Alper Sinan Akyurek

TESLA: Taylor expanded solar analog forecasting

With the increasing penetration of renewable energy resources within the Smart Grid, solar forecasting has become an important problem for hour-ahead and day-ahead planning. Within this work, we analyze the Analog Forecast method family, which uses past observations to improve the forecast product. We first show that the frequently used euclidean distance metric has drawbacks and leads to poor performance relatively. In this paper, we introduce a new method, TESLA forecasting, which is very fast and light, and we show through case studies that we can beat the persistence method, a state of the art comparison method, by up-to 50% in terms of root mean square error to give an accurate forecasting result. An extension is also provided to improve the forecast accuracy by decreasing the forecast horizon.

User Behavior Modeling for Estimating Residential Energy Consumption

Residential energy constitutes a significant portion of the total US energy consumption. Several researchers proposed energy-aware solutions for houses, promising significant energy and cost savings. However, it is important to evaluate the outcomes of these methods on larger scale, with hundreds of houses. This paper presents a human-activity based residential energy modeling framework, that can create power demand profiles considering the characteristics of household members. It constructs a mathematical model to show the detailed relationships between human activities and house power consumption. It can be used to create various house profiles with different energy demand characteristics in a reproducible manner. Comparison with real data shows that our model captures the power demand differences between different family types and accurately follows the trends seen in real data. We also show a case study that evaluates voltage deviation in a neighborhood, which requires accurate estimation of the trends in power consumption.

Distributed control of a swarm of buildings connected to a smart grid: demo abstract

Energy-efficient control mechanisms are necessary to manage the ever increasing energy demand. Recently several tools for building energy consumption control have been proposed for small (e.g. homes) [8] and large (e.g. offices) buildings [3][6][1]. The mechanism each tool uses is different, e.g. HVAC control [3] and appliance rescheduling [8], but they share the goal of improving consumption of the buildings with respect to a given cost function. Some examples of cost functions are reduced energy consumption, reduced electricity bill, lower peak power, and increased ancillary service participation. The tools however do not capture the impacts of their control actions on the grid. These actions can lead to supply/demand imbalance and voltage/frequency deviation and thus, threaten grid stability. Utilities can take protective actions against those who cause instability by increasing electricity price or even momentarily disconnecting them from the grid. The effects of these protective actions can be so severe that the savings obtained by building management tools might disappear.

A Modular Approach to Context-Aware IoT Applications

The Internet of Things (IoT) refers to an environment of ubiquitous sensing and actuation, where devices are connected to a distributed backend infrastructure. It offers the opportunity to access a large amount of input data, and process it into contextual information about different system entities for reasoning and actuation. State-of-the-art IoT applications are generally black-box, end-to-end application-specific implementations, and cannot keep up with timely resolution of all this live, continually updated, heterogeneous data. In this work, we propose a modular approach to these context-aware applications, breaking down monolithic applications into an equivalent set of functional units, or context engines. By exploiting the characteristics of context-aware applications, context engines can reduce compute redundancy and computational complexity. In conjunction with formal data specifications, or ontologies, we can replace application-specific implementations with a composition of context engines that use common statistical learning to generate output, thus improving context reuse. We implement interconnected context-aware applications using our approach, extracting both user activity and location context from wearable sensors. We compare our infrastructure to single-stage monolithic implementations, demonstrating a reduction in application latency by up to 65% and execution overhead by up to 50% with only a 3% reduction in accuracy.

ECO-DAC Energy Control over Divide and Control

The need for a smarter grid is emerging with the increase of peak demand and the integration of renewable resources. A great solution for peak shifting and renewable energy smoothing is through the usage of energy storage devices. This paper focuses on the energy storage power control problem in small to medium sized power distribution systems with loads, energy storage devices and renewable resources connected to the grid. To the best of our knowledge, solutions in this area either focus on the optimization problem with a convex optimization solver, that have high worst-case complexities or on sub-optimal heuristics. This paper provides a low-complexity solution, ECO-DAC, which is optimal in terms of minimizing a multi-tier cost function. We show on multiple case studies that it is possible to save up to 21% in costs for electricity drawn from the grid, compared to the no-battery case.

Modular and Personalized Smart Health Application Design in a Smart City Environment

The Internet of Things (IoT) envisions to create a smart, connected city that is composed of ubiquitous environmental and user sensing along with distributed, low-capacity computing. This provides ample information regarding the citizens in various smart environments. We can leverage this people-centric information, provided by the smart city infrastructure, to improve “smart health” applications: user data from connected wearable devices can be accompanied with ubiquitous environmental sensing and versatile actuation. The state-of-the-art in smart health applications is black-box, end-to-end implementations which are neither intended for use with heterogeneous data nor adaptable to a changing set of sensing and actuation. In this paper, we apply our modular approach for IoT applications—the context engine—to smart health problems, enabling the ability to grow with available data, use general-purpose machine learning, and reduce compute redundancy and complexity. For smart health, this improves response times for critical situations, more efficient identification of health-related conditions and subsequent actuation in a smart city environment. We demonstrate the potential with three sets of interconnected context-aware applications, extracting health-related people-centric context, such as user presence, user activity, air quality, and location from IoT sensors.

Optimal Distributed Nonlinear Battery Control

Energy storage plays a more important role than ever before, due to the transition to smart grid along with higher penetration of renewable resources. In this paper, we describe our optimal nonlinear battery control algorithm that can handle multiple batteries connected to the grid in a distributed and cost-optimal fashion, while maintaining low complexity of

Time-series clustering for data analysis in Smart Grid

O(N^{2})

Dynamic Optical Switching for Latency Sensitive Applications

. In contrast to the state-of-the-art models, we use a high accuracy nonlinear battery model with 2% error. We present three distributed solutions: 1) a circular negotiation ring, providing convergence rates independent of the number of batteries; 2) a mean circular negotiation ring, converging very quickly for a low number of batteries; 3) a bisection method has a convergence rate independent of battery capacities. We compare our algorithm to the state-of-the-art algorithms and show that we can decrease the utility cost of an actual building by up to 50% compared with the batteryless case by 30% over the load-following heuristic and by 60% over a state-of-the-art optimal control algorithm designed using a linear battery model. For a constant load profile, optimal linear control incurs costs higher by 150% for model predictive control and 250% for single-trajectory solutions than for our algorithm.

Context-aware and user-centric residential energy management

The increasingly pervasive deployment of networked sensors in the Smart Grid for monitoring energy consumption has resulted in an unprecedentedly large amount of data generation. Efficient methods are required to understand this high volume and high dimensional data on an embedded platform, which has many challenges due to memory, processing and power constraints. One of the popular methods to analyze time-series data is clustering. In this paper, we discuss D-Stream II, a common time-series clustering algorithm, and demonstrate that it fails to obtain clusters in sample Smart Grid applications. Then, we propose an enhanced version of this algorithm, which handles the scenarios where the original algorithm fails. We show the effectiveness of our algorithm using real residential power consumption data from Pecan Street database. Our enhanced algorithm efficiently handles the high volume energy consumption data and captures the everyday energy consumption patterns of each residential home, which allow the consumer to compare energy consumption with their neighbors or detect any abnormality such as a defective appliance. The consumers can also be clustered into different groups, which can be effectively used to enhance the demand response policies. Our algorithm can perform clustering on approximately half a million energy consumption data points using only 5KB max memory, 360J max overhead in energy consumption and can complete in 8 mins on a resource limited embedded platform (Raspberry Pi 2).