Geographical dispersion and output power fluctuations are the major barriers to efficient utilization and grid connection of building attached photovoltaic (BAPV). To eliminate these negative factors, a reliable energy management system and an accurate power forecasting model are necessary. In this paper, we first design an energy management micro-grid based on the energy Internet, which aims to tackle the problems faced by the grid-connected BAPV through the effective dual-flow management of energy and information. In the context of the proposed micro-grid, we propose a deep power forecasting model that employs a convolutional neural network to find the nonlinear relationship between meteorological information and BAPV power, while the data fed to the model are obtained through the 2-D Fourier transform of meteorological data. We evaluate the proposed model based on real-world meteorological and power data sets. Numerical results highlight the superiority of our forecasting model in terms of accuracy and reliability.
The energy Internet (EI) is an important infrastructure for effectively utilizing and intelligently managing renewable energy sources (RES). In this paper, we study the architecture design of the EI under the backdrop of large-scale RES grid connection and the efficient forecasting and optimal utilization of energy. The contribution of this paper is threefold. First, we design a hierarchical integration architecture for the EI and attempt to solve the issues of energy and information management that stem from large-scale RES grid connection. Second, we propose a novel energy forecasting scheme that significantly reduces the amount of effort and ensures the accuracy of formulating the energy forecasting as an instance of the matrix completion issue. Third, we take electric vehicle charging as a typical case and propose the use of reinforcement learning to achieve optimal energy delivery. An experimental evaluation of real-world data sets validates the expectations of the study and highlights the superiorities of our proposed approaches.
In this paper, we analyze the characteristics of the load forecasting task in the Energy Internet context and the deficiencies of existing methods and then propose a data driven approach for one-hour-ahead load forecasting based on the deep learning paradigm. The proposed scheme involves three aspects. First, we formulate a historical load matrix (HLM) with spatiotemporal correlation combined with the EI scenario and then create a three-dimensional historical load tensor (HLT) that contains the HLMs for multiple consecutive time points before the forecasted hour. Second, we preprocess the HLT leveraging a novel low rank decomposition algorithm and different load gradients, aiming to provide a forecasting model with richer input data. Third, we develop a deep forecasting framework (called the 3D CNN-GRU) featuring a feature learning module followed by a regression module, in which the 3D convolutional neural network (3D CNN) is used to extract the desired feature sequences with time attributes, while the gated recurrent unit (GRU) is responsible for mapping the sequences to the forecast values. By feeding the corresponding load label into the 3D CNN-GRU, our proposed scheme can carry out forecasting tasks for any zone covered by the HLM. The results of self-evaluation and a comparison with several state-of-the-art methods demonstrate the superiority of the proposed scheme.
While the sparse methods for 1-D direction-of-arrival (DOA) estimation are extensively studied in literature, the research for 2-D DOA estimation is rare. The main reason is that, for utilizing the on-grid or off-grid sparse methods, the 2-D continuous angle space should have to be discretized, which, however, may bring unacceptable computations due to the high dimensionality of the angle space. Hence, incorporating the gridless sparse methods which require no discretization into the 2-D DOA estimation is essential. In this paper, we propose a gridless 2-D DOA estimation method based on the low-rank matrix reconstruction and the Vandermonde decomposition theorem. We first reconstruct the covariance matrix with certain structure (i.e., low-rank, Toeplitz, and positive semidefinite), and then, retrieve the DOAs by using the Vandermonde decomposition theorem. We also present a theorem to guarantee that the true DOAs can be exactly recovered in certain condition. A faster algorithmic implementation is then given by deriving the dual problem of the original one. Our proposed method is applicable for both the uniform rectangular arrays and the sparse rectangular arrays. Extensive simulations are provided to evaluate its estimation performance and the adaptability to various array geometries.
With the rapid development of location-based service applications in recent years, indoor localization based on Wi-Fi channel state information (CSI) has attracted extensive attention. Although spatial variations are explicitly reflected in CSI measurements, representation differences caused by small contextual changes tend to be submerged by overall multipath fluctuations, especially in device-free localization. Most existing model-based solutions either underutilize the spatial and time-frequency information carried by CSI, or the employed models fail to capture small representational changes in CSI well, which makes them struggle to get satisfactory expectations. To address these problems, this paper proposes a progressive device-free localization scheme using multidimensional CSI features classification, named ProLoc. Different from previous works that directly extract the features of raw measurements, ProLoc constructs a model input containing multi-category information of CSI to perform better location inference. First, a time-frequency correlation matrix with low rank is created based on the CSI amplitude, and a 3-order CSI tensor is planned combined with spatial diversity. Second, using a novel low-rank matrix factorization algorithm and the time-frequency gradients of CSI, a 4-order tensor containing multidimensional information is generated as input data. Finally, employing 3D convolutional neural networks and gated recurrent unit, we build a progressive dual-model system to realize the mapping from CSI to target locations in various spatial contexts. Extensive self-evaluations and comparisons with several state-of-the-art methods highlight the superiority of the proposed ProLoc.
The popularisation of fingerprinting localisation technology has been hindered because of two major hurdles: (i) the accuracy bottleneck caused by unreliable location fingerprints and (ii) the huge effort required to construct a fingerprints database (or radio map) for the targeted area. To tackle the two problems, the authors propose an effective solution in this work. First, they exploit channel state information, which is a parameter depicting the frequency response of each subchannel, to design the location fingerprint, striving to eliminate the interferences of the complex indoor environment. Second, they propose an efficient construction scheme leveraging the matrix completion theory to improve the calibration efficiency, and employ a Bayes rule‐based fingerprint matching method to implement location estimation. Finally, they evaluate the authors’ localisation system in two typical scenarios, and the numerical results show that the proposal ensures superior performance while reducing the workload significantly.