During the past few years, Wireless Sensor Networks (WSNs) have become widely used due to their large amount of applications. The use of WSNs is an imperative necessity for future revolutionary areas like ecological fields or smart cities in which more than hundreds or thousands of sensor nodes are deployed. In those large scale WSNs, hierarchical approaches improve the performance of the network and increase its lifetime. Hierarchy inside a WSN consists in cutting the whole network into sub-networks called clusters which are led by Cluster Heads. In spite of the advantages of the clustering on large WSNs, it remains a non-deterministic polynomial hard problem which is not solved efficiently by traditional clustering. The recent researches conducted on Machine Learning, Computational Intelligence, and WSNs bring out the optimized clustering algorithms for WSNs. These kinds of clustering are based on environmental behaviors and outperform the traditional clustering algorithms. However, due to the diversity of WSN applications, the choice of an appropriate paradigm for a clustering solution remains a problem. In this paper, we conduct a wide review of proposed optimized clustering solutions nowadays. In order to evaluate them, we consider 10 parameters. Based on these parameters, we propose a comparison of these optimized clustering approaches. From the analysis, we observe that centralized clustering solutions based on the Swarm Intelligence paradigm are more adapted for applications with low energy consumption, high data delivery rate, or high scalability than algorithms based on the other presented paradigms. Moreover, when an application does not need a large amount of nodes within a field, the Fuzzy Logic based solution are suitable.

No abstract

No abstract

No abstract

Optimization energy is a technique helpful to manage electricity consumption of home devices according to the electric system. CBR is used to predict consumption but lacks to be generic. This paper intends to design a more generic CBR approach by relying on various intelligences. The retrieve process includes four steps. The first step is weight evaluation of attributes based on AHP. The second step exploits an adapted cosine model for distance similarity. The third and fourth steps use k-Means and k-NN to identify the most similar cases. The reuse process is defined as a linear programming problem solved by PSO. During revise, an algorithm based on the reuse model and SVR, derives the revised solution. Experiments on a dataset of 1096 samples are made for forecasting energy electricity consumption. CBR revise process is 99.35% accurate, improving the reuse accuracy by 11%. The proposed architecture is a potential in energy management as well as for other prediction problems.