wu

    Abstract

    With the rapid development of e-commerce, an increasing amount of manufacturers are implementing a dual-channel strategy, i.e., introducing an online direct sale channel on the existing traditional retail channel. However, it is noted that for the small and medium-sized manufacturers, they may face the resistance of the dominant retailer caused by this strategy. It is well known as channel conflict, and some challenging issues remain to be further explored, e.g., How to price and coordinate the dual-channel under different pricing power structures? And, if possible, when should the retailer offer the value-added service to enhance his competitiveness and mitigate the channel conflict? The paper compares the decision models under a different power structure (i.e., Manufacturer-Stackelberg, Retailer-Stackelberg, and Vertical Nash) and service strategy (providing the value-added service or not); the analyses show two important findings. First, for the manufacturer or the retailer, they both prefer to be the leader to enjoy the power advantage and higher profits. However, for the entire supply chain and the consumers, they prefer equal power structure to the other two power arrangements. Second, the retailer will get more profits while providingvalue-added service. Interestingly, it is also beneficial to the manufacturer.

12. Designing Value Co-creation for a Free-Floating e-Bike-Sharing System

    Christoph Heitz, Marc Blume, Corinne Scherrer, Raoul Stöckle, Thomas Bachmann

    Abstract

    Value co-creation requires a system that links actors together for mutual value creation. In our paper, we describe the development of such a system in the context of the new free-floating e-bike-sharing system (BSS) in Zurich, Switzerland. This BSS is based on the idea that users of the BSS co-create value by adapting their usage behavior such that the overall service level is maximized. This creates value for other users and reduces the provider’s costs for redistribution, but requires some kind of incentive system for influencing the user behavior. We describe a systematic approach of designing such a system by operationalizing the concept of value and value generation for the different actors: What exactly is the value that is to be created, and how can it be measured? By which activities is value created, and what are the options for stimulating these activities? Which design options maximize value creation? We found that this required combining two different research approaches: Empirical social research was necessary to understand user needs, value perception, motivational patterns in response to incentives, and communication needs. Operational research was necessary for assessing different options for the incentive system with respect to the value creation both for provider and users. By interlinking both research activities, we were able to design an incentive system that allows reducing the number of bikes by 30% without diminishing the service level. Users are offered a reward for dropping-off their bikes in dynamically changing reward zones whose locations are determined based on the bike distribution and the future demand pattern. These incentives lead to two distinct behavioral responses which were assessed and quantified in an extensive real-life field test during a period of 13 weeks. The impact of the measured behavioral change on service level and the required number of bikes was modeled via simulation.

13. Research on Electricity Falling Accident Based on Improved Bode Accident Causation Model

    Qian Yuanyuan, Xu Jie, Mi Chuanmin, Peng Qiwei

    Abstract

    In view of the applicability of accident causation models to different types of accidents, this paper comprehensively considers various kinds of accident causation theories, compares, and analyses seven kinds of accident causation models, and points out the limitations of existing models. Then, according to the mechanism of electricity falling accident, combining the theory of Bode accident causality chain with the theory of man–machine–environment system, this paper presents an accident causation model suitable for the analysis of electricity falling accident, which has a strong guiding significance for the prevention of electricity falling accident. Finally, the management improvement measures are put forward from the aspects of safety operation standard, safety production responsibility system, safety education, and training, in order to prevent the recurrence of electricity falling accident.

14. Crop Yield Prediction Using Deep Neural Networks

    Saeed Khaki, Lizhi Wang

    Abstract

    The world’s population is on the rise and in order to feed the world in 2050, food production will need to increase by 70% [1]. As a result, it is of great importance to construct powerful predictive models for phenotype prediction based on Genotype and Environment data (so-called G by E problem). The objective of the G by E analysis is to understand how genotype and the environment jointly determine the phenotype (such as crop yield and disease resistance) of plant or animal species. In this research, deep neural networks are trained and used as predictive models. Deep neural networks have become a popular tool in supervise learning due to considerable ability in training nonlinear features [5]. Recent articles have stated that the network depth is a vital factor in decreasing classification or regression error. But, deeper networks have a so-called vanishing/exploding gradients problem which makes the training and optimizing deeper networks difficult. He et al. proposed residual learning method which alleviates this problem very well and showed that deep residual networks are significantly better and more efficient than previous typical networks [5]. As a result, residual training has been used in this research to prevent gradient degradation and ease the optimization process. Finally, since it is difficult to predict the yield difference directly, two separate residual neural networks have been trained to predict yield and check yield. After training the networks, the RMSE for check yield and yield are 8.23 and 10.52, respectively, which are very good because of considerable amount of missing values, uncertainty, and complexity in the datasets.

15. Cloud-Based Life Sciences Manufacturing System: Integrated Experiment Management and Data Analysis via Amazon Web Services

    Pei Guo, Raymond Peterson, Paul Paukstelis, Jianwu Wang

    Abstract

    A vital need in the life sciences industry is software that manages large amounts of fast-moving data for manufacturing quality assurance, clinical diagnostics, and research. In the life sciences industry and research labs, lab information management systems (LIMS) are often used to manage expensive lab instruments. We propose a new software architecture for cloud-based life sciences manufacturing system through the following two advances: (1) full life cycle support of life science experiment through cloud services, (2) workflow-based easy and automatic experiment management and data analysis. This paper discusses our software architecture and implementation on top of Amazon Web Services by utilizing its services including Lambda architecture, API gateway, serverless computing, and Internet of Things (IoT) services. We demonstrate its usage through a real-world life sciences instrument and experimental use case. To our best knowledge, it is the first work on supporting integrated experiment design, experiment instrument operation, experiment data storage, and experiment data analysis all in the cloud for the life sciences.

16. Matching Anonymized Individuals with Errors for Service Systems

    Wai Kin (Victor) Chan

    Abstract

    Data privacy is of great importance for the healthy development of service systems. Companies and governments that provide services to people often have big concerns in sharing their data. Because of that, data must be preprocessed (e.g., anonymized) before they can be shared. However, without identification, it is difficult to match data from different sources and thus the data cannot be used together. This paper investigates how the performance of two simple individual matching methods was affected by errors in the similarity scores between individuals. The first method is a greedy method (GM) that simply matches individuals based on the maximum similarity scores. The second method is an optimal assignment problem (AP), which maximizes the total similarity scores of the matched individuals. Consistent with the literature, we found that GM outperforms AP in most situations. However, we also discovered that AP could be better in fixing errors.

17. Developing a Production Structure Model Using Service-Dominant Logic—A Hypergraph-Based Modeling Approach

    Mahei Manhai Li, Christoph Peters, Jan Marco Leimeister

    Abstract

    To make a fundamental shift toward value orientation, manufacturing companies strategically move to integrate services into their portfolio. While manufacturing firms rely on production information systems as the backbone of their operations, these systems are based on product structure models (e.g., bill of materials). This poses a problem because services do not adhere to the goods-dominant perspective of product structures. To solve this divide, this paper proposes an integrative mathematical model for both production systems and service systems. The model draws upon concepts of service-dominant logic and is based on hypergraph theory. To illustrate that the production structure model includes both product structures and process structures, we further demonstrate that the production structure model can be transformed into either. Therefore, our theoretical contribution lies in introducing a structural model into production systems that is compatible with structures of a service system model. For practice, the model enables the development of production information systems that can plan and control products, services, and hybrids.

18. Airworthiness Evaluation Model Based on Fuzzy Neural Network

    Jie-Ru Jin, Peng Wang, Yang Shen, Kai-Xi Zhang

    Abstract

    Based on fuzzy neural network, this study explores the quantifying calculation problem of airworthiness of specific rescue operations. Aiming at the problem of quantification assessment of flightworthiness for rescue flight operation, this study starts from the perspective of the mechanical properties of the aircraft itself, and use a fuzzy neural network model for rescue operations of airworthiness evaluation modeling. The rescue historical data of the EC-135 helicopter model is used for model training, in order to form a quantitative model of airworthiness assessment for flight operation. On the one hand, the quantitative output results provide qualitative guidance for aircraft’s competency, and on the other hand, it provides a basis for comparing the advantages and disadvantages of multitask allocation schemes. Optimization of the whole system task allocation effect is formed through the best way of individual utility. Aiming at the heterogeneous problem for rescue system, this paper introduces the concept of “task ability vector”, quantitative representing the ability of heterogeneous aircrafts and requirements of mission. The comprehensive ability of quantitative calculation of multi-aircrafts alliance is discussed, as well as the single and multi-aircrafts cooperative task ability.