Raffi Sevlian

Detection and Statistics of Wind Power Ramps

Ramps events are a significant source of uncertainty in wind power generation. Developing statistical models from historical data for wind power ramps is important for designing intelligent distribution and market mechanisms for a future electric grid. This requires robust detection schemes for identifying wind ramps in data. In this paper, we propose an optimal detection technique for identifying wind ramps for large time series. The technique relies on defining a family of scoring functions associated with any rule for defining ramps on an interval of the time series. A dynamic programming recursion is then used to find all such ramp events. Identified wind ramps are used to propose a new stochastic framework to characterize wind ramps. Extensive statistical analysis is performed based on this framework, characterizing ramping duration and rates as well as other key features needed for evaluating the impact of wind ramps in the operation of the power system. In particular, evaluation of new ancillary services and wind ramp forecasting can benefit from the proposed approach.

RSSI-Fingerprinting-Based Mobile Phone Localization With Route Constraints

Accurate positioning of a moving vehicle along a route enables various applications, such as travel-time estimation, in transportation. Global Positioning System (GPS)-based localization algorithms suffer from low availability and high energy consumption. A received signal strength indicator (RSSI) measured in the course of the normal operation of Global System for Mobile Communications (GSM)-based mobile phones, on the other hand, consumes minimal energy in addition to the standard cell-phone operation with high availability but very low accuracy. In this paper, we incorporate the fact that the motion of vehicles satisfies route constraints to improve the accuracy of the RSSI-based localization by using a hidden Markov model (HMM), where the states are segments on the road, and the observation at each state is the RSSI vector containing the detected power levels of the pilot signals sent by the associated and neighboring cellular base stations. In contrast to prior HMM-based models, we train the HMM based on the statistics of the average drivers behavior on the road and the probabilistic distribution of the RSSI vectors observed in each road segment. We demonstrate that this training considerably improves the accuracy of the localization and provides localization performance robust over different road segment lengths by using extensive cellular data collected in Istanbul, Turkey; Berkeley, CA, USA; and New Delhi, India.

Optimal electric energy storage operation

Estimating the arbitrage value of storage is an important problem in power systems planning. Various studies have reported different values based numerical solutions of variations of a basic model. In this paper, we instead rely on a closed form solution for storage control. The closed form highlights the right type of forecasting that is required and allows large horizon problems to be solved. We study various scenarios and provide a simple methodology for evaluating the arbitrage value of storage.

Channel characterization for 700 MHz DSRC vehicular communication

Adapting OFDM for vehicular communication requires extensive knowledge of anticipated multipath and Doppler environments. We present a GPS-enabled channel sounding system built and used to conduct a channel measurement campaign. Tests conducted at the 700MHz band in and around downtown Ann Arbor, Michigan, explored various vehicle-to-vehicle and vehicle-to-roadside channel scenarios. The measured channelmetrics are used to quantify the effects on guard interval, packet duration, and subcarrier spacing for a functional OFDM system at 700 MHz. This paper is one of the first to present vehicular-based channel-modeling results from measured data in the 700MHz band.

Wind power ramps: Detection and statistics

Ramps events are a significant source of uncertainty in wind power generation. Developing statistical models from historical data for wind power ramps is important for designing intelligent distribution and market mechanisms for a future electric grid. This requires robust detection schemes for identifying wind ramps in data. In this paper, use an optimal detection technique for identifying wind ramps for large time series. The technique relies on defining a family of scoring functions associated with any rule for defining ramps on an interval of the time series. A dynamic programming recursion is then used to find all such ramp events. Identified wind ramps are used to perform an extensive statistical analysis on the process, characterizing ramping duration and rates as well as other key features needed for developing future models.

Outage detection in power distribution networks with optimally-deployed power flow sensors

An outage detection framework for power distribution networks is proposed. The framework combines the use of optimally deployed real-time power flow sensors and that of load estimates via Advanced Metering Infrastructure (AMI) or load forecasting mechanisms. The distribution network is modeled as a tree network. It is shown that the outage detection problem over the entire network can be decoupled into detection within subtrees, where within each subtree only the sensors at its root and on its boundary are used. Outage detection is then formulated as a hypothesis testing problem, for which a maximum a-posteriori probability (MAP) detector is applied. Employing the maximum misdetection probability Pmaxe as the detection performance metric, the problem of finding a set of a minimum number of sensors that keeps Pmaxe below any given probability target is formulated as a combinatorial optimization. Efficient algorithms are proposed that find the globally optimal solutions for this problem, first for line networks, and then for tree networks. Using these algorithms, optimal three-way tradeoffs between the number of sensors, the load estimate accuracy, and the outage detection performance are characterized for line and tree networks using the IEEE 123 node test feeder system.

Value of aggregation in smart grids

The projected smart grid intends to deliver a variety of services to energy consumers. Many of these services rely on aggregates of energy consumers. Forecasting energy consumption of groups of users is a crucial aspect of this process. This paper addresses electricity load forecasting on varying scales of aggregation. Using city wide consumption data at 1 hour interval standard forecasting methods are applied to the data. It is shown that under some ideal conditions, forecasting errors decrease on as Θ(1/√N). We show empirically the true forecasting errors decrease as Θ(1/Nα) with α <; 0.5.

A model for the effect of aggregation on short term load forecasting

In this work, we propose a simple empirical scaling law that describes load forecasting accuracy at different levels of aggregation. We show that for the short term forecasting problem, aggregating more users will improve the relative forecasting performance up to a point. Beyond this point, no more improvement in relative performance can be obtained.

Segmenting consumers using smart meter data

Existing electricity market segmentation analysis techniques only make use of limited consumption statistics (usually averages and variances). In this paper we use power demand distributions (PDDs) obtained from fine-grain smart meter data to perform market segmentation based on distributional clustering. We apply this approach to mining 8 months of readings from about 1000 US Google employees.

Aggregation for load servicing

The proliferation of smart meters enables a load-serving entity (LSE) to aggregate customers according to their consumption patterns. We demonstrate a method for constructing groups of customers who will be the cheapest to service at wholesale market prices. Using smart meter data from a region in California, we show that by aggregating more of these customers together, their consumption can be forecasted more accurately, which allows an LSE to mitigate financial risks in its wholesale market transactions. We observe that the consumption of aggregates of customers with similar consumption patterns can be forecasted more accurately than that of random aggregates of customers. The model we propose enables an LSE to offer discounted rates to low-cost customers because it can purchase electricity for them more cheaply than it can for the general population.