Image clustering is an effective way to discover and analyze large quantities of image data. The HSV color space is particularly advantageous in image feature extraction because of its relatively prominent feature vector. The objective of this study is to develop an image clustering method using the active-constraint semi-supervised affinity propagation (ACSSAP) algorithm. The algorithm adds supervision to the affinity propagation (AP) clustering algorithm with pairwise constraints and uses active learning to guide the AP clustering algorithm. Active learning of pairwise constraints leads to an adjustment of the similarity matrix in AP at each iteration. In the experiments, the advantage of HSV space is analyzed and the ACSSAP algorithm is evaluated for data sets of different sizes in comparison with other algorithms. The result demonstrates that the ACSSAP has better performance.
A novel micro-pressure power generation system (PPG) is proposed. It mainly consists of two parts: a piezoelectric ceramic power generation device and a super capacitor energy storage module (SCESM). The SCESM is composed of a conditioning circuit and a control circuit. The control circuit has a proportional-integral (PI) controller to smooth the voltage output with pulse-width modulation. The PPG system can generate power and conserve energy as well as support renewable and low-power energy. The superiority of the PPG system is verified through simulation and actual experiments. The process of storing energy takes 572 seconds, and the maximum error of the voltage output is 5%. In future research, we will be studying conversion efficiency optimization for the proposed system, an embedded power supply for wearable devices.
The pumping cavity we propose for solar-pumped lasers consists of a compound parabolic concentrator (CPC) and a conventional cone-shaped cavity. We applied ray tracing and FEM simulation in designing the cavity. Simulation results suggest that the cavity increases absorption efficiency 1.2 times over the conventional cone-shaped cavity and decreases the standard deviation of absorption density distribution by 33% and thermal stress by 19% over the conventional. Our evaluation of the cavity in lasing experiments realized 80 W in stable laser output.
Iron ore sintering process is the secondary most energy consuming procedure in steel making industry. In this study, a discrete wavelet transfer based back-propagation neural network (BPNN) model is built to predict the carbon efficiency of an iron ore sintering process. The raw-material variables and manipulated variables are chosen to be the inputs of the predictive model. First, the input variables are decomposed into 5 components. Then, BPNN models of each component are built. Finally, the prediction results are obtained by adding the output from each wave series. Actual run data are collected to verify the validity of the predictive model. The results show the validity of the proposed method with a MSE of 0.7708, a MAPE of 0.0125, and a R2 of 0.7016.
A dedicated LiFePO4 battery management system (BMS) used in the electrically controlled pneumatic brake system of a heavy-haul train needs a reliable, efficient and safe charging system to guarantee an uninterrupted power supply. To achieve that, this paper proposes an adaptive fast and safe charging strategy for Li-ion batteries based on the Lyapunov stability theory. In the strategy, a sliding-mode state of charge (SOC) observer is designed, and based on this, the dynamical reference charging current profile is derived. With the estimated SOC and dynamical setting current, the adaptive current control law is proposed by using a Lyapunov function. Finally, experiments are conducted to verify the feasibility and superiority of the proposed method.
Due to the disadvantages of existing ultracapacitor (UC) charging topologies, e.g., slow transient response, an improved three-level (TL) buck charging topology is proposed. Voltage is divied by the energy transformation of the flying-capacitance. The model of the circuit is shifted adaptively on the basis of the dynamic parameters of the reference current, improving the speed of the output current adjustment. Finally, simulation results validate the feasibility of the proposed TL buck topology for the UC energy storage system.
This study investigates stability problems related to discrete-time randomly switched genetic regulatory networks (GRNs) with time-varying delays. A new discrete-time randomly switched GRN model with known sojourn probabilities is proposed. By utilizing the discrete Wirtinger-based inequality and a newly proposed constraint condition on the feedback regulatory function, which have not been fully used in stability analysis of discrete-time GRNs, we establish delay-dependent stability and robust stability criteria. These criteria possess the sojourn probabilities of randomly switched GRNs. Two numerical examples are provided to demonstrate the effectiveness of the established results.
Drillability is a precondition for drilling-trajectory planning and intelligent control, as well as an important foundation for achieving safety, high quality, and efficiency in deep drilling. A formation drillability modeling method based on the Nadaboost-ELM algorithm is proposed. First, well logging parameters are chosen as inputs of the extreme learning machine (ELM) model, whose output is the formation drillability. Then, the models are trained as weak learners using the improved Adaboost algorithm. Finally, the weak learners are combined into a strong learner. The proposed modeling method is used for regression and prediction. In the regression aspect, the comparison results indicate that the proposed method has higher accuracy than methods in other studies, e.g., multiple regression (MR), grey model (GM), particle-swarm optimization back-propagation (PSO-BP), etc. In the prediction aspect, the results show that the proposed method is better than other prediction methods, e.g., MR, GM, BP, PSO-BP, ELM, and Adaboost-ELM at improving the model’s prediction accuracy, which provides a foundation for intelligent geological drilling.
This study focuses on maximum power point tracking (MPPT) control for photovoltaic (PV) power generation systems under partial shading conditions. A mathematic model of the partially shaded solar cell is built. Then, the output characteristics of the partial-shade array are analyzed. Based on the model of the PV battery and the concept of the average-state switch cycle, an average-state mathematical model of the PV power generation system using a boost circuit for the realization circuit is established. A sliding mode controller based on the integral sliding mode function is designed to realize MPPT in the PV power generation system. Finally, simulations in MATLAB/Simulink confirm the functionality and performance of the proposed controller.
Sparse signal reconstruction (SSR) problems based on compressive sensing (CS) arise in a broad range of application fields. Among these are the so-called “block-structured” or “block sparse” signals with nonzero atoms occurring in clusters that occur frequently in natural signals. To make block-structured sparsity use more explicit, many block-structure-based SSR algorithms, such as convex optimization and greedy pursuit, have been developed. Convex optimization algorithms usually pose a heavy computational burden, while greedy pursuit algorithms are overly sensitive to ambient interferences, so these two types of block-structure-based SSR algorithms may not be suited for solving large-scale problems in strong interference scenarios. Sparse adaptive filtering algorithms have recently been shown to solve large-scale CS problems effectively for conventional vector sparse signals. Encouraged by these facts, we propose two novel block-structure-based sparse adaptive filtering algorithms, i.e., the “block zero attracting least mean square” (BZA-LMS) algorithm and the “block ℓ0-norm LMS” (BL0-LMS) algorithm, to exploit their potential performance gain. Experimental results presented demonstrate the validity and applicability of these proposed algorithms.
To detect demagnetization faults in real time based on an adaptive sliding mode observer, we used a permanent-magnet synchronous motor (PMSM). Demagnetization faults are first modeled for the PMSM using coordinates oriented to the magnetic field. To solve demagnetization faults problems as multiple parameters change, we used adaptive and sliding mode variable structure control and designed an adaptive sliding mode observer. The adaptive estimation algorithm of the permanent magnet flux is given and the observer’s stability is proven by Lyapunov stability theory. Simulation and experimental results demonstrate the feasibility and effectiveness of our proposal.
In this paper, local Gaussian process (GP) approximation is introduced to build the critic network of adaptive dynamic programming (ADP). The sample data are partitioned into local regions, and for each region, an individual GP model is utilized. The nearest local model is used to predict a given state-action point. With the two-phase value iteration method for a Gaussian-kernel (GK)-based critic network which realizes the update of the hyper-parameters and value functions simultaneously, fast value function approximation can be achieved. Combining this critic network with an actor network, we present a local GK-based ADP approach. Simulations were carried out to demonstrate the feasibility of the proposed approach.
Students have different levels of motivation, approaches to learning, and intellectual levels. The better that instructors understand these differences, the better the chances they have of improving their quality of teaching. To explore differences thoroughly, we focuses on three crucial factors in student learning features – i.e., personality, learning style and multiple intelligences – and propose an approach effective in classifying students for the purpose of instructing instructors while optimizing their teaching process. We collected data on learning features from a class of 58 college students and analyzed these data by using principal component analysis (PCA) and then classified them using Ward clustering. Results of experiments indicate that our proposal effectively classifies students based on their learning features and that classification results facilitate instructors in creating personalized teaching strategies.
Japan is an island nation that experiences frequent earthquakes. When an earthquake occurs, it is important to forecast its resultant tsunami: its size, location, time of arrival, etc. These forecasts are made using numerical simulations. The initial conditions are very important for numerical simulations, but the small number of tide stations makes it difficult to make highly precise forecasts. The distance between stations is normally several tens of km, and this lowers the precision of the initial data afforded by them. It is therefore common to use data interpolated from the sparse observation data at time t=0. Even so, high-resolution interpolation cannot be expected since the original data is of poor quality. In addition, the interpolated values may not be physically valid because the governing equation may not have been considered when the data were interpolated. We therefore propose a new method of estimating the initial value by using a characteristic equation. In this method, we replace the spatial resolution with time resolution. This results in a high-resolution initial value because the same place is measured more than once. In addition, the characteristic equation is based on the governing equation. Therefore, in this method, an accurate estimation of initial value is considered to be possible. In this paper, we show two applications of this approach, one for a dimensional shallow water wave equation and one for Euler’s equation. The shallow water wave equation is for the tsunami, and the Euler equation is the governing equation of the numerical weather forecast.
Stable control of the ball mill grinding process is very important to reduce energy losses, enhance operation efficiency, and recover valuable minerals. In this work, a controller for the ball mill grinding process is designed using a combination of model predictive control (MPC) with the equivalent-input-disturbance (EID) approach. MPC has been researched and applied widely as one of the multi-variable control algorithms for grinding. It is used to decouple in real time. The controller design does not deal with the disturbances directly. However, strong disturbances such as those caused by ore hardness and feed particle size exist in the ball mill grinding. EID estimates the equivalent disturbance of the grinding circuit in the control input channel and integrates this disturbance directly into the control law in order to suppress disturbances promptly and effectively. This results in good disturbance suppression performance. Simulation results demonstrate that the combination of MPC with EID for controlling the ball mill grinding circuit yields better performance in terms of disturbance rejection, rapid response, and strong robustness as compared to the performance of the MPC and proportional-integral (PI) decoupling control.
Pedestrian detection systems are increasing in popularity recently. These systems that work together with car-mounted cameras need to operate in real-time. A Field Programmable Gate Array (FPGA) is able to work in a highly optimized parallel process and hence it is expected to work in real-time. However, it is difficult for FPGA to calculate complex processes. Therefore, pedestrian detection methods must have low computational costs in order to implement the system using FPGA. This paper proposes a system that uses Binary Robust Independent Elementary Features (BRIEF) and a Neural Network (NN) as a pedestrian detection method. The system was simulated using a CPU and the human detection performance was evaluated. Additionally, the NN was trained using three Particle Swarm Optimization (PSO) methods. The performance of our approach was shown using a Receiver Operating Characteristic (ROC) curve with respect to each learning method. In the future, the system needs to improve the human detection rate and it will be implemented and simulated using an FPGA.
Precursory earthquake data are linked closely to the earthquake processes. Taking the Tibetan Autonomous Region’s Yushu County earthquake as an example, we analyzed three types of earthquake signals and studied a modeling method for self-adaptative matching warning data on precursory data’s fingerprint features. We calculated different timescale features of precursory fingerprint pattern images based on statistical physics and image matching. We also developed corresponding fuzzy discriminant rules and established a database of warning-image fingerprint pattern features for the Yushu County region and studied evolutionary laws for the data feature patterns under different time scales during abnormal development in front of and behind of abnormal development. Result were similar to the general “fingerprint” pattern feature among precursory earthquake data for different signal channels, but the details of these characteristics are completely different. This special “fingerprint” image pattern feature is useful as on early warning of possible geological follow-up activity. Our method could improve the limitations on and low efficiency of manual handling and could also heighten observational accuracy and work efficiency.
The objective of this paper is to present a novel method, based on a swarm intelligence algorithm, for ellipse detection in digital images of embryo. The process is carried out in several stages. First, edge detection is performed on the image. Then, line segments in the image are detected, and potential elliptical arc segments are extracted from the line segments. Afterward, the detection process is carried out using the Particle Swarm Optimization (PSO) method, which utilize the calculation of the fitness function from the arc segment previously detected. The PSO technique, which is the idea behind the proposed algorithm, is used to find the actual ellipses by combining potential elliptical arcs. The best combination of potential arcs is determined by means a voting technique that utilizes three important points on the arc, namely the starting point, midpoint, and endpoint, so the voting is more efficient than doing the voting for every single pixel in the image. Furthermore, this method is used an embryo image that has following the characteristics: multiple ellipses, a lot of noise, an incomplete ellipse, low image contrast, and overlapping cells. Experiment show that the proposed method detects the ellipses better than do several voting-based ellipse detection methods such as RHT, IRHT, and PSORHT. On the other hand, the experiments show that the proposed method has a higher average hit rate than do other methods. This research is used to increase the success rate of In-Vitro Fertilization (IVF).
Traffic flow prediction plays an important role in intelligent transportation systems. With the rapid growth of traffic flow data, fast and accurate traffic flow prediction methods are now required. In this paper, we propose a novel fast learning data-driven fuzzy approach for the traffic flow prediction problem. In the proposed approach, to achieve fast learning, an extreme learning machine is utilized to optimize the consequent parameters of the fuzzy rules. Further, a fuzzy rule pruning strategy that involves measuring the firing levels of the fuzzy rules is presented to obtain reduced fuzzy inference systems. To evaluate the performance of the proposed approach, it was experimentally applied to traffic flow prediction and its results compared with those of widely used methods. The experimental results verify that the proposed approach can achieve satisfactory performance. The comparisons show that the proposed approach can obtain better (sometimes similar) performances, but with a simpler structure, fewer parameters, and much faster learning speed than the other methods.