Explore the vast range of titles available.

A novel state estimation algorithm based on the parameters of a self-learning unscented Kalman filter (UKF) with a model parameter identification method based on a collaborative optimization mechanism is proposed in this paper. This algorithm can realize the dynamic self-learning and self-adjustment of the parameters in the UKF algorithm and the automatic optimization setting Sigma points without human participation. In addition, the multi-algorithm collaborative optimization mechanism unifies a variety of algorithms, so that the identification method has the advantages of member algorithms while avoiding the disadvantages of them. We apply the combination algorithm proposed in this paper for state of charge (SoC) estimation of power batteries and compare it with other model parameter identification algorithms and SoC estimation methods. The results showed that the proposed algorithm outperformed the other model parameter identification algorithms in terms of estimation accuracy and robustness.

With the advent of the Internet of Everything, the proliferation of data has put a huge burden on data centers and network bandwidth. To ease the pressure on data centers, edge computing, a new computing paradigm, is gradually gaining attention. Meanwhile, artificial intelligence services based on deep learning are also thriving. However, such intelligent services are usually deployed in data centers, which cause high latency. The combination of edge computing and artificial intelligence provides an effective solution to this problem. This new intelligence paradigm is called edge intelligence. In this paper, we focus on edge training and edge inference, the prior training models using local data at the resource-constrained edge devices. The latter deploying models at the edge devices through model compression and inference acceleration. This paper provides a comprehensive survey of existing architectures, technologies, frameworks and implementations in these two areas, and discusses existing challenges, possible solutions and future directions. We believe that this survey will make more researchers aware of edge intelligence.

Given the difficulty of accurate online detection for massive data collecting real-timely in a strong noise environment during the complex geological mineral grade analysis process, an order self-learning ARHMM (Autoregressive Hidden Markov Model) algorithm is proposed to carry out online outlier detection in the geological mineral grade analysis process. The algorithm utilizes AR model to fit the time series obtained from “Online x - ray Fluorescent Mineral Analyzer” and makes use of HMM as a basic detection tool, which can avoid the deficiency of presetting the threshold in traditional detection methods. The structure of traditional BDT (Brockwell-Dahlhaus-Trindade) algorithm is improved to be a double iterative structure in which iterative calculation from both time and order is applied respectively to update parameters of ARHMM online. With the purpose of reducing the influence of outlier on parameter update of ARHMM, the strategies of detection-before-update and detection-based-update are adopted, which also improve the robustness of the algorithm. Subsequent simulation by model data and practical application verify the accuracy, robustness, and property of online detection of the algorithm. According to the result, it is obvious that new algorithm proposed in this paper is more suitable for outlier detection of mineral grade analysis data in geology and mineral processing.

To realize a fast and high-precision online state-of-health (SOH) estimation of lithium-ion (Li-Ion) battery, this article proposes a novel SOH estimation method. This method consists of a new SOH model and parameters identification method based on an improved genetic algorithm (Improved-GA). The new SOH model combines the equivalent circuit model (ECM) and the data-driven model. The advantages lie in keeping the physical meaning of the ECM while improving its dynamic characteristics and accuracy. The improved-GA can effectively avoid falling into a local optimal problem and improve the convergence speed and search accuracy. So the advantages of the SOH estimation method proposed in this article are that it only relies on battery management systems (BMS) monitoring data and removes many assumptions in some other traditional ECM-based SOH estimation methods, so it is closer to the actual needs for electric vehicle (EV). By comparing with the traditional ECM-based SOH estimation method, the algorithm proposed in this article has higher accuracy, fewer identification parameters, and lower computational complexity.

The success of deep learning has brought breakthroughs in many fields. However, the increased performance of deep learning models is often accompanied by an increase in their depth and width, which conflicts with the storage, energy consumption, and computational power of edge devices. Knowledge distillation, as an effective model compression method, can transfer knowledge from complex teacher models to student models. Self-distillation is a special type of knowledge distillation, which does not to require a pre-trained teacher model. However, existing self-distillation methods rarely consider how to effectively use the early features of the model. Furthermore, most self-distillation methods use features from the deepest layers of the network to guide the training of the branches of the network, which we find is not the optimal choice. In this paper, we found that the feature maps obtained by early feature fusion do not serve as a good teacher to guide their own training. Based on this, we propose a selective feature fusion module and further obtain a new self-distillation method, knowledge fusion distillation. Extensive experiments on three datasets have demonstrated that our method has comparable performance to state-of-the-art distillation methods. In addition, the performance of the network can be further enhanced when fused features are integrated into the network.

Aiming at the requirements of high accuracy, lightweight and real-time performance of the panoptic driving perception system, this paper proposes an efficient multi-task network (YOLOMH). The network uses a shared encoder and three independent decoding heads to simultaneously complete the three major panoptic driving perception tasks of traffic object detection, road drivable area segmentation and road lane segmentation. Thanks to our innovative design of the YOLOMH network structure: first, we design an appropriate information input structure based on the different information requirements between different tasks, and secondly, we propose a Hybrid Deep Atrous Spatial Pyramid Pooling module to efficiently complete the feature fusion work of the neck network, and finally effective approaches such as Anchor-free detection head and Depthwise Separable Convolution are introduced into the network, making the network more efficient while being lightweight. Experimental results show that our model achieves competitive results in both accuracy and speed on the challenging BDD100K dataset, especially in terms of inference speed, The model’s inference speed on NVIDIA TESLA V100 is as high as 107 Frames Per Second (FPS), far exceeding the 49 FPS of the YOLOP network under the same experimental settings. This well meets the requirements of autonomous vehicles for high system accuracy and low latency.

Knowledge distillation is a commonly used method for model compression that has been widely utilized in various computer vision tasks. Many efforts have utilized attention mechanisms to guide the student networks during training, encouraging them to mimic the important features of the teacher. However, most of these efforts use either the channel attention map or the spatial attention map to guide the student, ignoring the importance of positional features. In this paper, we propose a new distillation framework transferring global and positional features (DFGPD), which consists of three parts: global and positional distillation, a generic teacher framework and a two-stage distillation method. DFGPD takes positional information into consideration for a more effective distillation process. We conduct extensive comparison experiments, ablation studies, and sensitivity studies to demonstrate the effectiveness and stability of DFGPD. Our results show that (1) DFGPD achieves comparable or even better performance compared to state-of-the-art methods; (2) DFGPD can alleviate the bigger-models-not-always-better-teachers issue to a certain extent.

The promotion and application of model-based systems engineering (MBSE) suffer from the lack of effective sharing of research and design (R&D) resources among enterprises in the networked collaborative design environment. This paper establishes a cross-enterprise R&D resource sharing incentive evolutionary game model, which considers multiple factors about resource sharers, sharing process and shared resource value information. The model analyses how various factors influence the evolution trend of sharing behavior. In addition, in order to realize the dynamic multidimensional evaluation and feedback of shared R&D resources and promote the positive development of resource sharing, this paper establishes a new dynamic double-objective resource evaluation model. The digital information resource in R&D resources is taken as an example to carry out experimental simulation and verification. Compared with the traditional evaluation methods, the new dynamic double-objective evaluation model established in this paper owns better performance on improving the cross-enterprise R&D resources collaborative sharing and further promoting the deep integration between R&D resources and R&D process under the networked collaborative design and MBSE development mode than others.

In order to solve the problem that the model-based State of Charge (SOC) estimation method is too dependent on the model parameters in the SOC estimation of electric vehicles, an improved genetic algorithm is proposed in this paper. The method has the advantages of being able to quickly determine the search range, reducing the probability of falling into local optimum, and having high recognition accuracy. Then we can realize online dynamic identification of power battery model parameters and improve the accuracy of model parameter identification. In addition, considering the complex application environment and operating conditions of electric vehicles, an SOC estimation method based on improved genetic algorithm and unscented particle filter (improved GA-UPF) is proposed. And we compare the improved GA-UPF algorithm with the least square unscented particle filter (LS-UPF) and improved GA unscented Kalman filter (improved GA-UKF) algorithm. The comparison results show that the improved GA-UPF algorithm proposed in this paper has higher estimation accuracy and better stability. It also reflects the practicability and accuracy of the improved GA parameter identification algorithm proposed in this paper.

The success of deep learning has brought breakthroughs in many fields. However, the increased performance of deep learning models is often accompanied by an increase in their depth and width, which conflicts with the storage, energy consumption, and computational power of edge devices. Knowledge distillation, as an effective model compression method, can transfer knowledge from complex teacher models to student models. Self-distillation is a special type of knowledge distillation, which does not to require a pre-trained teacher model. However, existing self-distillation methods rarely consider how to effectively use the early features of the model. Furthermore, most self-distillation methods use features from the deepest layers of the network to guide the training of the branches of the network, which we find is not the optimal choice. In this paper, we found that the feature maps obtained by early feature fusion do not serve as a good teacher to guide their own training. Based on this, we propose a selective feature fusion module and further obtain a new self-distillation method, knowledge fusion distillation. Extensive experiments on three datasets have demonstrated that our method has comparable performance to state-of-the-art distillation methods. In addition, the performance of the network can be further enhanced when fused features are integrated into the network.