Transactions of the Indian National Academy of Engineering, 2021
Computing the similarity between time series data is one of the core challenges frequently occurr... more Computing the similarity between time series data is one of the core challenges frequently occurring in time series analysis across various disciplines. Particularly, in building energy management, there are many applications that need to cluster the time series with simultaneous rise, peak, similar patterns, pattern-average allowing small amplitude shift and small phase shift as one group. Commonly used distance measures like Euclidean and DTW being sensitive to phase shift or amplitude shift, are unable to identify similarity between two time series with simultaneous rise, peak and similar patterns. To address this problem, we propose a novel time series similarity measure, sub-sequence based DTW, SeqDTW based on time series segmentation approach. SeqDTW attempts to find best match of each subsequence in the segmented time series. Overall distance is the weighted aggregate of distance between the matching subsequences. In addition to existing monotonicity in DTW warping path, SeqDTW reduces the number of computations across the distance matrix and avoids singularity. An exhaustive experiment shows the superiority of the proposed method in finding time series having simultaneous rise, peak and similar patterns.
Smart grids infrastructure is rapidly adopting the recent technology to optimize the power genera... more Smart grids infrastructure is rapidly adopting the recent technology to optimize the power generation and energy saving. The load forecasting in smart grids has been one such technology integration and accurate load forecasting models has been a challenge. With the advent of advanced infrastructure, huge data is being generated at different time frequencies, that can be used to build accurate load forecasting models. Focusing on the state-of-the-art machine learning techniques, in this work, we propose a load forecasting model of stacked dilated convolutional layers. The dilations efficiently captures the local trend and seasonality from the time series for future predictions. Proposed model has been trained on multiple time series data with varying frequencies. Results show that the proposed model is an improvement to the existing state-of-the-art. Keywords Time series data • Smart grids • Deep neural networks • Dilated convolution • Load forecasting This article is part of the topical collection "Computational Biology and Biomedical Informatics" guest edited by Dhruba Kr Bhattacharyya, Sushmita Mitra and Jugal Kr Kalita.
Proceedings of the 7th ACM IKDD CoDS and 25th COMAD, 2020
Focusing on the objectives of Demand Side Management (DSM), we propose a novel time series distan... more Focusing on the objectives of Demand Side Management (DSM), we propose a novel time series distance metric that better capture the information related to similar peaks/off-peaks. The proposed metric uses autocorrelation based segmentation and similar segment identification for computation of overall distance. Experiment shows the proposed distance advances the state-of-the-art.
Demand side management of Energy in smart buildings often needs as a prerequisite (or benefits fr... more Demand side management of Energy in smart buildings often needs as a prerequisite (or benefits from) future appliance usage schedules of the occupants. In this work, a datadriven appliance usage prediction framework is proposed. The dependence of consumption trends on recent appliance usage history is explored through the use of temporally sensitive feature representations. Multi-label classification architectures, with appliance level adaptive thresholding are used as the predictive models. A data parallel distributed architecture is explored for our machine learning models to feasibly analyze larger data volumes generated by a Smart Building. The framework is extensively tested on the publicly available GREEND dataset. Appliance usage patterns are obtained at multiple temporal resolutions with encouraging results with lower training times demonstrated by our models on the distributed architecture.
Integration of renewable sources into energy grids has reduced carbon emission, but their intermi... more Integration of renewable sources into energy grids has reduced carbon emission, but their intermittent nature is of major concern to the utilities. In order to provide an uninterrupted energy supply, a prior idea about the total possible electricity consumption of the consumers is a necessity. In this paper, we have introduced a deep learning based load forecasting model designed using dilated causal convolutional layers. The model can efficiently capture trends and multi-seasonality from historic load data. Proposed model gives encouraging results when tested on synthetic and real life time series datasets.
Transactions of the Indian National Academy of Engineering, 2021
Computing the similarity between time series data is one of the core challenges frequently occurr... more Computing the similarity between time series data is one of the core challenges frequently occurring in time series analysis across various disciplines. Particularly, in building energy management, there are many applications that need to cluster the time series with simultaneous rise, peak, similar patterns, pattern-average allowing small amplitude shift and small phase shift as one group. Commonly used distance measures like Euclidean and DTW being sensitive to phase shift or amplitude shift, are unable to identify similarity between two time series with simultaneous rise, peak and similar patterns. To address this problem, we propose a novel time series similarity measure, sub-sequence based DTW, SeqDTW based on time series segmentation approach. SeqDTW attempts to find best match of each subsequence in the segmented time series. Overall distance is the weighted aggregate of distance between the matching subsequences. In addition to existing monotonicity in DTW warping path, SeqDTW reduces the number of computations across the distance matrix and avoids singularity. An exhaustive experiment shows the superiority of the proposed method in finding time series having simultaneous rise, peak and similar patterns.
Smart grids infrastructure is rapidly adopting the recent technology to optimize the power genera... more Smart grids infrastructure is rapidly adopting the recent technology to optimize the power generation and energy saving. The load forecasting in smart grids has been one such technology integration and accurate load forecasting models has been a challenge. With the advent of advanced infrastructure, huge data is being generated at different time frequencies, that can be used to build accurate load forecasting models. Focusing on the state-of-the-art machine learning techniques, in this work, we propose a load forecasting model of stacked dilated convolutional layers. The dilations efficiently captures the local trend and seasonality from the time series for future predictions. Proposed model has been trained on multiple time series data with varying frequencies. Results show that the proposed model is an improvement to the existing state-of-the-art. Keywords Time series data • Smart grids • Deep neural networks • Dilated convolution • Load forecasting This article is part of the topical collection "Computational Biology and Biomedical Informatics" guest edited by Dhruba Kr Bhattacharyya, Sushmita Mitra and Jugal Kr Kalita.
Proceedings of the 7th ACM IKDD CoDS and 25th COMAD, 2020
Focusing on the objectives of Demand Side Management (DSM), we propose a novel time series distan... more Focusing on the objectives of Demand Side Management (DSM), we propose a novel time series distance metric that better capture the information related to similar peaks/off-peaks. The proposed metric uses autocorrelation based segmentation and similar segment identification for computation of overall distance. Experiment shows the proposed distance advances the state-of-the-art.
Demand side management of Energy in smart buildings often needs as a prerequisite (or benefits fr... more Demand side management of Energy in smart buildings often needs as a prerequisite (or benefits from) future appliance usage schedules of the occupants. In this work, a datadriven appliance usage prediction framework is proposed. The dependence of consumption trends on recent appliance usage history is explored through the use of temporally sensitive feature representations. Multi-label classification architectures, with appliance level adaptive thresholding are used as the predictive models. A data parallel distributed architecture is explored for our machine learning models to feasibly analyze larger data volumes generated by a Smart Building. The framework is extensively tested on the publicly available GREEND dataset. Appliance usage patterns are obtained at multiple temporal resolutions with encouraging results with lower training times demonstrated by our models on the distributed architecture.
Integration of renewable sources into energy grids has reduced carbon emission, but their intermi... more Integration of renewable sources into energy grids has reduced carbon emission, but their intermittent nature is of major concern to the utilities. In order to provide an uninterrupted energy supply, a prior idea about the total possible electricity consumption of the consumers is a necessity. In this paper, we have introduced a deep learning based load forecasting model designed using dilated causal convolutional layers. The model can efficiently capture trends and multi-seasonality from historic load data. Proposed model gives encouraging results when tested on synthetic and real life time series datasets.
Uploads
Papers by Kakuli Mishra