Decision and prediction support in cyberphysical systems based on computational and machine learning tools

Abstract

Modern trends tend to exploit increased computational power to develop cyber-physical systems that will support or act as decision-makers in complex problems. This Thesis focuses on environmental problems such as energy consumption and water resource management and how automated systems can assist decisions and predictions related to those matters using modern scientific methods such as probabilistic and Machine Learning tools. A key consideration was that pressure on agricultural systems is increasing, and water scarcity is expected to become an important constraint to sustainable development. An increase in agricultural demand and water price is very realistic and will certainly affect the profit, making predictions an even more useful tool in the farming economy. Ultimately, data engineering and Machine Learning methodologies will work combined to produce reliable predictions in the agriculture domain in terms of maximizing estimated yield profit while preserving and well-managing water resources. The utmost target of this research is to develop a Machine Learning framework that performs crop yield predictions, is custom-tailored on farms and focuses in parallel on maintaining low water resources usage. The goal for this framework is to receive datasets related to various features specific to crop farms and then be able to perform preparation and analysis on them, choose the most useful features after examining their correlation and ultimately work on a set of given Machine Learning methods to produce the best prediction possible and optimally adjust to the nature of the data.

Οι σύγχρονες τάσεις τείνουν να εκμεταλλεύονται την αυξημένη υπολογιστική ισχύ για την ανάπτυξη κυβερνοφυσικών συστημάτων που θα υποστηρίζουν ή θα λαμβάνουν αποφάσεις σε πολύπλοκα προβλήματα. Αυτή η διατριβή επικεντρώνεται σε περιβαλλοντικά προβλήματα όπως η κατανάλωση ενέργειας και η διαχείριση των υδατικών πόρων και πώς τα αυτοματοποιημένα συστήματα μπορούν να συνδράμουν σε αποφάσεις και προβλέψεις που σχετίζονται με αυτά τα θέματα χρησιμοποιώντας σύγχρονες επιστημονικές μεθόδους όπως πιθανολογικά εργαλεία και εργαλεία μηχανικής μάθησης. Μία βασική εξέταση ήταν ότι η πίεση στα γεωργικά συστήματα αυξάνεται και η λειψυδρία αναμένεται να αποτελέσει σημαντικό περιορισμό για την αειφόρο ανάπτυξη τους. Η αύξηση της γεωργικής ζήτησης και της τιμής του νερού είναι πολύ ρεαλιστικά και σίγουρα θα επηρεάσουν το κέρδος, καθιστώντας τις προβλέψεις ακόμη πιο χρήσιμο εργαλείο στην αγροτική οικονομία. Τελικά, οι μεθοδολογίες μηχανικής δεδομένων και μηχανικής μάθησης θα λειτουργήσουν συνδυαστικά για να παράγουν αξιόπιστες προβλέψεις στον τομέα της γεωργίας όσον αφορά τη μεγιστοποίηση του εκτιμώμενου κέρδους της γεωργικής απόδοσης με παράλληλη διαφύλαξη και σωστή διαχείριση των υδατικών πόρων. Ο απώτατος στόχος αυτής της έρευνας είναι η ανάπτυξη ενός πλαισίου μηχανικής μάθησης που εκτελεί προβλέψεις απόδοσης των καλλιεργειών, είναι προσαρμοσμένο στις καλλιεργήσιμες εκτάσεις και ταυτόχρονα εστιάζει στη διατήρηση χαμηλών επιπέδων χρήσης υδατικών πόρων. Ο στόχος αυτού του πλαισίου είναι να λαμβάνει σύνολα δεδομένων που σχετίζονται με διάφορα χαρακτηριστικά των γεωργικών εκτάσεων και στη συνέχεια να μπορεί να προετοιμάζει και να αναλύει αυτά τα δεδομένα, να επιλέξει τα πιο χρήσιμα χαρακτηριστικά αφού εξετάσει τη συσχέτισή τους και τελικά να εργασθεί σε ένα σύνολο μεθόδων μηχανικής μάθησης για την εξαγωγή της καλύτερης δυνατής πρόβλεψης και τη βέλτιστη προσαρμογή στη φύση των δεδομένων.

Tendințele moderne tind să exploateze puterea de calcul crescută pentru a dezvolta sisteme ciber-fizice care să sprijine sau să acționeze ca factori de decizie în probleme complexe. Această teză se concentrează pe problemele de mediu, cum ar fi consumul de energie și gestionarea resurselor de apă și modul în care sistemele automatizate pot asista deciziile și predicțiile legate de aceste aspecte folosind metode științifice moderne, cum ar fi instrumentele probabilistice și de învățare automată. Un aspect important la constituit presiunea în creștere asupra sistemelor agricole în care deficitul de apă este de așteptat să devină o constrângere importantă pentru dezvoltarea durabilă. O creștere a cererii de produse agricole precum şi creşterea prețului apei sunt previzibile și cu siguranță vor afecta profitul, făcând predicțiile un instrument și mai util în economia activităţii agricole. Metodele de inginerie a datelor și de învățare automată vor fi utilizate pentru a genera predicții fiabile în domeniul agriculturii în ceea ce privește maximizarea profitului estimat, păstrând și gestionând bine resursele de apă, în acelaşi timp. Obiectivul general al acestei cercetări este dezvoltarea unui cadru de învățare automată care realizează predicții privind randamentul culturilor, care este personalizat pe ferme și se concentrează în paralel pe menținerea consumului scăzut al resurselor de apă. Scopul este de a utiliza seturi de date legate de diverse caracteristici specifice fermelor de cultură și pe baza analizei acelor date, de a alege cele mai utile caracteristici după examinarea corelației lor și, în final, de a aplica metode de învățare automată adecvate pentru a produce cea mai bună predicție posibilă și pentru a se adapta în mod optim la natura datelor.