RDI: A Real-time Decision Support System Applied to Dispatch Decision Problem

Resumo

Several important services with limited resources require uninterrupted support for a vast amount of customers. For example, internet providers, hospitals, distribution utilities and so on. When a customer’s call is received in the proper channels, the reported problem pass through a screening phase in which the operator judge whether or not a support should be sent to the customer. However, not all problems are responsibility of the service provider. For instance, when a distribution utility sends a maintenance team to solve an energy issue that is out of company scope’s, this action generates an improper dispatch problem. Improper dispatches bring high costs regarding fuel and logistics, and can result in heavy penalties to the company. For tackling this problem, we propose RDI, a decision support system that combines supervised machine learning algorithm and model predictive control (MPC) techniques. RDI receives customer’s calls information and recommends when a maintenance team should be dispatched or not. Our first results indicate an assertiveness of 83% in the number of true positives (proper dispatches) and a decrease of 51% in the number of false positives (improper dispatches) within a real dataset from the industry. Moreover, RDI is capable of calculating the associated risk of each occurrence and by predicting changes in the current number of unsolved customer’s calls using a Markov chain model. We show how we built this system, how this solution was applied for diminishing dispatch costs inside a distribution utility and possible directions for further research.