ANNO: A Time Series Annotation Tool to Evaluate Event Detection Algorithms

2020 ◽  
pp. 70-87
Author(s):  
Jana Huchtkoetter ◽  
Andreas Reinhardt ◽  
Sakif Hossain
Keyword(s):  
Time Series ◽  
Event Detection ◽  
Annotation Tool ◽  
Detection Algorithms

scholarly journals A Demand-Side Load Event Detection Algorithm Based on Wide-Deep Neural Networks and Randomized Sparse Backpropagation

2021 ◽  
Vol 9 ◽  
Author(s):  
Chen Li ◽  
Gaoqi Liang ◽  
Huan Zhao ◽  
Guo Chen
Keyword(s):  
Machine Learning ◽  
Time Series ◽  
Deep Learning ◽  
Event Detection ◽  
Detection Algorithm ◽  
Demand Side ◽  
Rule Based ◽  
Detection Algorithms ◽  
Deep Model ◽  
Conventional Machine

Event detection is an important application in demand-side management. Precise event detection algorithms can improve the accuracy of non-intrusive load monitoring (NILM) and energy disaggregation models. Existing event detection algorithms can be divided into four categories: rule-based, statistics-based, conventional machine learning, and deep learning. The rule-based approach entails hand-crafted feature engineering and carefully calibrated thresholds; the accuracies of statistics-based and conventional machine learning methods are inferior to the deep learning algorithms due to their limited ability to extract complex features. Deep learning models require a long training time and are hard to interpret. This paper proposes a novel algorithm for load event detection in smart homes based on wide and deep learning that combines the convolutional neural network (CNN) and the soft-max regression (SMR). The deep model extracts the power time series patterns and the wide model utilizes the percentile information of the power time series. A randomized sparse backpropagation (RSB) algorithm for weight filters is proposed to improve the robustness of the standard wide-deep model. Compared to the standard wide-deep, pure CNN, and SMR models, the hybrid wide-deep model powered by RSB demonstrates its superiority in terms of accuracy, convergence speed, and robustness.

scholarly journals Explaining Bad Forecasts in Global Time Series Models

2021 ◽  
Vol 11 (19) ◽  
pp. 9243
Author(s):  
Jože Rožanec ◽  
Elena Trajkova ◽  
Klemen Kenda ◽  
Blaž Fortuna ◽  
Dunja Mladenić
Keyword(s):  
Time Series ◽  
Model Performance ◽  
Time Series Forecasting ◽  
Modular Architecture ◽  
Detection Algorithms ◽  
Training Samples ◽  
Forecasting Performance ◽  
Global Time ◽  
Real World Datasets ◽  
Value Changes

While increasing empirical evidence suggests that global time series forecasting models can achieve better forecasting performance than local ones, there is a research void regarding when and why the global models fail to provide a good forecast. This paper uses anomaly detection algorithms and explainable artificial intelligence (XAI) to answer when and why a forecast should not be trusted. To address this issue, a dashboard was built to inform the user regarding (i) the relevance of the features for that particular forecast, (ii) which training samples most likely influenced the forecast outcome, (iii) why the forecast is considered an outlier, and (iv) provide a range of counterfactual examples to understand how value changes in the feature vector can lead to a different outcome. Moreover, a modular architecture and a methodology were developed to iteratively remove noisy data instances from the train set, to enhance the overall global time series forecasting model performance. Finally, to test the effectiveness of the proposed approach, it was validated on two publicly available real-world datasets.

scholarly journals STABILIZATION OF SENTINEL-1 SAR TIME-SERIES USING CLIMATE AND FOREST STRUCTURE DATA FOR EARLY TROPICAL DEFORESTATION DETECTION

2020 ◽  
Vol V-3-2020 ◽  
pp. 89-96
Author(s):  
J. Doblas ◽  
A. Carneiro ◽  
Y. Shimabukuro ◽  
S. Sant’Anna ◽  
L. Aragão ◽  
...  
Keyword(s):  
Time Series ◽  
Forest Structure ◽  
Detection Algorithm ◽  
Structure Information ◽  
Detection Algorithms ◽  
Alternative State ◽  
Level Data ◽  
Relevant Variables ◽  
Long Time ◽  
Omission Errors

Abstract. In this study we analyse the factors of variability of Sentinel-1 C-band radar backscattering over tropical rainforests, and propose a method to reduce the effects of this variability on deforestation detection algorithms. To do so, we developed a random forest regression model that relates Sentinel-1 gamma nought values with local climatological data and forest structure information. The model was trained using long time-series of 26 relevant variables, sampled over 6 undisturbed tropical forests areas. The resulting model explained 71.64% and 73.28% of the SAR signal variability for VV and VH polarizations, respectively. Once the best model for every polarization was selected, it was used to stabilize extracted pixel-level data of forested and non-deforested areas, which resulted on a 10 to 14% reduction of time-series variability, in terms of standard deviation. Then a statistically robust deforestation detection algorithm was applied to the stabilized time-series. The results show that the proposed method reduced the rate of false positives on both polarizations, especially on VV (from 21% to 2%, α=0.01). Meanwhile, the omission errors increased on both polarizations (from 27% to 37% in VV and from 27% to 33% on VV, α=0.01). The proposed method yielded slightly better results when compared with an alternative state-of-the-art approach (spatial normalization).

Combing empirical matched field processing at IMS station SPITS and convolutional neural networks for calving event detection in Svalbard

2021 ◽  
Author(s):  
Andreas Köhler ◽  
Steffen Mæland
Keyword(s):  
Machine Learning ◽  
Neural Networks ◽  
Time Series ◽  
Convolutional Neural Networks ◽  
Event Detection ◽  
The Arctic ◽  
Beam Power ◽  
Seismic Events ◽  
Tidewater Glaciers ◽  
Seismic Arrays

<p>We combine the empirical matched field (EMF) method and machine learning using Convolutional Neural Networks (CNNs) for calving event detection at the IMS station SPITS and GSN station KBS on the Arctic Archipelago of Svalbard. EMF detection with seismic arrays seeks to identify all signals similar to a single template generated by seismic events in a confined target region. In contrast to master event cross-correlation detectors, the detection statistic is not the waveform similarity, but the array beam power obtained using empirical phase delays (steering parameters) between the array stations. Unlike common delay-and-sum beamforming, the steering parameters do not need to represent a plane wave and are directly computed from the template signal without assuming a particular apparent velocity and back-azimuth. As for all detectors, the false alarms rate depends strongly on the beam power threshold setting and therefore needs appropriate tuning or alternatively post-processing. Here, we combine the EMF detector using a low detection threshold with a post-detection classification step. The classifier uses spectrograms of single-station three-component records and state-of-the-art CNNs pre-trained for image recognition. Spectrograms of three-component seismic data are hereby combined as RGB images. We apply the methodology to detect calving events at tidewater glaciers in the Kongsfjord region in Northwestern Svalbard. The EMF detector uses data of the SPITS array, at about 100 km distance to the glaciers, while the CNN classifier processes data from the single three-component station KBS at 15 km distance using time windows where the event is expected according to the EMF detection. The EMF detector combines templates for the P and for the S wave onsets of a confirmed, large calving event. The CNN spectrogram classifier is trained using classes of confirmed calving signals from four different glaciers in the Kongsfjord region, seismic noise examples, and regional tectonic seismic events. By splitting the data into training and test data set, the CNN classifier yields a recognition rate of 89% on average. This is encouragingly high given the complex nature of calving signals and their visually similar waveforms. Subsequently, we process continuous data of 6 months in 2016 using the EMF-CNN method to produce a time series of glacier calving. About 90% of the confirmed calving signals used for the CNN training are detected by EMF processing, and around 80% are assigned to the correct glacier after CNN classification. Such calving time series allow us to estimate and monitor ice loss at tidewater glaciers which in turn can help to better understand the impact of climate change in Polar regions. Combining the superior detection capability of (less common) seismic arrays at a larger source distance with a powerful machine learning classifier at single three-component stations closer to the source, is a promising approach not only for environmental monitoring, but also for event detection and classification in a CTBTO verification context.</p>

scholarly journals Enhanced Heartbeat Graph for emerging event detection on Twitter using time series networks

2019 ◽  
Vol 136 ◽  
pp. 115-132 ◽  
Author(s):  
Zafar Saeed ◽  
Rabeeh Ayaz Abbasi ◽  
Imran Razzak ◽  
Onaiza Maqbool ◽  
Abida Sadaf ◽  
...  
Keyword(s):  
Time Series ◽  
Event Detection
Sign in / Sign up

Export Citation Format

Share Document