Blackout

About: Blackout is a research topic. Over the lifetime, 2088 publications have been published within this topic receiving 30433 citations.

BlackOut: Speeding up Recurrent Neural Network Language Models With Very Large Vocabularies

Abstract: We propose BlackOut, an approximation algorithm to efficiently train massive recurrent neural network language models (RNNLMs) with million word vocabularies. BlackOut is motivated by using a discriminative loss, and we describe a new sampling strategy which significantly reduces computation while improving stability, sample efficiency, and rate of convergence. One way to understand BlackOut is to view it as an extension of the DropOut strategy to the output layer, wherein we use a discriminative training loss and a weighted sampling scheme. We also establish close connections between BlackOut, importance sampling, and noise contrastive estimation (NCE). Our experiments, on the recently released one billion word language modeling benchmark, demonstrate scalability and accuracy of BlackOut; we outperform the state-of-the art, and achieve the lowest perplexity scores on this dataset. Moreover, unlike other established methods which typically require GPUs or CPU clusters, we show that a carefully implemented version of BlackOut requires only 1-10 days on a single machine to train a RNNLM with a million word vocabulary and billions of parameters on one billion words. Although we describe BlackOut in the context of RNNLM training, it can be used to any networks with large softmax output layers.

Genetic epidemiology of alcohol-induced blackouts.

Abstract: Background Alcohol-induced blackouts (ie, periods of anterograde amnesia) have received limited recent research attention. Objective To examine the genetic epidemiology of lifetime blackouts and having had 3 or more blackouts in a year, including analyses controlling for the frequency of intoxication. Design, Setting, and Participants Members of the young adult Australian Twin Register, a volunteer twin panel born between January 1, 1964, and December 31, 1971, were initially registered with the panel as children by their parents between 1980 and 1982. They underwent structured psychiatric telephone interviews from February 1996 through September 2000. The current sample contains 2324 monozygotic and dizygotic twin pairs (mean [SD] age 29.9 [2.5] years) for whom both twins' responses were coded for blackout questions and for frequency of intoxication. Main Outcome Measure Data on lifetime blackouts and having had 3 or more blackouts in a year were collected within an examination of the genetic epidemiology of alcoholism. Results A lifetime history of blackouts was reported by 39.3% of women and 52.4% of men; 11.4% of women and 20.9% of men reported having had 3 or more blackouts in a year. The heritability of lifetime blackouts was 52.5% and that of having had 3 or more blackouts in a year was 57.8%. Models that controlled for frequency of intoxication found evidence of substantial genetic contribution unique to risk for the blackouts and a significant component of genetic risk shared with frequency of intoxication. Conclusions The finding of a substantial genetic contribution to liability for alcohol-induced blackouts including a component of genetic loading shared with frequency of intoxication may offer important additional avenues to investigate susceptibility to alcohol-related problems.

Load recognition for automated demand response in microgrids

Abstract: Microgrids are well-suited for electrification of remote off-grid areas. This paper sketches the concept of a plug-and-play microgrid with a minimum of configuration effort needed for setup. When the load of such an off-grid microgrid grows over the generation capacity and energy storage is not sufficient, demand has to be reduced to prevent a blackout. In order to decide which loads are inessential and can be shedded, automated load recognition on the basis of measured power consumption profiles is needed. Two promising approaches from the area of speech recognition, Dynamic Time Warping and Hidden Markov Models, are compared for this application. It is found that a key feature to achieve good recognition efficiency is a careful selection of the features extracted from the measured power data.

Interim report : Causes of the August 14th blackout in the United States and Canada

Abstract: A U.S.-Canada Power System Outage Task Force was created following a massive power outage on August 14, 2003 which affected approximately 50 million people in the Midwest and Northeast United States and Ontario, Canada. The Task Force divided its work into two phases. The first phase of work was to investigate the outage, and to determine its causes and why it was not contained. The second phase involved recommendations to reduce the possibility of future outages and to minimize the scope of any that may occur. The chapters of this report dealt with the following issues: the North American electric power system and its reliability organizations; status of the Northeastern power grid before the blackout sequence began; how and why the blackout began; the cascade stage of the blackout; the August 14 blackout compared with previous major North American outages; performance of nuclear power plants affected by the blackout; and, physical and cyber security aspects of the blackout. Three public forums will be held in which the public will have the opportunity to comment on the report and present recommendations for consideration by the Task Force. The report indicates that the loss of FirstEnergy's overloaded Sammis-Star line triggered the cascade. Its 345-kV line into northern Ohio from eastern Ohio began tripping out because the lines were in contact with overgrown trees. The loss of the line created major and unsustainable burdens on lines in adjacent areas. The cascade spread rapidly as lines and generating units automatically took themselves out of service to avoid physical damage. The blackout had many contributing factors in common with earlier outages including: inadequate tree trimming, failure to identify emergency conditions, inadequate operator training, and inadequate regional-scale visibility over the power system. Contributing factors to the August 14 blackout included: inadequate inter-regional visibility over the power system; dysfunction of control area SCADA/EMS system; and, a lack of adequate backup capability to that system. It was noted that the nuclear power plants did not trigger the blackout. All nuclear plants shutdown and responded in a manner consistent with plant designs and safety functions. Analysis to date shows no evidence that a malicious cyber event played a role in the power outage. 4 tabs., 47 figs., 4 appendices.

Reactive power dispatch incorporating voltage stability

Abstract: The cause of the 1977 New York blackout has been proved to be the reactive-power problem. The 1987 Tokyo blackout was believed to be due to a reactive-power shortage and a voltage collapse at the summer peak load. These events have confirmed the importance of reactive-power planning and dispatching in maintaining the security of modern power systems. Appropriate control of system voltage profiles can enhance system security and may also reduce system losses. Previous investigations by the same authors have shown that a voltage collapse proximity indicator based on the optimal impedance solution of a two-bus system can be extended to an actual network. The authors describe a linear reactive power dispatch algorithm which incorporates this indicator to minimise the possibility of voltage collapse in the system.