In this work, an extension of the federated averaging algorithm, FedAvg-Gaussian, is applied to train probabilistic neural networks. The performance advantage of probabilistic prediction models is demonstrated and it is shown that federated learning can improve driving range prediction. Using probabilistic predictions, routing and charge planning based on destination attainability can be applied. Furthermore, it is shown that probabilistic predictions lead to reduced travel time.
This work aims at improving the energy consumption forecast of electric vehicles by enhancing the prediction with a notion of uncertainty. The algorithm itself learns from driver and traffic data in a training set to generate accurate, driver-individual energy consumption forecasts.
Properly determining the driving range is critical for accurately predicting the sales and social benefits of battery electric vehicles (BEVs). This study proposes a framework for optimizing the driving range by minimizing the sum of battery price, electricity cost, and range limitation cost referred to as the "range-related cost" as a measurement of range anxiety. The objective function is linked to policy-relevant parameters, including battery cost and price markup, battery utilization, charging infrastructure availability, vehicle efficiency, electricity and gasoline prices, household vehicle ownership, daily driving patterns, discount rate, and perceived vehicle lifetime. Qualitative discussion of the framework and its empirical application to a sample (N=36664) representing new car drivers in the United States is included. The quantitative results strongly suggest that ranges of less than 100 miles are likely to be more popular in the BEV market for a long period of time. The average optimal range among U.S. drivers is found to be largely inelastic. Still, battery cost reduction significantly drives BEV demand toward longer ranges, whereas improvement in the charging infrastructure is found to significantly drive BEV demand toward shorter ranges. In conclusion, the bias of a single-range assumption and the effects of range optimization and diversification in reducing such biases are both found to be significant.
The principal advantage of smart electricity meters is their ability to transfer digitized electricity consumption data to remote processing systems. The data collected by these devices make the realization of many novel use cases possible, providing benefits to electricity providers and customers alike. This book includes 14 research articles that explore and exploit the information content of smart meter data, and provides insights into the realization of new digital solutions and services that support the transition towards a sustainable energy system. This volume has been edited by Andreas Reinhardt, head of the Energy Informatics research group at Technische Universität Clausthal, Germany, and Lucas Pereira, research fellow at Técnico Lisboa, Portugal.
As the nation's economic activities, security concerns, and stewardship of natural resources become increasingly complex and globally interrelated, they become ever more sensitive to adverse impacts from weather, climate, and other natural phenomena. For several decades, forecasts with lead times of a few days for weather and other environmental phenomena have yielded valuable information to improve decision-making across all sectors of society. Developing the capability to forecast environmental conditions and disruptive events several weeks and months in advance could dramatically increase the value and benefit of environmental predictions, saving lives, protecting property, increasing economic vitality, protecting the environment, and informing policy choices. Over the past decade, the ability to forecast weather and climate conditions on subseasonal to seasonal (S2S) timescales, i.e., two to fifty-two weeks in advance, has improved substantially. Although significant progress has been made, much work remains to make S2S predictions skillful enough, as well as optimally tailored and communicated, to enable widespread use. Next Generation Earth System Predictions presents a ten-year U.S. research agenda that increases the nation's S2S research and modeling capability, advances S2S forecasting, and aids in decision making at medium and extended lead times.
As an important enabler for changing people’s lives, advances in artificial intelligence (AI)-based applications and services are on the rise, despite being hindered by efficiency and latency issues. By focusing on deep learning as the most representative technique of AI, this book provides a comprehensive overview of how AI services are being applied to the network edge near the data sources, and demonstrates how AI and edge computing can be mutually beneficial. To do so, it introduces and discusses: 1) edge intelligence and intelligent edge; and 2) their implementation methods and enabling technologies, namely AI training and inference in the customized edge computing framework. Gathering essential information previously scattered across the communication, networking, and AI areas, the book can help readers to understand the connections between key enabling technologies, e.g. a) AI applications in edge; b) AI inference in edge; c) AI training for edge; d) edge computing for AI; and e) using AI to optimize edge. After identifying these five aspects, which are essential for the fusion of edge computing and AI, it discusses current challenges and outlines future trends in achieving more pervasive and fine-grained intelligence with the aid of edge computing.
Over the last few years, interest in the industrial applications of AI and learning systems has surged. This book covers the recent developments and provides a broad perspective of the key challenges that characterize the field of Industry 4.0 with a focus on applications of AI. The target audience for this book includes engineers involved in automation system design, operational planning, and decision support. Computer science practitioners and industrial automation platform developers will also benefit from the timely and accurate information provided in this work. The book is organized into two main sections comprising 12 chapters overall: •Digital Platforms and Learning Systems •Industrial Applications of AI
This open access book is part of the LAMBDA Project (Learning, Applying, Multiplying Big Data Analytics), funded by the European Union, GA No. 809965. Data Analytics involves applying algorithmic processes to derive insights. Nowadays it is used in many industries to allow organizations and companies to make better decisions as well as to verify or disprove existing theories or models. The term data analytics is often used interchangeably with intelligence, statistics, reasoning, data mining, knowledge discovery, and others. The goal of this book is to introduce some of the definitions, methods, tools, frameworks, and solutions for big data processing, starting from the process of information extraction and knowledge representation, via knowledge processing and analytics to visualization, sense-making, and practical applications. Each chapter in this book addresses some pertinent aspect of the data processing chain, with a specific focus on understanding Enterprise Knowledge Graphs, Semantic Big Data Architectures, and Smart Data Analytics solutions. This book is addressed to graduate students from technical disciplines, to professional audiences following continuous education short courses, and to researchers from diverse areas following self-study courses. Basic skills in computer science, mathematics, and statistics are required.
