As globalization, intense competition, demand, and supply uncertainty rise, supply chains are becoming more and more information intensive. Thus, Artificial Intelligence could turn into a useful decision-aid tool to facilitate information exchange, agility, and flexibility in the supply chain. However, in this area, the wide potential of AI has not been fully explored yet. Therefore, the objective of the present thesis is to review past and current applications of AI to SCM and to experiment AI on a real case. In order to develop an overall understanding, a systematic literature review has been conducted. Five most interesting areas of AI were identified for further research: Multi-Agent Systems, Genetic Algorithms, Swarm Optimization, Neural Network, Bayesian Network; their application in several temporal and functional dimensions of planning were investigating by collecting, selecting and analyzing a wide range of articles. Furthermore, the applications of these AI techniques, were explored in recently emerged research topics: Supply Chain Risk Management and Human Resources Management. Finally, a benchmark has been conducted considering 6 dimensions, to identify the most promising areas of applications and literature gaps for future research. This review and framework of AI applications to SC planning represent the first contribution of our work. Relying on the knowledge base developed through the literature review, AI was applied to a real case, the eVai case. eVai is an Italian car-sharing company that aims at meeting customers' demand minimizing the costs of moving operators and cars. Neural Networks and Genetic Algorithms were applied to the eVai case, respectively to perform prediction and optimization. After data collection and pre-processing, a neural network model was implemented on Jupiter Lab to predict the arrival station of a potential customer request considering the type of car and the departure station, reaching an accuracy of 56%. Then, a mathematical model was developed, to represent the optimization problem, and a genetic algorithm was implemented using Python, showing good performances in terms of accuracy, efficiency and flexibility. The concurrent application of NN and GA to a such complex real problem represents the second contribution of our work In conclusion, the main achievements of the present research are the development of a performing AI system combining Neural Networks and Genetic Algorithms for prediction and optimization. By focusing on relatively underexplored aspects of research, and addressing real industry problems, the present work widens the range of applications of both GA and NN. Moreover, the most interesting characteristic of the model is its flexibility, which is not only represented by its ability to adapt to problem-specific characteristics, but also to change easily when some data, parameters or starting conditions change through time. However, the main limitations of the present study consist of the lack of data, possible drawbacks from an oversimplification of some aspects of the problem, and lack of past research on the topic.
Per via della globalizzazione, la forte concorrenza, la grande domanda e l'incertezza dell'offerta, le catene di approvvigionamento stanno diventando sempre più competitive e dipendenti da informazioni a tempo reale. Pertanto, l'Intelligenza Artificiale (IA) potrebbe rivelarsi un utile strumento di aiuto decisionale per facilitare lo scambio di informazioni, l'agilità e la flessibilità nella catena di approvvigionamento. Tuttavia, in questo settore, l'ampio potenziale dell'IA non è stato ancora completamente esplorato. Pertanto, l'obiettivo della presente tesi è di valutare le attuali applicazioni dell'IA in SCM e di sperimentare le applicazioni più promettenti dell'IA su un caso reale. Al fine di sviluppare una comprensione generale è stata condotta una revisione sistematica della letteratura. Dopo aver sviluppato una comprensione generale di SCM e IA, cinque aree più interessanti dell'IA sono state identificate per ulteriori ricerche: Sistemi Multi-Agente, Algoritmi Genetici, Ottimizzazione dello Sciame, Rete Neurale, Rete Bayesiana. Inoltre, sono state esplorate le applicazioni di queste tecniche di intelligenza artificiale nelle aree nella gestione della catena di approvvigionamento e gestione delle risorse umane. Infine, è stato condotto un benchmark che considera 6 dimensioni, per identificare le aree più promettenti di applicazioni e lacune nella letteratura per la ricerca futura. Questa panoramica e schematizzazione delle applicazioni di IA a SC planning rappresenta la prima parte del contributo del nostro lavoro. Basandosi sulla comprensione sviluppata attraverso lo studio della letteratura, l'IA è stata applicata al caso eVai. eVai è una società italiana di car sharing che mira a soddisfare la domanda dei clienti minimizzando i costi di spostamento di operatori e auto. reti neurali e algoritmi genetici sono stati applicati al caso eVai, per predizione e ottimizzazione, rispettivamente. Dopo che i dati sono state raccolti e pre-processati, un modello di Neural Network è stato implementation in Jupiter Lab per predirre la stazione di arrivo di potenziali richeste di clienti, considerando il tipo di vettura e la stazione di partenza, e raggiungendo un’accuratezza del 56%. Successivamente, un modello matematico è stato definito per rapprentare il problema di ottimizzazione, e un algoritmo genetico è stato sviluppato usando Python, mostrando buene perfromance in termini di accuratezza, efficienza e flessibilità. L’applicazione di NN e GA a un problema reale e altamente complesso rappresenta il secondo contributo del nostro lavoro. In conclusione, i principali risultati della presente ricerca sono lo sviluppo di un sistema di intelligenza artificiale performante che combina reti neurali e algoritmi genetici per la previsione e l'ottimizzazione. Concentrandosi su aspetti relativamente poco esplorati della ricerca e affrontando problemi reali del settore. Il presente lavoro amplia la gamma di applicazioni di GA e NN. Inoltre, la caratteristica più interessante del modello è la sua flessibilità, che non è solo rappresentata dalla sua capacità di adattarsi alle caratteristiche specifiche del problema, ma anche di cambiare facilmente quando alcuni dati, parametri o condizioni inizialmente impostano cambiamenti nel tempo. Tuttavia, i limiti principali del presente studio consistono nella mancanza di dati, nei possibili svantaggi della semplificazione eccessiva di alcuni aspetti del problema e nella mancanza di ricerche passate sull'argomento.
Review of AI in SC planning and application of NN and GA to a complex transportation planning problem : the case of eVai
CHIOCCHIA, IRENE;Chen, Rosalina
2019/2020
Abstract
As globalization, intense competition, demand, and supply uncertainty rise, supply chains are becoming more and more information intensive. Thus, Artificial Intelligence could turn into a useful decision-aid tool to facilitate information exchange, agility, and flexibility in the supply chain. However, in this area, the wide potential of AI has not been fully explored yet. Therefore, the objective of the present thesis is to review past and current applications of AI to SCM and to experiment AI on a real case. In order to develop an overall understanding, a systematic literature review has been conducted. Five most interesting areas of AI were identified for further research: Multi-Agent Systems, Genetic Algorithms, Swarm Optimization, Neural Network, Bayesian Network; their application in several temporal and functional dimensions of planning were investigating by collecting, selecting and analyzing a wide range of articles. Furthermore, the applications of these AI techniques, were explored in recently emerged research topics: Supply Chain Risk Management and Human Resources Management. Finally, a benchmark has been conducted considering 6 dimensions, to identify the most promising areas of applications and literature gaps for future research. This review and framework of AI applications to SC planning represent the first contribution of our work. Relying on the knowledge base developed through the literature review, AI was applied to a real case, the eVai case. eVai is an Italian car-sharing company that aims at meeting customers' demand minimizing the costs of moving operators and cars. Neural Networks and Genetic Algorithms were applied to the eVai case, respectively to perform prediction and optimization. After data collection and pre-processing, a neural network model was implemented on Jupiter Lab to predict the arrival station of a potential customer request considering the type of car and the departure station, reaching an accuracy of 56%. Then, a mathematical model was developed, to represent the optimization problem, and a genetic algorithm was implemented using Python, showing good performances in terms of accuracy, efficiency and flexibility. The concurrent application of NN and GA to a such complex real problem represents the second contribution of our work In conclusion, the main achievements of the present research are the development of a performing AI system combining Neural Networks and Genetic Algorithms for prediction and optimization. By focusing on relatively underexplored aspects of research, and addressing real industry problems, the present work widens the range of applications of both GA and NN. Moreover, the most interesting characteristic of the model is its flexibility, which is not only represented by its ability to adapt to problem-specific characteristics, but also to change easily when some data, parameters or starting conditions change through time. However, the main limitations of the present study consist of the lack of data, possible drawbacks from an oversimplification of some aspects of the problem, and lack of past research on the topic.| File | Dimensione | Formato | |
|---|---|---|---|
|
Review of AI in SC Planning and application of NN and GA to a complex transportation planning problem, the case of Evai (1).pdf
solo utenti autorizzati dal 09/07/2023
Descrizione: Relying on the knowledge base developed through the literature review, AI was applied to a real case, the eVai case. eVai is an Italian car-sharing company that aims at meeting customers' demand minimizing the costs of moving operators and cars. Neural Networks and Genetic Algorithms were applied to the eVai case, respectively to perform prediction and optimization. .The main achievements of the present research are the development of a performing AI system combining Neural Networks and Genetic Algorithms for prediction and optimization
Dimensione
3.46 MB
Formato
Adobe PDF
|
3.46 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/166909