A Modified Jellyfish Search Optimizer with Opposition Based Learning and Biased Passive Swarm Motion

Jellyfish Search Optimizer (JSO) is one of the latest nature inspired optimization algorithms. This paper aims to improve the convergence speed of the algorithm. For the purpose, it identifies two modifications to form a proposed variant. First, it proposes improvement of initial population using Opposition based Learning (OBL). Then it introduces a probability-based replacement of passive swarm motion into moves biased towards the global best. OBL enables the algorithm to start with an improved set of population. Biased moves towards global best improve the exploitation capability of the algorithm. The proposed variant has been tested over 30 benchmark functions and the real world problem of 10-bar truss structure design optimization. The proposed variant has also been compared with other algorithms from the literature for the 10-bar truss structure design. The results show that the proposed variant provides fast convergence for benchmark functions and accuracy better than many algorithms for truss structure design.

A High-robust Sensor Activity Control Algorithm for Wireless Sensor Networks

In wireless sensor networks, it is important to use the best number of sensors to optimize the network and consider the key design and cost. Owing to the limited power of sensors, how controlling the state of the sensor through an automatic control algorithm and power-saving and efficient distribution of work have become important issues. However, sensor nodes are usually deployed in dangerous or inaccessible locations. Therefore, it is difficult and impractical to supply power to sensors through humans. In this study, we propose a high-reliability control algorithm with fast convergence and strong self-organization ability, called sensor activity control algorithm (SACA), which can efficiently control the number of sensors in the active state and extend their use time. SACA considers the relationship between the total number of inactive sensors and the target value and determines the state of the sensor in the next round. The data transmission technology of random access is used between the sensor and the base station; therefore, the sensor in the sleep state does not need to receive the feedback packet from the base station. The sensor can achieve true dormancy and power-saving effects. The experimental results show that SACA has fast convergence, strong self-organization capabilities, and power-saving advantages.

Understanding approximate Fisher information for fast convergence of natural gradient descent in wide neural networks*

Natural gradient descent (NGD) helps to accelerate the convergence of gradient descent dynamics, but it requires approximations in large-scale deep neural networks because of its high computational cost. Empirical studies have confirmed that some NGD methods with approximate Fisher information converge sufficiently fast in practice. Nevertheless, it remains unclear from the theoretical perspective why and under what conditions such heuristic approximations work well. In this work, we reveal that, under specific conditions, NGD with approximate Fisher information achieves the same fast convergence to global minima as exact NGD. We consider deep neural networks in the infinite-width limit, and analyze the asymptotic training dynamics of NGD in function space via the neural tangent kernel. In the function space, the training dynamics with the approximate Fisher information are identical to those with the exact Fisher information, and they converge quickly. The fast convergence holds in layer-wise approximations; for instance, in block diagonal approximation where each block corresponds to a layer as well as in block tri-diagonal and K-FAC approximations. We also find that a unit-wise approximation achieves the same fast convergence under some assumptions. All of these different approximations have an isotropic gradient in the function space, and this plays a fundamental role in achieving the same convergence properties in training. Thus, the current study gives a novel and unified theoretical foundation with which to understand NGD methods in deep learning.

An improved 2-level MPPT scheme for photovoltaic systems using a novel high-frequency learning based adjustable gain-MRAC controller

Under rapidly changing environmental conditions, the model reference adaptive control (MRAC) based MPPT schemes need high adaptation gain to achieve fast convergence and guaranteed transient performance. The high adaptation gain causes high-frequency oscillations in the control signals resulting in numerical instability and inefficient operation. This paper proposes a novel high-frequency learning-based adjustable gain MRAC (HFLAG-MRAC) for a 2-level MPPT control architecture in photovoltaic (PV) systems to ensure maximum power delivery to the load under rapidly changing environmental conditions. In the proposed 2-level MPPT control architecture, the first level is the conventional ripple correlation control (RCC) that yields a steady-state ripple-free optimum duty cycle. The duty cycle obtained from the first level serves as an input to the proposed HFLAG-MRAC in the second level. In the proposed adaptive law, the adaptation gain varies as a function of the high-frequency ripple content of the tracking error. These high-frequency contents are the difference between the tracking error and its low-pass filtered version representing the fluctuations in output due to rapid changes in the environmental conditions. Thus, adjusting the adaptation gain by high-frequency content of the tracking error ensures fast convergence, guaranteed transient performance, and overall system stability without needing high adaptation gain. The adaptive law of the proposed HFLAG-MRAC is derived using the Lyapunov theory. Simulation studies, experimental analysis, and performance comparison with recent similar work validate the effectiveness of the proposed work.

Sensitivity calibration of a three-axis accelerometer under different temperature conditions using the hybrid GA–PSO–BPNN algorithm

Purpose Three-axis accelerometers play a vital role in monitoring the vibrations in aircraft machinery, especially in variable flight temperature environments. The sensitivity of a three-axis accelerometer under different temperature conditions needs to be calibrated before the flight test. Hence, the authors investigated the efficiency and sensitivity calibration of three-axis accelerometers under different conditions. This paper aims to propose the novel calibration algorithm for the three-axis accelerometers or the similar accelerometers. Design/methodology/approach The authors propose a hybrid genetic algorithm–particle swarm optimisation–back-propagation neural network (GA–PSO–BPNN) algorithm. This method has high global search ability, fast convergence speed and strong non-linear fitting capability; it follows the rules of natural selection and survival of the fittest. The authors describe the experimental setup for the calibration of the three-axis accelerometer using a three-comprehensive electrodynamic vibration test box, which provides different temperatures. Furthermore, to evaluate the performance of the hybrid GA–PSO–BPNN algorithm for sensitivity calibration, the authors performed a detailed comparative experimental analysis of the BPNN, GA–BPNN, PSO–BPNN and GA–PSO–BPNN algorithms under different temperatures (−55, 0 , 25 and 70 °C). Findings It has been showed that the prediction error of three-axis accelerometer under the hybrid GA–PSO–BPNN algorithm is the least (approximately ±0.1), which proved that the proposed GA–PSO–BPNN algorithm performed well on the sensitivity calibration of the three-axis accelerometer under different temperatures conditions. Originality/value The designed GA–PSO–BPNN algorithm with high global search ability, fast convergence speed and strong non-linear fitting capability has been proposed to decrease the sensitivity calibration error of three-axis accelerometer, and the hybrid algorithm could reach the global optimal solution rapidly and accurately.

Grey Wolf Resampling-Based Rao-Blackwellized Particle Filter for Mobile Robot Simultaneous Localization and Mapping

An artificial intelligent grey wolf optimizer (GWO)-assisted resampling scheme is applied to the Rao-Blackwellized particle filter (RBPF) in the simultaneous localization and mapping (SLAM). By doing this, we can make the diversity of the particles resampling and then obtain a better localization accuracy and fast convergence to realize indoor mobile robot SLAM. In addition, we propose an adaptive local data association (Range-SLAM) scheme to improve the computational efficiency for the algorithm of the nearest neighbor (NN) data association in the iteration of the RBPF prediction. Through the experiment and simulations, the proposed SLAM schemes have fast convergence, accuracy, and heuristics. Therefore, the improved RBPF and new data association schemes presented in this paper can provide a feasible method for the indoor mobile robot SLAM.