Showing papers by "Wenjiang Ji published in 2020"

Motion trajectory prediction based on a CNN-LSTM sequential model

Abstract: Accurate monitoring the surrounding environment is an important research direction in the field of unmanned systems such as bio-robotics, and has attracted much research attention in recent years. The trajectories of surrounding vehicles should be predicted accurately in space and time to realize active defense and running safety of an unmanned system. However, there is uncertainty and uncontrollability in the process of trajectory prediction of surrounding obstacles. In this study, we propose a trajectory prediction method based on a sequential model, that fuses two neural networks of a convolutional neural network (CNN) and a long short-term memory network (LSTM). First, a box plot is used to detect and eliminate abnormal values of vehicle trajectories, and valid trajectory data are obtained. Second, the trajectories of surrounding vehicles are predicted by merging the characteristics of CNN space expansion and LSTM time expansion; the hyper-parameters of the model are optimized according to a grid search algorithm, which satisfies the double-precision prediction requirement in space and time. Finally, data from next generation simulation (NGSIM) and Creteil roundabout in France are taken as test cases; the correctness and rationality of the method are verified by prediction error indicators. Experimental results demonstrate that the proposed CNN-LSTM method is more accurate and features a shorter time cost, which meets the prediction requirements and provides an effective method for the safe operation of unmanned systems.

From Hardware to Operating System: A Static Measurement Method of Android System Based on TrustZone

Abstract: Android system has been one of the main targets of hacker attacks for a long time. At present, it is faced with security risks such as privilege escalation attacks, image tampering, and malicious programs. In view of the above risks, the current detection of the application layer can no longer guarantee the security of the Android system. The security of mobile terminals needs to be fully protected from the bottom to the top, and the consistency test of the hardware system is realized from the hardware layer of the terminal. However, there is not a complete set of security measures to ensure the reliability and integrity of the Android system at present. Therefore, from the perspective of trusted computing, this paper proposes and implements a trusted static measurement method of the Android system based on TrustZone to protect the integrity of the system layer and provide a trusted underlying environment for the detection of the Android application layer. This paper analyzes from two aspects of security and efficiency. The experimental results show that this method can detect the Android system layer privilege escalation attack and discover the rootkit that breaks the integrity of the Android kernel in time during the startup process, and the performance loss of this method is within the acceptable range.

High-Speed Train Emergency Brake Modeling and Online Identification of Time-Varying Parameters

Abstract: By analyzing the mechanism of pure air emergency brake for high-speed train, the discrete emergency brake model is established. Aiming at the problem that time-varying hidden parameters cannot be observed directly, the sliding window-based expectation maximization is proposed, and the unobserved time-varying brake parameters are identified. Firstly, the position and size of the sliding window are selected; then, the sliding window-based expectation maximization is used for brake parameter identification; finally, combined with the gradient optimization, the optimal identifications of emergency brake parameters are obtained. The simulation results show that the brake parameters can be identified quickly and accurately by the proposed method. Under uniform noise, the identification errors of friction coefficient and braking ratio are ±0.0068 and ±0.0349, respectively, and the maximum relative errors between the identifications and true values are 2.4807% and 1.3154%, respectively, which can meet the actual requirements of the brake system. The effectiveness and practicability of the proposed model and method are verified.

Fault Diagnosis of Rolling Bearing Based on Improved LeNet-5 CNN

Abstract: To solve the problem of fault diagnosis of rolling bearing caused by large amount of data and difficulties of processing those data on to bearing set, based on Convolution Neural Network, a new method of data processing is proposed in this paper. With this method, one-dimensional time domain signal can be transformed into two-dimensional images, which is more suitable for Convolutional Neural Network processing. Meanwhile, the traditional machine learning method has the disadvantage of low robustness and low recognition rate with noise interference. Therefore, based on the feature extraction of Convolution Neural Network, in this paper we proposed an improved LeNet-5 Convolution Neural Network model, that is, adding a convolution layer and a pooling layer to the classic LeNet-5 model. The hidden layer features are extracted by using the trainable convolution kernel, while the extracted implicit features are reduced by the pooling layer, the Softmax classifier is used for classification and recognition of rolling bearing faults. In this paper we verified the effectiveness of the improved LeNet-5 model for fault diagnosis of rolling bearing by using the rolling bearing data to train the classic LeNet-5 model and the improved model.

Knowledge graph storage and verification method based on blockchain and IPFS

Abstract: The invention discloses a knowledge graph storage and verification method based on a blockchain and an IPFS. The method specifically comprises the steps that firstly, a process file of the knowledge graph is exported from a graph database, a file hash value is calculated, the process file is stored in an IPFS inter-satellite file system, a file index hash value returned by the IPFS is compared with the previously calculated hash value, and if the file index hash value is equal to the previously calculated hash value, the file index hash value is stored in a block chain, and safety of the process file is guaranteed. The file index hash value is stored into the blockchain through processes of data packaging, signing, block packaging, block broadcasting and the like, querying the file index hash value in the blockchain, comparing the file index hash value with a file index hash value obtained from an IPFS before, if the file index hash value is the same as the file index hash value, entering a file processing module, and otherwise, entering a file warning module. Through combination of the blockchain and the IPFS, reliable and rapid storage of the knowledge graph can be realized, andthe security and traceability of knowledge graph process files are ensured.

Analyzing the Collision Probability of Autonomous Vehicle at Crossroad

Abstract: It is essential to assess the autonomous vehicle operation safety during driving, which can avoid or reduce the collision risk by evaluating the drive safety. In this paper, a collision risk assessment algorithm is proposed, that is quantifies the auto-drive vehicle collision risk by the Time to Collision (TTC) frequency. Firstly, the Long Short Time Memory network (LSTM) is used to predict the surrounding vehicle trajectory; Moreover, the collision point between the auto-drive vehicle and the surrounding vehicle is determined, and the frequency distribution result of TTC is calculated by the Monte Carlo simulation method; Finally, the running speed & hazard probability is obtained by changing the running speed of auto-drive vehicle, and the running speed & safety probability is obtained further. It can be seen from the result that the proposed method can provide an effective evidence for decision-making layer of auto-drive vehicle, improve the running safety of vehicle, and reduce the operation risk of auto-drive vehicle.

University computer basic exercise recommendation method based on knowledge graph

Abstract: The invention discloses a university computer basic exercise recommendation method based on a knowledge graph. Based on a knowledge graph constructed by college computer basic knowledge points, a graph team detection algorithm is used for analyzing relevance among knowledge points from a relation structure among the knowledge points, so that the knowledge points with relatively high relevance withwrong knowledge points of students are recommended to the students for learning; exercise semantic features are extracted through Chinese word segmentation to construct a word2vec model, the similarity of wrong questions of students and other exercises in content is calculated by using an RWMD method, and exercises with high similarity are selected for recommendation. Corresponding knowledge points and exercises are recommended to students by analyzing wrong questions of the students, the students can be helped to sweep knowledge blind spots as soon as possible, the college computer base course can be better mastered, and meanwhile the pressure of course teachers is relieved.

Classification and Recognition of Unknown Network Protocol Characteristics.

Abstract: In recent years, unscrupulous hacker attacks have led to the information leakage of enterprise and individual network users, which makes the network security issue unprecedented concerned. Botnet and dark network, which use C & C channel of unknown protocol format to communicate, are the important parts. With the development of wireless mobile networks technology, this problem becomes more prominent. Classifying and identifying the unknown protocol features can help us to judge and predict the unknown attack behavior in the Internet of things environment, so as to protect the network security. Firstly, this paper compares the protocol features to be detected with the existing protocol features in the feature base through the vectorization operation of protocol features, selects the feature set with high recognition rate, and judges the similarity between protocols. The extracted composite features are digitized to generate 0-1 matrix, then Principal Component Analysis (PCA) dimension reduction is processed, and finally clustering analysis is carried out. A Clique to Protocol Feature Vectorization (CPFV) algorithm is designed to improve the efficiency of protocol clustering and finally generate a new protocol format. The experimental results show that compared with the traditional Clique and BIRCH algorithms, the proposed optimization algorithm improves the accuracy by 20% and the stability by 15%. It can cluster and identify unknown protocols accurately and quickly.

Network Flow Analytics: Multi-Class Classification of DDoS Attacks Based on OKNN

Abstract: in the past decade DDoS attacks detection solutions have been one hot area in the research of cyberspace. Different techniques including machine learning and recent complex deep learning models were used and improved. However, only a few have used real network data and multiclass classification. In this paper, we challenged an Optimized K-Nearest Neighbor (OKNN) on a recent public dataset of the real network containing labeled classes of normal network flow and DDoS attacks. While performing minimum preprocessing to keep data original, OKNN with a tune of hyper-parameters such as; n_neighbors, metric, weights, n_jobs, has identified a normal traffic flow and DDoS attacks with high accuracy. The results of the experiment show that our model would perform better than its counterparts if only they are trained in the same conditions.

Weighted Sparse Graph Non-negative Matrix Factorization based on L 21 norm

Abstract: Dimension reduction is widely concerned because of the rapid development of all walks of life, which leads to the exponential growth of data dimension. As a basic method of dimension reduction, non-negative matrix factorization is easily affected by noise, but its improved algorithm L 21 NMF is not sensitive to noise. Therefore, the paper proposes a weighted sparse graph non-negative matrix factorization based on L 21 norm(LSL 21 NMF). In this method, L 21 norm is used as the measure criterion, the non-negative matrix is decomposed into the summation of one error matrix and the product of two matrices non-negative, the weight sparse graph is applied to the regularization term to preserved geometrical structure of data. An efficient iterative approach is developed to solve the optimization problem of LSL 21 NMF. On the standard noisy public dataset,the experimental results show that the method is more effective and robust than other methods.

A Framework for Audit Analysis of Heterogeneous Logs from PaaS

Abstract: PaaS platform generates a large number of heterogeneous logs in the process of providing services. To analyze these heterogeneous logs together requires a unified analysis framework. We designed a framework for audit analysis in which the raw logs, the annotated logs, and the statistics generated from the logs are combined into a unified log for processing.

Tramcar conflict area evaluation method based on multi-tramcar advancing environment

Abstract: The invention discloses a tramcar conflict area evaluation method based on a multi-vehicle advancing environment. The method comprises the steps that: a used vehicle movement data set is preprocessed,a data set of the operation states of a target vehicle and surrounding vehicles is obtained, and then a tramcar observation area and an early warning area are divided; a data set obtained through processing is visually observed through multi-vehicle motion state drawing, the lane changing intention of the target vehicle is analyzed, a track prediction model of vehicles around a tramcar is constructed based on a long-term and short-term memory network, and through a collision conflict area constructed through a determined track of the tramcar and a predicted track of the target vehicle, the collision danger degree of the target vehicle entering the early warning area and the traveling tramcar is evaluated according to the collision grade division condition. The problem that in the prior art, a tramcar is prone to colliding with a third-party vehicle in a mixed road right mode, and consequently traffic accidents are caused is solved.

Deep Learning as a Service Based on Encrypted Data

Abstract: With the widespread promotion of 4G/5G wireless network and the rapid development of Internet of Things (IoT), more and more enterprises begin to make use of their powerful computing power and technology to provide pre-trained deep neural networks for ordinary users to help them complete classification, regression, image recognition, NLP and other services. This not only brings convenience to people, but also easily causes the leakage of users' private data. In this paper, we combine deep learning with homomorphic encryption algorithm and design a deep learning network model based on secure Multi-party computing (MPC). In the whole process, we realize that the cloud only owns the encryption samples of users, and users do not own any parameters or structural information related to the model. In the experimental part, we input the encrypted Mnist and Cifar-l0 datasets into the model for testing, and the results show that the classification accuracy rate of the encrypted Mnist can reach 99.21%, which is very close to the result under plaintext. The classification accuracy rate of encrypted cifar-l0 can reach 91.35%, slightly lower than the test result in plaintext, and better than the existing deep learning network model that can realize data privacy protection.