Routing Vehicles for the Distribution of Postal Parcels in Metropolitan Areas, Using Text Addresses in the Time Windows Requested By Customers

In this research, it was tried to optimize the routing of the mail delivery fleet using text addresses and in different time windows of customers. Therefore, the two main phases include: 1. Geocoding and 2. Routing; it was determined to do this. For this purpose, using natural language processing (NLP) and using machine learning methods, the body of text addresses was extracted cleanly, then using support vector machine (SVM) and random forest methods. (RF), the addresses obtained in the previous step were compared with the addresses in the database, and as a result, the input addresses were matched with the corresponding geographical coordinates (or points) (if any). Then, in the next step, the obtained geographic coordinates were routed using the cold and warm simulation algorithm. By conducting a study on two areas of Tehran (Shaharak Gharb and Monirieh), the accuracy criteria of the geocoding process in SVM and RF methods are 85.7% and 83.2%, respectively, and the accuracy criteria are equal to 93.6% and 91.2%. was obtained. Also, each of the two areas was served with three optimal routes Also, each of the two areas was served with three optimal routes. For Shaharak Gharb, the first, second, and third routes passed 18, 16, and 18 points, respectively, and the travel times of each were106,79, and 104 minutes, respectively. Also, for Monirieh, the first, second, and third routes passed 17, 20, and 19 points, respectively, and the travel times of each were 44, 52, and 43 minutes, respectively