Search

Nonideal switching characteristics of voltage source inverter (VSI) will result in distortions in both phase currents and reference voltages of a permanent magnet synchronous machine (PMSM) drive system, which will also consequently introduce ripples into the output speed and shaft torque, and result in a performance deterioration in the state observation control. To solve this issue, a method of directly extracting and compensating the nonlinear voltage disturbance due to VSI nonlinearity employed in PMSM drives is proposed in this paper. Firstly, the distorted voltage due to VSI nonlinearity is directly extracted via a harmonic separation scheme, which can be easily compensated by a proportional integral controller afterwards. Since both 6th harmonic and DC terms of dq-axis distorted voltages due to VSI nonlinearity are well compensated, the proposed method shows sufficiently good performance under both $i_{d}=0$ and maximum torque per ampere controls ( $i_{d} < 0$ ), as verified on two prototype PMSMs being salient pole and non-salient pole, respectively, both developed for a low-speed rotary actuator. In addition, the application of proposed method to state observation controls such as the model-based sensorless control and the deadbeat predictive current control, is finally investigated. Test results manifest that the proposed method can effectively improve the accuracy of observed rotor position/speed, and the ability of current tracking.

Fast and reliable target recognition of the synthetic aperture radar (SAR) images has been widely used in the fields of the marine monitoring, military reconnaissance and strike all over the world. However, due to the difficulty of the intra-class difference and inter-class similarity of the multiclass targets in the high resolution SAR images, the existing methods are difficult to recognize the targets accurately when facing the spaceborne platforms with the high resource constraints. Therefore, in order to solve the above problems, we propose a novel recognition method based on the convolutional neural network (CNN). Firstly, we propose a lightweight CNN framework which regards densely connected convolutional network (DenseNet) as the baseline. Secondly, we advocate a strong discriminative loss function which efficiently improves the recognition accuracy of the targets in the spaceborne SAR images. Experiments are conducted on the TerraSAR dataset and MSTAR dataset to evaluate the proposed method. The results show that our method performs better than the baseline on the both benchmark datasets.

Hierarchical Neural Networks [e.g., deep neural networks (DNNs)] have recently gained increasing attention for image classification task. Most of the previous works devote to stack multiple levels of feature learning modules with the hope that higher level modules can represent more abstract semantics of the image data. However, training of such hierarchical network is typically cast as a time-consuming non-convex optimization problem, and its effectiveness for image feature representation critically depends on expertise in parameter tuning with various ad hoc tricks. To address these issues, we advocate a biologically-inspired hierarchical neural network. One key philosophy is that the higher-level modules in the network should correct misclassification induced by the lower ones. Given this idea, a sequential stacking strategy of basic feature learning modules is presented. In practical, an efficient layer wise learning and decision aggregation method is applied to boost the hierarchical network, which is different from naive layers cascading and end-to-end finetuning, and we term it as AdaboostNet. Moreover, the basic feature learning module is set as extreme learning machine (ELM), an effective and cheaply-optimized model. Experiments are conducted on benchmark datasets to evaluate our claims. It shows that the proposed network achieves comparable or better image classification accuracy and training efficiency than traditional DNNs and hierarchical ELMs.

The training and running of neural network require large computational space and memory space, which makes it difficult to deploy on a resource-constrained embedded sys-tems. To address this limitation, we introduce a two-stage pipeline: Depth separable, local binary. Our method is divided into two steps. Firstly, a deep separable convolution is applied to each input channel. Then, point-by-point convolution is applied to the feature map obtained by the filter. In the second step, we use local binarization method to initialize the filter corresponding to the input channel into sparse binary. In network training, the sparse binary filter remains fixed and only needs to train convolution of 1×1 size. Our experimental results show that, on the basis of approximate accuracy with the original network, we have reduced the number of convolution parameters by 9x to 10x, and reduced the training time and testing time to by 2x. Our compression method helps to deploy complex neural networks on resource-constrained embedded platform.

Remote sensing technology is widely used in agriculture monitoring because of the advantages of large-area simultaneous observation, low cost and dynamic monitoring of time and space. However, a manual visual interpretation method is often used to extract the information behind the remote sensing images, which is time and labor consuming. Moreover, handcraft features such as texture and structure of crop images are applied to classify crop planting area while these features are not robust. In order to reduce the cost and enhance the classification accuracy, we improved the state-of-the-art image semantic segmentation network SegNet in crop planting area classification which can speed up the convergence and reduce the model size largely with small gains in accuracy. The experiment result shows that the classification accuracy of the improved SegNet is slightly increased compared with SegNet, and the computational cost (FLOPs) of the improved SegNet is much less than SegNet.

Vehicle detection in aerial image is the foundation of some applications, such as the traffic management, parking lot utilization, etc. Recently, universal object detection methods based on the convolutional neural networks have achieved state-of-the-art performances, this is mainly because CNNs can extract more effective features compared with the handcrafted features in early mainstream methods. Extracting effective feature is crucial for vehicle detection in aerial image where the vehicles are small and the background is rather complicated. As a result, these methods based on CNNs have been used to detect the vehicles in aerial image. However, the performance may be poor when directly performing these universal methods. Firstly, these existing methods mostly detect the vehicles with horizontal bounding boxes. But these horizontal boxes do not match the vehicles in aerial image with arbitrary orientations and multiply aspect ratios. As a result, this kind of box would harm the detection accuracy directly. In addition, these methods are mostly computationally expensive, so they are not suitable for the platform with limited computational resources, for example the unmanned aerial vehicle. To address these problems above, by introducing the rotated bounding box regression and a lightweight network into SSD, we propose our vehicle detection networks, VDetor. Specifically, we use rotated bounding box rather than horizontal one to match the vehicle well and use a lightweight network, PeleeNet, as the backbone network of SSD to speed up the inference. Experiments on VEDAI dataset illustrate that our model performs better than SSD in terms of the detection accuracy and detection speed.

Load aggregator is one of the most effective means of integrating demand-side resources into the power system. However, with the complexity of the load aggregator members, especially with the introduction of "peaking devices" represented by batteries, the traditional method based on energy produced can not evaluate the role of participants properly. Aiming at this problem, this paper presents a distribution-allocation bi-level optimization model based on the improved Shapley value. The upper layer takes the relative income balance of each member as the optimization target, while the lower layer treats wind power, battery and interruptible load as the member of load aggregator. The power flow of load aggregator is optimized. In this paper, the operability and rationality of the proposed improved profit allocation method are proved by case study. It is also verified that the load aggregator is an effective way to integrate demand-side resources into the power system.

Measuring the impact of the user side on the carbon emission of the power system scientifically and rationally can promote the enthusiasm of power users participating in low-carbon scheduling, and provide guidance for the formulation of demand response policies, low-carbon scheduling and low-carbon planning. In this paper, the physical and mathematical significance of locational marginal emission (LME) is analyzed, and the difference between LME and node carbon potential is clarified. Then, a method for calculating the probability distribution of LME based on probabilistic power flow and the application model of LME in demand side management are proposed. An example is given to verify the significance of LME based on probabilistic power flow in improving the potential of reducing emissions on the user side and reducing the carbon emission in power systems.

This paper proposes a fast Galerkin projection method for model order reduction of large-scale dynamic systems. We first define a novel dynamic optimization problem in which the projection modes are chosen to minimize the difference between the output from the original model and the output from the Galerkin reduced model. Next, we approximate this dynamic optimization problem by a static problem that can be solved efficiently using standard nonlinear optimization algorithms. The solution of the static problem defines a set of suboptimal projection modes that (approximately) maximize the accuracy of the corresponding reduced model. This is different to the well-known proper orthogonal decomposition and Galerkin projection method (POD-Galerkin method), in which the accuracy of the orthogonal state projection, rather than the accuracy of the reduced model's output, is maximized. We compare the new method with the POD-Galerkin method for a one-dimensional heat transfer system.

With the increasing spatial and temporal resolutions of acquired remote sensing (RS) images, effective image compression is becoming more and more important. RS image compression technologies have been extensively studied in the past a few decades, and various algorithms have been developed accordingly. In this paper, we provide an overview of practically deployed RS image compression approaches, including predictive coding and transform coding approaches that have been adopted in different satellite systems. In addition, some newly derived RS image compression methods are discussed, with highlights on the new trends of the on-going design and developments of RS image compression.