Refine
Year of publication
Document Type
- Article (53)
- Conference Proceeding (29)
- Master's Thesis (14)
- Working Paper (12)
- Report (11)
- Bachelor Thesis (8)
- Part of a Book (3)
- Study Thesis (3)
- Preprint (2)
- Book (1)
Language
- English (137) (remove)
Has Fulltext
- yes (137) (remove)
Keywords
- 1D-CNN (1)
- 2 D environment Laser data (1)
- 3D Extended Object Tracking (1)
- 3D shape tracking (1)
- Abschätzung (1)
- Accelerometer (1)
- Additive manufacturing (1)
- Agile administration (1)
- Alpine area (1)
- Alternative Energy Production (1)
Institute
- Fakultät Architektur und Gestaltung (2)
- Fakultät Bauingenieurwesen (16)
- Fakultät Elektrotechnik und Informationstechnik (6)
- Fakultät Informatik (12)
- Fakultät Maschinenbau (6)
- Fakultät Wirtschafts-, Kultur- und Rechtswissenschaften (12)
- Institut für Angewandte Forschung - IAF (18)
- Institut für Optische Systeme - IOS (9)
- Institut für Strategische Innovation und Technologiemanagement - IST (2)
- Institut für Systemdynamik - ISD (13)
Further applications of the Cauchon algorithm to rank determination and bidiagonal factorization
(2018)
For a class of matrices connected with Cauchon diagrams, Cauchon matrices, and the Cauchon algorithm, a method for determining the rank, and for checking a set of consecutive row (or column) vectors for linear independence is presented. Cauchon diagrams are also linked to the elementary bidiagonal factorization of a matrix and to certain types of rank conditions associated with submatrices called descending rank conditions.
The class of square matrices of order n having a negative determinant and all their minors up to order n-1 nonnegative is considered. A characterization of these matrices is presented which provides an easy test based on the Cauchon algorithm for their recognition. Furthermore, the maximum allowable perturbation of the entry in position (2,2) such that the perturbed matrix remains in this class is given. Finally, it is shown that all matrices lying between two matrices of this class with respect to the checkerboard ordering are contained in this class, too.
In 1970, B.A. Asner, Jr., proved that for a real quasi-stable polynomial, i.e., a polynomial whose zeros lie in the closed left half-plane of the complex plane, its finite Hurwitz matrix is totally nonnegative, i.e., all its minors are nonnegative, and that the converse statement is not true. In this work, we explain this phenomenon in detail, and provide necessary and sufficient conditions for a real polynomial to have a totally nonnegative finite Hurwitz matrix.
Positive systems play an important role in systems and control theory and have found applications in multiagent systems, neural networks, systems biology, and more. Positive systems map the nonnegative orthant to itself (and also the non-positive orthant to itself). In other words, they map the set of vectors with zero sign variation to itself. In this article, discrete-time linear systems that map the set of vectors with up to k-1 sign variations to itself are introduced. For the special case k = 1 these reduce to discrete-time positive linear systems. Properties of these systems are analyzed using tools from the theory of sign-regular matrices. In particular, it is shown that almost every solution of such systems converges to the set of vectors with up to k-1 sign variations. It is also shown that these systems induce a positive dynamics of k-dimensional parallelotopes.
Forecasting is crucial for both system planning and operations in the energy sector. With increasing penetration of renewable energy sources, increasing fluctuations in the power generation need to be taken into account. Probabilistic load forecasting is a young, but emerging research topic focusing on the prediction of future uncertainties. However, the majority of publications so far focus on techniques like quantile regression, ensemble, or scenario-based methods, which generate discrete quantiles or sets of possible load curves. The conditioned probability distribution remains unknown and can only be estimated when the output is post-processed using a statistical method like kernel density estimation.
Instead, the proposed probabilistic deep learning model uses a cascade of transformation functions, known as normalizing flow, to model the conditioned density function from a smart meter dataset containing electricity demand information for over 4,000 buildings in Ireland. Since the whole probability density function is tractable, the parameters of the model can be obtained by minimizing the negative loglikelihood through the state of the art gradient descent. This leads to the model with the best representation of the data distribution.
Two different deep learning models have been compared, a simple three-layer fully connected neural network and a more advanced convolutional neural network for sequential data processing inspired by the WaveNet architecture. These models have been used to parametrize three different probabilistic models, a simple normal distribution, a Gaussian mixture model, and the normalizing flow model. The prediction horizon is set to one day with a resolution of 30 minutes, hence the models predict 48 conditioned probability distributions.
The normalizing flow model outperforms the two other variants for both architectures and proves its ability to capture the complex structures and dependencies causing the variations in the data. Understanding the stochastic nature of the task in such detail makes the methodology applicable for other use cases apart from forecasting. It is shown how it can be used to detect anomalies in the power grid or generate synthetic scenarios for grid planning.
Probabilistic Short-Term Low-Voltage Load Forecasting using Bernstein-Polynomial Normalizing Flows
(2021)
The transition to a fully renewable energy grid requires better forecasting of demand at the low-voltage level. However, high fluctuations and increasing electrification cause huge forecast errors with traditional point estimates. Probabilistic load forecasts take future uncertainties into account and thus enables various applications in low-carbon energy systems. We propose an approach for flexible conditional density forecasting of short-term load based on Bernstein-Polynomial Normalizing Flows where a neural network controls the parameters of the flow. In an empirical study with 363 smart meter customers, our density predictions compare favorably against Gaussian and Gaussian mixture densities and also outperform a non-parametric approach based on the pinball loss for 24h-ahead load forecasting for two different neural network architectures.
Non-volatile NAND flash memories store information as an electrical charge. Different read reference voltages are applied to read the data. However, the threshold voltage distributions vary due to aging effects like program erase cycling and data retention time. It is necessary to adapt the read reference voltages for different life-cycle conditions to minimize the error probability during readout. In the past, methods based on pilot data or high-resolution threshold voltage histograms were proposed to estimate the changes in voltage distributions. In this work, we propose a machine learning approach with neural networks to estimate the read reference voltages. The proposed method utilizes sparse histogram data for the threshold voltage distributions. For reading the information from triple-level cell (TLC) memories, several read reference voltages are applied in sequence. We consider two histogram resolutions. The simplest histogram consists of the zero-and-one ratios for the hard decision read operation, whereas a higher resolution is obtained by considering the quantization levels for soft-input decoding. This approach does not require pilot data for the voltage adaptation. Furthermore, only a few measurements of extreme points of the threshold voltage distributions are required as training data. Measurements with different conditions verify the proposed approach. The resulting neural networks perform well under other life-cycle conditions.
The growing error rates of triple-level cell (TLC) and quadruple-level cell (QLC) NAND flash memories have led to the application of error correction coding with soft-input decoding techniques in flash-based storage systems. Typically, flash memory is organized in pages where the individual bits per cell are assigned to different pages and different codewords of the error-correcting code. This page-wise encoding minimizes the read latency with hard-input decoding. To increase the decoding capability, soft-input decoding is used eventually due to the aging of the cells. This soft-decoding requires multiple read operations. Hence, the soft-read operations reduce the achievable throughput, and increase the read latency and power consumption. In this work, we investigate a different encoding and decoding approach that improves the error correction performance without increasing the number of reference voltages. We consider TLC and QLC flashes where all bits are jointly encoded using a Gray labeling. This cell-wise encoding improves the achievable channel capacity compared with independent page-wise encoding. Errors with cell-wise read operations typically result in a single erroneous bit per cell. We present a coding approach based on generalized concatenated codes that utilizes this property.
In the field of autonomously driving vehicles the environment perception containing dynamic objects like other road users is essential. Especially, detecting other vehicles in the road traffic using sensor data is of utmost importance. As the sensor data and the applied system model for the objects of interest are noise corrupted, a filter algorithm must be used to track moving objects. Using LIDAR sensors one object gives rise to more than one measurement per time step and is therefore called extended object. This allows to jointly estimate the objects, position, as well as its orientation, extension and shape. Estimating an arbitrary shaped object comes with a higher computational effort than estimating the shape of an object that can be approximated using a basic geometrical shape like an ellipse or a rectangle. In the case of a vehicle, assuming a rectangular shape is an accurate assumption.
A recently developed approach models the contour of a vehicle as periodic B-spline function. This representation is an easy to use tool, as the contour can be specified by some basis points in Cartesian coordinates. Also rotating, scaling and moving the contour is easy to handle using a spline contour. This contour model can be used to develop a measurement model for extended objects, that can be integrated into a tracking filter. Another approach modeling the shape of a vehicle is the so-called bounding box that represents the shape as rectangle.
In this thesis the basics of single, multi and extended object tracking, as well as the basics of B-spline functions are addressed. Afterwards, the spline measurement model is established in detail and integrated into an extended Kalman filter to track a single extended object. An implementation of the resulting algorithm is compared with the rectangular shape estimator. The implementation of the rectangular shape estimator is provided. The comparison is done using long-term considerations with Monte Carlo simulations and by analyzing the results of a single run. Therefore, both algorithms are applied to the same measurements. The measurements are generated using an artificial LIDAR sensor in a simulation environment.
In a real-world tracking scenario detecting several extended objects and measurements that do not originate from a real object, named clutter measurements, is possible. Also, the sudden appearance and disappearance of an object is possible. A filter framework investigated in recent years that can handle tracking multiple objects in a cluttered environment is a random finite set based approach. The idea of random finite sets and its use in a tracking filter is recapped in this thesis. Afterwards, the spline measurement model is included in a multi extended object tracking framework. An implementation of the resulting filter is investigated in a long-term consideration using Monte Carlo simulations and by analyzing the results of a single run. The multi extended object filter is also applied to artificial LIDAR measurements generated in a simulation environment.
The results of comparing the spline based and rectangular based extended object trackers show a more stable performance of the spline extended object tracker. Also, some problems that have to be addressed in future works are discussed. The investigation of the resulting multi extended object tracker shows a successful integration of the spline measurement model in a multi extended object tracker. Also, with these results some problems remain, that have to be solved in future works.
In 3D extended object tracking (EOT), well-established models exist for tracking the object extent using various shape priors. A single update, however, has to be performed for every measurement using these models leading to a high computational runtime for high-resolution sensors. In this paper, we address this problem by using various model-independent downsampling schemes based on distance heuristics and random sampling as pre-processing before the update. We investigate the methods in a simulated and real-world tracking scenario using two different measurement models with measurements gathered from a LiDAR sensor. We found that there is a huge potential for speeding up 3D EOT by dropping up to 95\% of the measurements in our investigated scenarios when using random sampling. Since random sampling, however, can also result in a subset that does not represent the total set very well, leading to a poor tracking performance, there is still a high demand for further research.
In this paper, a novel measurement model based on spherical double Fourier series (DFS) for estimating the 3D shape of a target concurrently with its kinematic state is introduced. Here, the shape is represented as a star-convex radial function, decomposed as spherical DFS. In comparison to ordinary DFS, spherical DFS do not suffer from ambiguities at the poles. Details will be given in the paper. The shape representation is integrated into a Bayesian state estimator framework via a measurement equation. As range sensors only generate measurements from the target side facing the sensor, the shape representation is modified to enable application of shape symmetries during the estimation process. The model is analyzed in simulations and compared to a shape estimation procedure using spherical harmonics. Finally, shape estimation using spherical and ordinary DFS is compared to analyze the effect of the pole problem in extended object tracking (EOT) scenarios.
In the past years, algorithms for 3D shape tracking using radial functions in spherical coordinates represented with different methods have been proposed. However, we have seen that mainly measurements from the lateral surface of the target can be expected in a lot of dynamic scenarios and only few measurements from the top and bottom parts leading to an error-prone shape estimate in the top and bottom regions when using a representation in spherical coordinates. We, therefore, propose to represent the shape of the target using a radial function in cylindrical coordinates, as these only represent regions of the lateral surface, and no information from the top or bottom parts is needed. In this paper, we use a Fourier-Chebyshev double series for 3D shape representation since a mixture of Fourier and Chebyshev series is a suitable basis for expanding a radial function in cylindrical coordinates. We investigate the method in a simulated and real-world maritime scenario with a CAD model of the target boat as a reference. We have found that shape representation in cylindrical coordinates has decisive advantages compared to a shape representation in spherical coordinates and should preferably be used if no prior knowledge of the measurement distribution on the surface of the target is available.