Data clustering is the process of partitioning a set of data objects into meaning clusters or groups. Due to the vast usage of clustering Algorithms in many fields, a lot of research is still going on to find the best and efficient clustering Algorithm to partition the data items. K-means is simple and easy to implement, but it suffers from initialization of cluster center and hence trapped in local optimum. In this paper, a new hybrid data clustering approach which combines the modified Krill Herd and K-means Algorithms, named as K-MKH, is proposed. K-MKH Algorithm utilizes the power of quick convergence behaviour of K-means and efficient global exploration of Krill Herd and random phenomenon of Levy flight method. The Krill-Herd Algorithm is modified by incorporating Levy flight into it to improve the global exploration. The proposed Algorithm is tested on artificial and real life datasets. The simulation results are compared with other methods such as K-means, Particle Swarm Optimization (PSO), Original Krill Herd (KH), hybrid K-means and KH. Also the proposed Algorithm is compared with other evolutionary Algorithms such as hybrid modified cohort intelligence and K-means (K-MCI), Simulated Annealing (SA), Ant Colony Optimization (ACO), Genetic Algorithm (GA), Tabu Search (TS), Honey Bee Mating Optimization (HBMO) and K-means++. The comparison shows that the proposed Algorithm improves the clustering results and has high convergence speed.

1. Introduction An effective way to achieve an economical design of water distribution networks (WDNs) is to utilize the metaheuristic optimization Algorithms profited by swarm intelligence. In this research, Krill Herd (KH) Optimization Algorithm was applied to obtain the optimum design of water distribution networks (WDNs). For this purpose, the KH Algorithm was linked with EPANET hydraulic software (Rossman 2000) in MATLAB. The capital cost was considered as the objective function in Kadu and Khorramshahr WDNs herein. The obtained optimum design cost for two WDNs was compared with the solutions published using other approaches...

In this study first, we investigated the equations in movement of an underwater robot and also the state-space model of system is expressed with linearizing existing equations. Then, an energy efficient path is planned using dynamic equations and dynamic planning optimization method. There are moving obstacles in environment in which a robot is moving. It can be seen that the planned path is flat and energy consumption is minimized. The main objective of the study is to present an appropriate controller for the provided state-space model of system. For this purpose, by studying the system controller designing using LQR optimal controller, an appropriate controller for model has been presented. The planning a path to a target for the underwater robot has been presented using combination of optimization Algorithm with learning automata and the Krill Herd optimization Algorithm. In other words, the study has applied hybrid Algorithm in order to find optimal path for the underwater robot to move in a static environment which is expressed through the map with the nodes and links.

Objective: Portfolio optimization is one of the most important issues in investment. Harry Markowitz was the first person who applied risk with this regard. This issue was later studied from different perspectives, using various risk measures, optimization methods, and considering transaction costs. In this research, we aim to use the Krill Herd metaheuristic Algorithm in portfolio optimization, and examine its possible advantages. Methods: In the present study, we try to solve the portfolio optimization problem and to find the efficient frontier using Krill Herd’ s novel Algorithm. We also consider three different measures for risk: variance, semi-variance, and expected shortfall. Our data consists of adjusted returns of the top fifty stocks in Tehran Stock Exchange from 2012 to 2018. Results: At first, the efficient frontiers of the optimal portfolios, using different measures for risk were plotted. The relative similarity of the three plots indicates the stability of the Krill Herd Algorithm in obtaining efficient frontiers. Then, we observed that the Sharpe Ratios of this Algorithm are higher than those of Imperialist Competitive and Particles Swarm Algorithms. Conclusion: The Krill Herd Algorithm has a better performance finding efficient frontier and optimized portfolios in comparison to the other common Algorithms; therefore, it can be used instead of the other Algorithms to obtain better results.

Internet of Things (IoT) technology involves a large number of sensor nodes that generate large amounts of data. Optimal energy consumption of sensor nodes is a major challenge in this type of network. Clustering sensor nodes into separate categories and exchanging information through headers is one way to improve energy consumption. This paper introduces a new clustering-based routing protocol called KHCMSBA. The proposed protocol biologically uses fast and efficient search features inspired by the Krill Herd optimization Algorithm based on Krill feeding behavior to cluster the sensor nodes. The proposed protocol also uses a mobile well to prevent the hot spot problem. The clustering process at the base station is performed by a centralized control Algorithm that is aware of the energy levels and position of the sensor nodes. Unlike protocols in other research, KHCMSBA considers a realistic energy model in the grid that is tested in the Opnet simulator and the results are compared with AFSRP (Artifical Fish Swarm Routing ProtocolThe simulation results show better performance of the proposed method in terms of energy consumption by 12.71%, throughput rate by 14.22%, end-to-end delay by 76.07%, signal-to-noise ratio by 82.82%. 46% compared to the AFSRP protocol

One of the most fundamental problems in investment decisions and portfolio optimization is choosing a suitable measure for risk assessment and management. In this study, the performance of the Krill Herd Algorithm is investigated for solving the mean-value at risk and mean-conditional value at risk portfolio optimization models considering the cardinality constraints, among 35 active companies in Tehran Stock Exchange. For Algorithm training, the roller window method has been used in 2011-2018 and 2012-2019. The Sharpe ratio and the conditional Sharpe ratio of the models have been evaluated and they are compared using the Wilcoxon test. According to the numerical results, the mean–conditional value at risk model outperforms the mean–value at risk model in terms of the rate of return. Also, the model’s profitability improved using cardinality constraint with 5 stocks. Based on the empirical studies, we concluded that there is no significant difference between the performance of the value at risk and conditional value at risk based models. Furthermore, the portfolios with lower number of stocks have shown the better performance.

Purpose: One of the topics for manufacturers is to discuss the diversity of customer tastes. To manage this situation with the least change in products, multiple assembly lines make the necessary flexibility to produce the products. In multi-model assembly lines, different product types in different batches are produced and there is a setup time to prepare assembly lines between two types of products to produce another product type. This paper aims to investigate multi-model assembly lines and their sequencing, balancing, and worker assignment due to the existence of various tasks for workers according to learning and disremembering effects. Frequent changes in the product design of multi-model assembly lines according to customer demands can reduce the learning effect of workers and increase task times, while in another view, repeating tasks, particularly for products with more demands can increase the learning effect and reduce the task times. Therefore, in this study, the effects of workers' learning and disremembering multi-model assembly line balancing, sequencing, and worker assignment are investigated to minimize the number of workstations for a given cycle time not only to cover the different tastes of customers, but also indirectly minimize the costs of building stations, hiring, and employing manpower. Design/methodology/approach: In this paper, as an innovation, a mixed-integer mathematical model for multi-model assembly line balancing, sequencing, and worker assignment with different workers' skill levels and learning and disremembering rates has been developed to minimize the number of stations. Based on the nature of the multi-model, random demand for each product has been considered. After mathematical modeling, different small-sized problems have been solved by the GAMS software. Results and sensitivity analysis underlined the validity of the proposed model. Since this problem is typically NP-hard, GAMS software cannot solve medium and large-sized problems in a reasonable time. Therefore, the Krill Herd optimization and Particle Swarm Optimization (PSO) Algorithms have been used for medium and large-sized problems, which have not been used earlier in similar cases. The Krill Herd optimization Algorithm has been used as the proposed Algorithm and PSO has been used as a competing Algorithm. The parameters of both Algorithms have been adjusted by the Taguchi method, and the best level has been selected for each parameter. Findings: 12 test problems were solved with different sizes. Results indicated that only five GAMS problems could reach the optimal solution. For better comparison of the Krill Herd optimization and the particle swarm optimization Algorithm, each test problem was run 30 times and minimum, maximum, and average objective function and their running times were reported. The results indicated that the objective function of both metaheuristic Algorithms was the same but the Krill Herd optimization Algorithm can achieve optimal or near-optimal answers in less time than GAMS and the PSO Algorithm declared the efficiency of the proposed Algorithm in solving these problems. Research limitations/implications: One of the limitations in this research was the lack of cooperation of factories whose assembly lines were similar to the problem considered in this study, and in this regard, the real-world data was not accessible. Therefore, the standard test problems were used that existed in the famous database of assembly line balancing problems. Since the problem in this paper was new, some other required data, and different examples in different ways needed to be considered, randomly. Another limitation of using this research in a real-world situation was the challenge of exact determination of learning and disremembering rate of each worker which can be solved by using experts in the field of assessment and training. Originality/value: In this paper, a mathematical model was developed for multi-model assembly line balancing, sequencing, and worker assignment according to the learning and disremembering effect. Since the problem was NP-Hard, as well as GAMS software, two metaheuristic Algorithms were applied for a similar problem, and their efficiency was compared with each other. The two-mentioned Algorithms have not been used in previous studies. Both academic researchers and production managers can benefit from applying the findings of this study.

Data mining techniques have been applied to analyze, predict and diagnose diseases. The prediction of disease becomes meaningless when there is no proper recommendation of a drug to the patient. A drug recommendation method called Artificial Neural Network (ANN) with side effect constraints was proposed to recommend drug names for multiple diseases such as Chronic Kidney Disease (CKD), diabetic and heart disease based on the interaction between drug and disease and their side effects. In this drug recommendation method, multiple attributes of drugs and patients were collected from different sources and the hidden relationship between the attributes was predicted by using a Hidden Markov Model (HMM). In addition to this, statistical features were calculated and added as additional features. The collected and calculated features were used in ANN with side effect constraint classifier which predicted drug name for multiple diseases with the consideration of side effects. However, there is a high dimensionality problem in the recommended method due to more number of features. Moreover, it leads to more computational and space complexity in the ANN classifier. In this paper, an efficient Krill Herd (KH) Algorithm for optimization is introduced to solve the above-mentioned problems in the drug recommendation method. According to the Herding behavior of the likeness of the Krill individuals, KH selects the optimal features. The multiple attributes of drugs and patients are collected in a different time slots. The KH Algorithm is also used to select the optimal time slot. Then, the optimal time slot and features are given as input to ANN which predicts drug names for multiple diseases with high accuracy and low computational complexity.

In this paper, two recent heuristic optimization Algorithms are presented to optimally manage the operation of the microgrid(MG) with installed renewable energy sources (RESs); Krill Herd (KH) optimization and ant lion optimizer (ALO)Algorithms. The first Algorithm is used for solving single-objective function represents either total operation cost or totalpollutant emission injected from the installed generating units while ALO is applied to solve the multi-objective function ofboth total operating cost and emission. The problem is formulated as nonlinear constrained objective function with equalityand inequality constraints. In this work; the devices installed in MGs are photovoltaic panel (PV), wind turbine (WT), microturbine(MT), fuel cell (FC), battery and grid. Two scenarios are studied; the first one is optimizing MG with installing allRESs within specified limits in addition to grid, while the second scenario is operating both PV and WT at their rated powers. The obtained results are compared with different reported Algorithms like genetic Algorithm (GA), Fuzzy self-adaptivePSO (FSAPSO) and others programmed like particle swarm optimization (PSO), grey-wolf optimizer (GWO) and whaleoptimization Algorithm (WOA). For first scenario; the proposed KH gives the best optimal cost of 105. 94 € ct while the bestemission is 420. 57 kg, the best optimal cost and emission of 592. 86 € ct 339. 71 kg are obtained via KH in the second scenario.

The Internet of Things is a new technology that communicates with the surrounding objects through the Internet and is used for the purpose of remote measurement and control. In the field of Internet of Things (IoT) network security, it is very important to accurately identify the types of attacks on these networks that are launched by zombie hosts under the control of the attacker. In this article, a new neural network is proposed to improve the detection of intrusion into the Internet of Things network based on the ALEXNET convolutional neural network and chaotic Krill optimization Algorithm (MONANET). In the MONANET network, in order to improve the accuracy in detecting intrusion into the IoT network and not need to manually adjust the parameters, the hyperparameters of the neural network are dynamically selected using the chaotic Krill Algorithm. The value of the loss function of the validation set obtained from the first training of the neural network model using the Danmini doorbell dataset is considered as the CKH fitness value. The comprehensive performance of the proposed network and GRU, ANN, SVM, LSTM, R-CNN, and APSO-CNN Algorithms have been compared in five evaluation indices and 12 times independent experiments. The obtained results show the improvement of intrusion detection to the Internet of Things network. The proposed Algorithm has been able to accurately detect %99.89 attacks on the Internet of Things network. The experimental results show the superiority of the proposed method over other knowledge boundary methods in terms of improving classification accuracy.