Search

EnergyPlus is the U.S. Department of Energy’s flagship whole-building energy simulation engine and provides extensive simulation capabilities. However, the computational cost of these capabilities has resulted in annual building simulations that typically requires 2–3 min of wall-clock time to complete. While EnergyPlus’s overall speed is improving (EnergyPlus 7.0 is 25–40% faster than EnergyPlus 6.0), the overall computational burden still remains and is the top user complaint. In other engineering domains, researchers substitute surrogate or approximate models for the computationally expensive simulations to improve simulation and reduce calibration time. Previous work has successfully demonstrated small-scale EnergyPlus surrogate models that use 10–16 input variables to estimate a single output variable. This work leverages feed forward neural networks and Lasso regression to construct robust large-scale EnergyPlus surrogate models based on 3 benchmark datasets that have 7–156 inputs. These models were able to predict 15-min values for most of the 80–90 simulation outputs deemed most important by domain experts within 5% (whole building energy within 0.07%) and calculate those results within 3 s, greatly reducing the required simulation runtime for relatively close results. The techniques shown here allow any software to be approximated by machine learning in a way that allows one to quantify the trade-off of accuracy for execution time.

Building energy modeling BEM is an approach to model the energy usage in buildings for design and retrofit purposes. EnergyPlus is the flagship Department of Energy software that performs BEM for different types of buildings. The input to EnergyPlus can often extend in the order of a few thousand parameters that have to be calibrated manually by an expert for realistic energy modeling. This makes it challenging and expensive thereby making BEM unfeasible for smaller projects. In this paper, we describe the 'Autotune' research that employs machine learning algorithms to generate agents for the different kinds of standard reference buildings in the US building stock. The parametric space and the variety of building locations and types make this a challenging computational problem necessitating the use of supercomputers. Millions of EnergyPlus simulations are run on supercomputers that are subsequently used to train machine learning algorithms to generate agents. These agents, once created, can then run in a fraction of the time thereby allowing cost-effective calibration of building models. Published 2014. This article is a US Government work and is in the public domain in the USA.

Traditional whole building energy modeling suffers from several factors, including the large number of inputs required for building characterization, simplifying assumptions, and the gap between the as-designed and as-built building. Prior work has attempted to mitigate these problems by using sensor-based machine learning approaches to statistically model energy consumption, applying the techniques primarily to commercial building data, which makes use of hourly consumption data. It is unclear, however, whether these techniques can translate to residential buildings, since the energy usage patterns may vary significantly. Until now, most residential modeling research only had access to monthly electrical consumption data. In this article, we report on the evaluation of seven different machine learning algorithms applied to a new residential data set that contains sensor measurements collected every 15 min, with the objective of determining which techniques are most successful for predicting next hour residential building consumption. We first validate each learner's correctness on the ASHRAE Great Energy Prediction Shootout, confirming existing conclusions that Neural Network-based methods perform best on commercial buildings. However, our additional results show that these methods perform poorly on residential data, and that Least Squares Support Vector Machines perform best – a technique not previously applied to this domain.

This report provides a summary of Phase I activities conducted to support an Integrated Evaluation of Mercury in Liquid Waste System (LWS) Processing Facilities. Phase I activities included a review and assessment of the liquid waste inventory and chemical processing behavior of mercury using a system by system review methodology approach. Gaps in understanding mercury behavior as well as action items from the structured reviews are being tracked. 64% of the gaps and actions have been resolved.

Many key building design policies are made using sophisticated computer simulations such as EnergyPlus (E+), the DOE flagship whole-building energy simulation engine. E+ and other sophisticated computer simulations have several major problems. The two main issues are 1) gaps between the simulation model and the actual structure, and 2) limitations of the modeling engine's capabilities. Currently, these problems are addressed by having an engineer manually calibrate simulation parameters to real world data or using algorithmic optimization methods to adjust the building parameters. However, some simulations engines, like E+, are computationally expensive, which makes repeatedly evaluating the simulation engine costly. This work explores addressing this issue by automatically discovering the simulation's internal input and output dependencies from ~20 Gigabytes of E+ simulation data, future extensions will use ~200 Terabytes of E+ simulation data. The model is validated by inferring building parameters for E+ simulations with ground truth building parameters. Our results indicate that the model accurately represents parameter means with some deviation from the means, but does not support inferring parameter values that exist on the distribution's tail.

Building Energy Modeling (BEM) is an approach to model the energy usage in buildings for design and retrofit purposes. EnergyPlus is the flagship Department of Energy software that performs BEM for different types of buildings. The input to EnergyPlus can often extend in the order of a few thousand parameters which have to be calibrated manually by an expert for realistic energy modeling. This makes it challenging and expensive thereby making building energy modeling unfeasible for smaller projects. In this paper, we describe the "Autotune" research which employs machine learning algorithms to generate agents for the different kinds of standard reference buildings in the U.S. building stock. The parametric space and the variety of building locations and types make this a challenging computational problem necessitating the use of supercomputers. Millions of EnergyPlus simulations are run on supercomputers which are subsequently used to train machine learning algorithms to generate agents. These agents, once created, can then run in a fraction of the time thereby allowing cost-effective calibration of building models.

Probabilistic latent semantic analysis (PLSA) has been widely used in the machine learning community. However, the original PLSAs are not capable of modeling real-valued observations and usually have severe problems with over fitting. To address both issues, we propose a novel, regularized Gaussian PLSA (RG-PLSA) model that combines Gaussian PLSAs and hierarchical Gaussian mixture models (HGMM). We evaluate our model on supervised human action recognition tasks, using two publicly available datasets. Average classification accuracies of 97.69% and 93.72% are achieved on the Weizmann and KTH Action Datasets, respectively, which demonstrate that the RG-PLSA model outperforms Gaussian PLSAs and HGMMs, and is comparable to the state of the art.

There is a growing interest in building Internet-scale sensor networks that integrate sensors from around the world into a single unified system. In contrast, robotics application development has primarily focused on building specialized systems. These specialized systems take scalability and reliability into consideration, but generally neglect exploring the key components required to build a large scale system. Integrating robotic applications with Internet-scale sensor networks will unify specialized robotics applications and provide answers to large scale implementation concerns. We focus on utilizing Internet-scale sensor network technology to construct a framework for unifying robotic systems. Our framework web-enables a surveillance robot's sensor observations and provides a web-interface to the robot's actuators. This lets robots seamlessly integrate into web applications. In addition, the framework eliminates most prerequisite robotics knowledge, allowing for the creation of general web-based robotics applications. The framework also provides mechanisms to create applications that can interface with any robot. Frameworks such as this one are key to solving large scale mobile robotics implementation problems. We provide an overview of previous Internet-scale sensor networks, Sensorpedia (an ad-hoc Internet-scale sensor network), our framework for integrating robots with Sensorpedia, two applications which illustrate our frameworks ability to support general web-based robotic control, and offer experimental results that illustrate our framework's scalability, feasibility, and resource requirements.

The classical technique for identifying cells engaged in DNA synthesis is by their uptake of [(3)H]-thymidine, detected using autoradiography. However, this method can be inconvenient, as specialized darkroom and radioisotope facilities are required, with the potential health hazard that handling isotopes entails. Bromodeoxyuridine (BrdU), the halogenated 5-substituted derivative of deoxyuridine, is a thymidine analog specifically incorporated into the DNA of proliferating cells during S phase. This is now a well-established alternative to (3)H thymidine, since it has been shown that labeling indices for the two molecules are the same (1,2). The development of a monoclonal antibody (3) that recognizes BrdU incorporated into single-stranded DNA has resulted in several techniques using immunocytochemical staining to detect incorporated BrdU in frozen, paraffin- and plastic-embedded sections of tissue by light microscopy. It has also proved extremely valuable for studies in conjunction with flow cytometry and even, for in vivo studies of human tumor cell kinetics (see this vol., Chapter 43 ). We describe here a method to detect DNA synthesis by in vivo labeling of nuclei with BrdU, followed by indirect immunological detection in paraffin-embedded tissue (4).

Erythropoietic protoporphyria patients can develop cholestasis, severe hepatic damage, fibrosis, and cirrhosis. We modeled this hepatic pathology in C57BL/6J and BALB/c mice using griseofulvin and analyzed 3,127 genes for alteration of expression in the liver before and during the onset of protoporphyria, cholestasis, inflammation, and hepatic fibrosis. The two mouse strains developed different levels of pathologic damage in response to the griseofulvin. Characteristic gene expression profiles could be associated with griseofulvin-induced gene expression, disruption of lipid metabolism, and the pathologic states of inflammation, early fibrosis, and cholestasis. Additionally, some genes individually indicated an alteration of homeostasis. or pathologic state; for example, fibroblast proliferation was potentially indicated by increased calcyclin (SA100a6) expression. Changes in cytochrome P450 (Cyp) gene expression were particularly pronounced, with increased expression of the Cyp2a, Cyp2b, and Cyp3a families. Decreased Cyp4a10 and Cyp4a14 expression was observed that could be associated with early pathologic change. A potential decrease in bile acid and steroid biosynthesis was indicated by the decreased expression of Cyp7b1 and Hsd3b4, respectively. DNA damage was indicated by induction of GADD45. This study illustrates how transcriptional programs can be associated with different stimuli in the same experiment. The time course of change in the gene expression profile compared with changes in pathology and clinical chemistry shows the potential of this approach for modeling causative, predictive, and adaptive changes in gene expression during pathologic change.

This report presents data obtained from two line-oriented simulations conducted in the Boeing 757/767 simulators. The purpose of these simulations was to assess the impact of two navigation- and two flight control modes on pilot visual behavior during an entire flight, from takeoff to touchdown. The two navigation modes were the traditional VOR mode, in which a compass rose was presented on the HSI, and the MAP mode, in which a pictorial representation of the airplane's flight path was presented on the HSI. The flight control modes were manual and coupled flight. The results indicated that: (a) the electronic map did not alter the basic “T” scan pattern, (b) the pilots' basic scan pattern did not differ in the two navigation modes when flying manually, (c) pilot scan patterns did vary as a function of flight control condition, with the basic “T” scan pattern accurately characterizing pilot visual behavior in the manual mode but not in the coupled mode, (d) several visual performance measures were sensitive to changes in flight phase, navigation mode, and flight control mode, and (e) no differences in pilot scan patterns were observed between an EICAS- equipped cockpit and a cockpit with conventional engine instruments.

In two experiments, each S received four successive trials in a distractor-type short-term memory task. A rest period was interpolated between Trials 3 and 4. Proactive interference (PI) developed across the first three trials and partially dissipated with rest. However, for all rest intervals tested (up to 5 min), the recovery from PI was increased by a shift in taxonomic class. It was suggested that performance in the release-from-PI paradigm is mediated by a combination of long-term and short-term memory processes.

Two experiments investigated the release from PI following a shift in mode of presentation of the study words. In each experiment, the interpolated task required processing of both auditory (A) and visual (V) information. The results were the same with mixed lists (Experiment I) and with independent groups (Experiment II): A release from PI was obtained following an A to V shift but not following a V to A shift. It was concluded that the mixed-modality filler task has functional characteristics very similar to those of an A filler and that the filler task influences the memory code for study items.