Search

Deep learning has become a research hotspot in the field of network intrusion detection. In order to further improve the detection accuracy and performance, we proposed an intrusion detection model based on improved deep belief network (DBN). Traditional neural network training methods, like Back Propagation (BP), start to train a model with preset parameters such as the randomly initialized weights and thresholds, which may bring some issues, e.g., attracting the model to the local optimal solutions, or requiring a long training period. We use the Kernel-based Extreme Learning Machine (KELM) with the supervised learning ability to replace the BP algorithm in DBN in a bid to ameliorate the situation. Considering the problem of poor classification performance usually caused by randomly initializing kernel parameters with KELM, an enhanced grey wolf optimizer (EGWO) is designed to optimize the parameters of KELM. In order to improve the search ability and optimization ability of the traditional grey wolf optimizer algorithm, a novel optimization strategy combining the inner and outer hunting is introduced. Experiments on KDDCup99, NSL-KDD, UNSW-NB15 and CICIDS2017 datasets show that the proposed DBN-EGWO-KELM algorithm has greater advantages in terms of its accuracy, precision, true positive rate, false positive rate and other evaluation indices compared with BP, RBF, SVM, KELM, LIBSVM, CNN, DBN-KELM and other intrusion detection models, and can effectively meet the requirements of intrusion detection of complex networks.

Pneumatic soft robots show great potential application value in the field of hand rehabilitation. This paper presents a novel pneumatic soft robotic glove for hand rehabilitation with a portable pneumatic box. The glove uses positive-negative pneumatic actuator made of bellows, which is very light (only 149g), easy to wear, comfortable, customizable, safe, easy to make, and low-cost. The matching portable pneumatic box is compact in size and lightweight, which not only has six outputting gas paths, but also adjust pressure and flow. Within the allowable range of motion, the glove can assist affected hand to achieve flexion and extension of five fingers, and adduction and abduction of the thumb. The output force of the actuators is experimentally measured. For the flexion/extension actuator, the maximum force is 4.6N when extending and 1.9N when flexing. For the adduction/abduction actuator, the maximum force is 8.1N when adducting and 5.7N when abducting. In addition, demonstration with complete system is performed to do a functional evaluation, which verified the practical availability of the glove in rehabilitation training.

A 3D measurement system consisting of a 3D scanner and an industrial robot (eye-in-hand) is commonly used to scan large object under test (OUT) from multiple fieldof-views (FOVs) for complete measurement. A data stitching process is required to align multiple FOVs into a single coordinate system. Marker-free stitching assisted by robot’s accurate positioning becomes increasingly attractive since it bypasses the cumbersome traditional fiducial marker-based method. Most existing methods directly use initial Denavit-Hartenberg (DH) parameters and hand-eye calibration to calculate the transformations between multiple FOVs. Since accuracy of DH parameters deteriorates over time, such methods suffer from high stitching errors (e.g., 0.2 mm) in long-term routine industrial use. This paper reports a new robot-scanner calibration approach to realize such measurement with low data stitching errors. During long-term continuous measurement, the robot periodically moves towards a 2D standard calibration board to optimize kinematic model’s parameters to maintain a low stitching error. This capability is enabled by several techniques including virtual arm-based robot-scanner kinematic model, trajectory-based robot-world transformation calculation, nonlinear optimization. Experimental results demonstrated a low data stitching error (< 0.1 mm) similar to the cumbersome marker-based method and a lower system downtime (< 60 seconds vs. 10-15 minutes by traditional DH and hand-eye calibration).

The Proportional Slowdown Differentiation (PSD) is an important Quality of Service (QoS) model for services in the Cloud. It aims to maintain consistent pre-specified slowdown ratios between the requests of different service classes. In this paper, a novel online fuzzy learning algorithm is proposed to facilitate the PSD control for distributed services. The advantages of the proposed approach are self-optimization of its fuzzy rules and robust control performance. We built a prototype to realize the fuzzy learning algorithm and the PSD control mechanism based on an open source middleware platform. Comprehensive experiments over a wide range of parameters and workloads have been conducted. Experimental results show that the proposed fuzzy learning algorithm can achieve superior performance over several other controllers.

Aiming at nonlinear control system on-line modeling, the effects of training data distribution to the performance of SVR are analyzed, and a new modeling method based on modified support vector regression (SVR) is proposed. The analyzed results indicate that the data near to the new added sample should be preserved when training data are updated, which can improve the performance; the fixed parameters of SVR can not fit the entire on-line modeling process because the distribution is time-varying, and SVR may have a substandard performance when the system output changes in a small range. Therefore, three data updating criteria are proposed, and the width of the Gaussian kernel is set according to the correlation of training data in sampling time. The method is employed in multichannel electrohydraulic force servo synchronous loading system to predict the load output during 1280 sampling periods, the training time and the prediction mean absolute percentage error are 20.3387s and 0.357% respectively, and the results show its effectiveness.

Reinforcement Learning (RL) based approaches are now widely applied to building autonomous management applications with self-optimizing features. In this paper, we engineer an autonomous QoS controller based on a combination of fuzzy logic and RL to guarantee differentiated average response times for distributed services. We argue that there are two advantages of the fuzzy/RL QoS controller: the ability to autonomously tune if-then rules, and efficient modeling of the system state through fuzzy logic. Experiments conducted on a prototype show that the online learning fuzzy/RL controller can optimize its fuzzy rules effectively and efficiently and achieve better performance than manually tuned fuzzy controllers and the Self-Tuning Fuzzy Controller (STFC) used in [1].

QoS control is complicated by limited knowledge of system characteristics and continuous evolving features in modern distributed systems. In this paper, we propose a practical reinforcement learning based self-optimized QoS controller algorithm with the ability to guarantee differentiated average response time requirements for different service classes. The proposed solution consists of two key contributions: the reinforcement learning based self-optimizing control algorithm and full evaluations of this control algorithm. Experiments on the prototype show that standard reinforcement learning can learn and self- optimize the control knowledge efficiently in feasible training time with only partial system knowledge.