Gabriele Maria Lozito

Swarm/flock optimization algorithms as continuous dynamic systems

A new general typology of optimization algorithms, inspired to classical swarm intelligence, is presented. They are obtained by translating the numerical swarm/flock-based algorithms into differential equations in the time domain and employing analytical closed-forms written in the continuum. The use of circulant matrices for the representation of the connections among elements of the flock allowed us to analytically integrate the differential equations by means of a time-windowing approach. The result of this integration provides functions of time that are closed-forms, suitable for describing the trajectories of the flock members: they are directly used to update the position and the velocity of each bird/particle at each step (time window) and consequently they substitute in the continuous algorithm the classical updating rules of the numerical algorithms. Thanks to the closed forms it is also possible to analyze the effects due to the tuning of parameters in terms of exploration or exploitation capabilities. In this way we are able to govern the behavior of the continuous algorithm by means of non stochastic tuning of parameters. The proposed continuous algorithms have been validated on famous benchmark functions, comparing the obtained results with the ones coming from the corresponding numerical algorithms.

On training efficiency and computational costs of a feed forward neural network: a review

A comprehensive review on the problem of choosing a suitable activation function for the hidden layer of a feed forward neural network has been widely investigated. Since the nonlinear component of a neural network is the main contributor to the network mapping capabilities, the different choices that may lead to enhanced performances, in terms of training, generalization, or computational costs, are analyzed, both in general-purpose and in embedded computing environments. Finally, a strategy to convert a network configuration between different activation functions without altering the network mapping capabilities will be presented.

Dynamic hysteresis modelling of magnetic materials by using a neural network approach

The modelling of the dynamic behavior of hysteretic materials and devices must take into account magnetodynamic effects. In the present paper these tasks are simultaneously modelled by means of an ad-hoc Neural System (NS) based on an array of 3-input 1-output Feed Forward NNs. Each NN is aimed to a particular typology of the excitation field (prediction of flux density from a known waveform of the magnetic field strength or vice-versa) and manages just a fixed portion of the dynamic hysteresis loop. The whole hysteretic curve is simulated by linking the evaluations made by different NNs of the NS. The NS is able to perform the simulation of any kind of dynamic loop (saturated and non-saturated, symmetric or asymmetric) generated by any assigned arbitrarily distorted excitations into a fixed range of frequencies. Numerical validations are presented both on a virtual magnetic device and on a non-oriented Fe-(3 wt%) Si laminations (thickness ~0.35 mm).

Microcontroller based maximum power point tracking through FCC and MLP Neural Networks

This paper covers the study towards the implementation of a Neural Network based approach for the efficiency control of Photovoltaic systems. The algorithm aims to track the maximum power point for the PV device whenever abrupt changes in climatic conditions occur. The core of the algorithm is a Neural Network (NN) trained by using a suitable mathematical model of the photovoltaic device. Different NN architectures and several optimization solutions were studied to investigate the best approach in terms of computational efficiency, memory footprint and prediction accuracy. The best found architecture has been implemented and tested on the microcontroller unit LM4F120H5QR by Texas Instruments by using a prototype board to drive the operating point of a low-power solar cell.

An Efficient Architecture for Floating Point Based MISO Neural Neworks on FPGA

The present paper documents the research towards the development of an efficient algorithm to compute the result from a multiple-input-single-output Neural Network using floating-point arithmetic on FPGA. The proposed algorithm focus on optimizing pipeline delays by splitting the Multiply and accumulate algorithm into separate steps using partial products. It is a revisit of the classical algorithm for NN computation, able to overcome the main computation bottleneck in FPGA environment. The proposed algorithm can be implemented into an architecture that fully exploits the pipeline performance of the floating-point arithmetic blocks, thus allowing a very fast computation for the neural network. The performance of the proposed architecture is presented using as target a Cyclone II FPGA Device.

Very Fast and Accurate Procedure for the Characterization of Photovoltaic Panels from Datasheet Information

In recent years several numerical methods have been proposed to identify the five-parameter model of photovoltaic panels from manufacturer datasheets also by introducing simplification or approximation techniques. In this paper we present a fast and accurate procedure for obtaining the parameters of the five-parameter model by starting from its reduced form. The procedure allows characterizing, in few seconds, thousands of photovoltaic panels present on the standard databases. It introduces and takes advantage of further important mathematical considerations without any model simplifications or data approximations. In particular the five parameters are divided in two groups, independent and dependent parameters, in order to reduce the dimensions of the search space. The partitioning of the parameters provides a strong advantage in terms of convergence, computational costs, and execution time of the present approach. Validations on thousands of photovoltaic panels are presented that show how it is possible to make easy and efficient the extraction process of the five parameters, without taking care of choosing a specific solver algorithm but simply by using any deterministic optimization/minimization technique.

Two FPGA-Oriented High-Speed Irradiance Virtual Sensors for Photovoltaic Plants

Knowing solar irradiance value allows an optimized management of photovoltaic (PV) power plants in terms of produced energy. Unfortunately, although sensing temperature is easy, the measurement of solar irradiance is expensive. In this paper, two circuit architectures for the estimation of the solar irradiance based on simple measurements are proposed. They are thought to be part of a centralized system implemented on field programmable gate array (FPGA) for sensing and monitoring of solar irradiance in a whole PV plant. The FPGA centralized architecture could allow for a real-time irradiance mapping by exploiting information coming from several low-cost measuring circuits suitably allocated on the PV modules. Validations on real irradiance data collected by the U.S. Department of Energy’s National Renewable Energy Laboratory are presented.

On the generalization capabilities of the ten-parameter jiles-atherton model

This work proposes an analysis on the generalization capabilities for the modified version of the classic Jiles-Atherton model for magnetic hysteresis. The modified model takes into account the use of dynamic parameterization, as opposed to the classic model where the parameters are constant. Two different dynamic parameterizations are taken into account: a dependence on the excitation and a dependence on the response. The identification process is performed by using a novel nonlinear optimization technique called Continuous Flock-of-Starling Optimization Cube (CFSO3), an algorithm belonging to the class of swarm intelligence. The algorithm exploits parallel architecture and uses a supervised strategy to alternate between exploration and exploitation capabilities. Comparisons between the obtained results are presented at the end of the paper.

Implementation of a neural MPPT algorithm on a low-cost 8-bit microcontroller

This work proposes a maximum power point tracking algorithm based on neural networks embedded in a low-cost 8-bit microcontroller. The obtained device can correctly track the maximum power point even under abrupt changes in solar irradiance and improves the dynamic performance of the power converter that connects photovoltaic power plants into the ac grid. Indeed, traditional maximum power point tracking algorithms such as “perturb & observe” and “incremental conductance” are able to track the point of maximum power in most cases but they can fail under rapidity changing atmospheric conditions. The use of a microcontroller allows for easy updates and enhancement by simply adding code libraries. Furthermore, it can be interfaced via standard communication means to other control devices, integrated into control schemes and remote-controlled through its embedded web server. The proposed approach has been validated through experimental and simulated results.

An empirical investigation on the static Jiles-Atherton model identification by using different set of measurements

Since the first presentation of the static Jiles-Atherton (JA) model, several authors discussed about its identification process and many of them have emphasized that the JA model performs poorly on waveforms different from the one used for its identification. As a consequence, this problem is usually approached by modifying the original model, or by introducing new parameters, or making the parameters dependent on the field intensity, using innovative optimization algorithms. The aim of this paper is to empirically demonstrate that a relationship between the found parameters and the identification process exists. Indeed, we prove that more than one numerical solution can reproduce the same saturated hysteresis loop under a quite low threshold error, and a wrong choice of data utilized for finding the parameters drastically affects the final results. Different shapes of hysteresis loops have been used as identification patterns and the generalization capabilities of the model has been inspected on different distorted excitation waveforms used for validation. The obtained results confirm that the simulation capabilities of the static JA model improve if a suitable choice of data is made during the identification process.