Fabio Massimo Frattale Mascioli

Adaptive resolution min-max classifiers

A high automation degree is one of the most important features of data driven modeling tools and it should be taken into consideration in classification systems design. In this regard, constructive training algorithms are essential to improve the automation degree of a modeling system. Among neuro-fuzzy classifiers, Simpsons (1992) min-max networks have the advantage of being trained in a constructive way. The use of the hyperbox, as a frame on which different membership functions can be tailored, makes the min-max model a flexible tool. However, the original training algorithm evidences some serious drawbacks, together with a low automation degree. In order to overcome these inconveniences, in this paper two new learning algorithms for fuzzy min-max neural classifiers are proposed: the adaptive resolution classifier (ARC) and its pruning version (PARC). ARC/PARC generates a regularized min-max network by a succession of hyperbox cuts. The generalization capability of ARC/PARC technique mostly depends on the adopted cutting strategy. By using a recursive cutting procedure (R-ARC and R-PARC) it is possible to obtain better results. ARC, PARC, R-ARC, and R-PARC are characterized by a high automation degree and allow to achieve networks with a remarkable generalization capability. Their performances are evaluated through a set of toy problems and real data benchmarks. The paper also proposes a suitable index that can be used for the sensitivity analysis of the classification systems under consideration.

A constructive algorithm for binary neural networks: the oil-spot algorithm

This paper presents a constructive training algorithm for supervised neural networks. The algorithm relies on a topological approach, based on the representation of the mapping of interest onto the binary hypercube of the input space. It dynamically constructs a two-layer neural network by involving successively binary examples. A convenient treatment of real-valued data is possible by means of a suitable real-to-binary codification. In the case of target functions that have efficient halfspace union representations, simulations show the constructed networks result optimized in terms of number of neurons.

A constructive approach to neuro-fuzzy networks

Abstract A new net of ANFIS class, with promising performances in terms of structural complexity, generalisation capability and speed of convergence, is derived by means of an optimal set of fuzzy If–Then rules. The rules are automatically obtained by relying on Simpsons Min–Max technique and on the constructive approach suggested by Learning Theory, with the aim of extracting the maximum possible information content from the training set. Both classification and prediction benchmarks are considered for illustration.

Genetic optimization of a fuzzy control system for energy flow management in micro-grids

In this paper we present an interesting application of Computational Intelligence techniques for the power demand side and flow management optimization in a microgrid. In particular, we used a Fuzzy Logic Controller (FLC) for Time-of use Cost Management program in the microgrid. FLC can either sell and buy energy from outside the microgrid making use of an aggregate of energy storage capacity realized with lithium ion batteries. According to the hybrid Fuzzy-GA paradigm, the Fuzzy Logic Controller that operates decision making on energy flows is optimized by a Genetic Algorithm. The experimental results show that the proposed control system can manage effectively the energy trade with the main grid on the basis of real time prices.

A recursive algorithm for fuzzy min-max networks

An algorithm to train min-max neural models is proposed. It is based on the adaptive resolution classifier (ARC) technique, which overcomes some undesired properties of the original Simpsons (1992) algorithm. In particular, training results do not depend on pattern presentation order and hyperbox expansion is not limited by a fixed maximum size, so that it is possible to have different covering resolutions. ARC generates the optimal min-max network by a succession of hyperbox cuts. The generalization capability of the ARC technique depends mostly on the adopted cutting strategy. A new recursive cutting procedure allows ARC technique to yield a better performance. Some real data benchmarks are considered for illustration.

Genetic Optimization of a PD Diagnostic System for Cable Accessories

An automatic procedure, based on a genetic algorithm capable of optimizing a diagnostic system for the recognition and identification of partial-discharge (PD) pulse patterns in the terminations and joints of solid dielectric extruded power distribution cables, is described. The core of the diagnostic system is a fuzzy neural network, namely a Min-Max classifier. The genetic optimization is capable for reducing the system complexity, while enhancing its diagnostic performance. The developed procedure is sufficiently general to be applied to PD source identification in the cables themselves as well as other electric power apparatus.

A Novel Mechanical Analogy-Based Battery Model for SoC Estimation Using a Multicell EKF

The future evolution of technological systems dedicated to improve energy efficiency will strongly depend on effective and reliable energy storage systems, as key components for smart grids, microgrids, and electric mobility. Besides possible improvements in chemical materials and cells design, the battery management system is the most important electronic device that improves the reliability of a battery pack. In fact, a precise state of charge (SoC) estimation allows the energy flows controller to better exploit the full capacity of each cell. In this paper, we propose an alternative definition for the SoC, explaining the rationales by a mechanical analogy. We introduce a novel cell model, conceived as a series of three electric dipoles, together with a procedure for parameters estimation relying only on voltage measures and a given current profile. The three dipoles represent the quasi-stationary, the dynamic, and the instantaneous components of voltage measures. An extended Kalman filter (EKF) is adopted as a nonlinear state estimator. Moreover, we propose a multicell EKF system based on a round-robin approach to allow the same processing block to keep track of many cells at the same time. Performance tests with a prototype battery pack composed by 18 A123 cells connected in series show encouraging results.

Comparison between time-constrained and time-unconstrained optimization for power losses minimization in Smart Grids using genetic algorithms

The power losses reduction is one of the main targets for any electrical energy distribution company. This paper studies the applicability of a control system based on a Genetic Algorithm (GA) on a portion of the actual Italian electric distribution network located in Rome and surroundings, managed by the ACEA Distribuzione S.p.A. The joint optimization of both power factor correction (PFC) and distributed feeder reconfiguration (DFR) is faced. The PFC is performed tuning the phases of the distributed generators (DGs) and the output voltage of the Thyristor Voltage Regulator (TVR). The DFR is performed by opening and closing the available breakers according to a graph based algorithm that is able to find all the possible radial configurations of the network. The joint PFC and the DFR optimization problem are faced by solving a suitable optimization problem, defining the fitness function that drives the GA. In order to have the opportunity to study a realistic future scenario, the actual network has been modified by introducing a few extra distributed generators. Aiming to validate the applicability of the proposed algorithm to an operative scenario, two different tests have been performed. The first one, referred to as time-unconstrained optimization, represents an ideal scenario where there are no constraints on time available for optimization. The second one, referred to as time-constrained optimization, represents a real scenario where the optimization must be completed within a time slot of one hour. Both tests have been performed by feeding the developed simulation tool with real data concerning dissipated and generated active and reactive power values. The comparison between results obtained in the two tests campaigns furnishes the opportunity to evaluate the effectiveness of the proposed control algorithm in real time, relying on the computational performances of an entry-level workstation. The obtained results encourage the use of derivative free methods in a real-time control scenario, showing that the performances achieved by the time-constrained optimization procedures are very close in terms of objective function values to the ones obtained by the time-unconstrained procedure.

Design and implementation of a fast recharge station for electric vehicles

In this paper it is described an innovative project born by the cooperation between Polo per la Mobilità Sostenibile (POMOS) and Citroën Italia. The project concerned the implementation of a high power fast recharge station for electric vehicles in the parking lot of the Citroën dealer. This station is free and available for all the owners of electric vehicles with no restriction on the vehicle brand. The fast recharge station is able to restore the battery pack charge to 80% of maximum value in about 20 minutes using direct current charging at 400 V. The whole project includes in addition to the fast recharge dock two traditional recharge docks driven by a vertical axis innovative mini-wind generator able to produce from 6 to 30 kW. The wind generator is allowed to exchange power with the electric grid implementing the concept of smart grid. The recharge station has been inaugurated by the Italian environment minister July 25-th 2012.

A new Granular Computing approach for sequences representation and classification

In this paper we present an innovative procedure for sequence mining and representation. It can be used as its own in Data Mining problems or as the core of a classification system based on a Granular Computing approach to represent sequences in a suited embedding space. By adopting an inexact sequence matching procedure, the algorithm is able to extract a symbols alphabet of frequent subsequences to be used as prototypes for the embedding stage. Experimental evaluation over both synthetically generated and biological datasets confirms that the modeling system is able to synthesize effective models when facing even complex and noisy problems defined by frequency-based classification rules.