Balakrishnan Narayanaswamy

Learning to be energy-wise: discriminative methods for load disaggregation

In this paper we describe an ongoing project which develops an automated residential Demand Response (DR) system that attempts to manage residential loads in accordance with DR signals. In this early stage of the project, we propose an approach for identifying individual appliance consumption from the aggregate load and discuss the effectiveness of load disaggregation techniques when total load data also includes appliances that are unmonitored even during the training phase. We show that simple discriminative methods can directly predict the appliance states (e.g. on, off, standby) and the predicted state can be used to calculate energy consumed by the appliances. We also show that these methods perform substantially better than the generative models of energy consumption that are commonly used. We evaluated the proposed approach using publicly available REDD data set, and our experimental evaluation demonstrates the improvement in accuracy.

Iterative Cross-Entropy Encoding for Memory Systems with Stuck-At Errors

In this paper, a novel iterative encoding scheme is proposed for memory systems suffering from stuck-at errors. The stuck-at errors can be efficiently managed by using side information about stuck-at memory cells during encoding, while encoding for unconstrained number of stuck-at errors is intractable due to its exponential complexity. The proposed coding scheme employs an iterative encoding algorithm using cross-entropy method, which has a polynomial time complexity. In addition, any linear block code (LBC) can be concatenated with the proposed code, to correct for both residual stuck-at errors and random (soft) errors. The proposed coding schemes are evaluated by numerical simulations using a memory channel undergoing both stuck-at and random errors. Simulation results show that the cross-entropy based coding scheme provides an improved block error rate (BLER) performance, or alternatively, a higher overall storage capacity.

Prediction based storage management in the smart grid

Economic and environmental concerns have fostered interest in incorporating greater amounts of electricity from renewable energy sources into the grid. Unfortunately, the intermittent availability of renewable power has raised a barrier to the inclusion of these sources. Distributed storage is perceived as a means to extract value from the different resources. However, the large cost of storage requires the design of algorithms that can manage intermittent resources with minimum storage size. At the same time, advances in metering, communication, and weather prediction allow real time management of energy generation, distribution and consumption based on predictions of the future. In this paper, we focus on online algorithms for local storage management that use short term predictions of intermittent renewable resource availability. In contrast to prior work, we develop algorithms that come with theoretical bounds on performance even when demand, prices and availability are arbitrary (possibly non-stochastic), and the utility functions non-concave. Our fundamental contribution is to prove how appropriate discounting of future welfare leads to storage management algorithms that exhibit excellent practical performance even in the worst-case scenario. We substantiate these theoretical guarantees with experiments that demonstrate the effectiveness of our algorithms and the value of storage in the smart grid.

Hedging strategies for renewable resource integration and uncertainty management in the smart grid

Increased environmental and economic concerns have set the stage for an increase in the fraction of electricity supplied using renewable sources. Recent advances in wind prediction offer hope that reduction in the uncertainty of wind availability will lead to an increase in its value. Model based methods that predict future wind availability and then optimize local generation have been seen to be successful for both economic dispatch and demand management. In this paper we evaluate model free hedging strategies for renewable resource integration and uncertainty management in the smart grid. We compare the performance of these two classes of algorithms for intelligent generator scheduling using simple wind speed forecasters in both simulations and on real wind traces. We also suggest that algorithms based on online convex optimization can be applied to demand management problems and evaluate hedging algorithms for smart demand response, highlighting the reduction in costs possible when renewable energy is combined with demand response.

Estimating the wake losses in large wind farms: A machine learning approach

Estimating the wake losses in a wind farm is critical in the short term forecast of wind power, following the Numerical Weather Prediction (NWP) approach. Understanding the intensity of the wakes and the nature of its propagation within the wind farm still remains a challenge to scientist, engineers and utility operators. In this paper, five different machine learning methods are used to estimate the power deficit experienced by wind turbines due to the wake losses. Production data from the Horns Rev offshore wind farm, Denmark, have been used for the study. The methods used are linear regression, linear regression with feature engineering, nonlinear regression, Artificial Neural Networks (ANN) and Support Vector Regression (SVR). Power developed by individual turbines located at different positions within the farm were computed based on the above methods and compared with the actual power measurements. With the respective Variance Normalized Root Mean Square Error (VNRMSE) of 0.21 and 0.22, models based on ANN and SVR could estimate the wind farm wake effects at an acceptable accuracy level. The study shows that suitable machine learning methods can effectively be used in estimating the power deficits due to wake effects experienced in large wind farms.

Optimal utilization of power transformers through virtual sensing

Power transformers are the most critical and expensive assets for utility companies and are expected to last for at-least 30-40 years. Unfortunately, many of them have failed before reaching their rated life. In order to prevent such premature failures, utility companies usually deploy real time asset management system by installing thermal sensors (e.g. fiber-optical sensor) inside the tank of large transformers (~500MVA). However, being expensive, sensor deployment may not be commercially viable for small and medium size (<;=250MVA) transformers. This paper proposes an asset management scheme of such power transformers through virtual (sensor-less) sensing. It simulates transformer internal heating phenomena (like hot-spot temperature, insulation aging, loss of life, etc.) using easily available SCADA measurements, ambient temperature from on-line weather forecast, and transformer assets data. It predicts future load and incentive and optimizes transformer operation by analyzing the economic incentive for carrying power and payoff for the loss of life calculated from virtual sensing. Proposed scheme is evaluated and implemented in Fortums transformers in Finland and experimental results are presented.

Congestion control of smart distribution grids using state estimation

Power utilities worldwide face two major challenges - peak demand and power (supply-demand) imbalance. In the midst of these difficulties faced by utilities, growing fuel costs, environmental awareness and government directives have increased the push to deploy Electric Vehicles (EVs). One single EV being charged at its peak rate imposes an instantaneous load equivalent to that of 10 average households on the grid, making it essential to schedule the EV charging in order to prevent grid failures. Our approach to this problem is motivated by parallels to the development of the internet and in particular internet protocols such as TCP, where agents respond to signals from the central authority to curtail load when the grid is congested. We show that using high resolution measurements from smart meters and distribution feeders and without measurements at any intermediate nodes, we can use recently developed semi-definite programming based state estimation techniques to accurately infer the state of the gird. We then show how to convert this line level congestion information into signal loads to users to curtail usage. In combination with smart home agents that automatically control consumption, we show how this state estimation and signaling protocol leads to reduced congestion and losses while minimizing user inconvenience.

Loss localisation in smart distribution networks

A sizable fraction of expensive resources such as water, gas, and electricity are lost in the process of distribution due to leaks, theft, and inefficiencies like aging infrastructure. Utilities worldwide are looking for solutions to localise losses and improve the overall efficiency of their distribution systems. The introduction of new sensing, communication, and control infrastructure in the next generation smart distribution networks offers hope that fine grained knowledge of the system will allow for more efficient operation. In this work, we present a novel loss localisation system that uses only a time series of inflow and outflow measurements to localise and quantify losses in tree distribution networks. The inflow and outflow measurements are used to set up a system of linear equations based on the principle of conservation of flows and these are solved to estimate loss rates of source destination paths and then localise losses on each internal link in the tree. We draw parallels with loss localisation in computer networks and show how ideas from network tomography could be applied in these settings. We demonstrate the efficacy of proposed estimators using preliminary simulation experiments.