IoT based Generator Fuel Level Severity Prediction Using Machine Learning Method for Cell Towers

Abstract

The major infrastructure element in mobile networks is the Base Transceiver Station (BTS), connecting client equipment to the cellular network. As the number of mobile customers and base stations increases, the demand of fuel for the BTS standby generator and installing reservoir fuel tank is also too. However, unexpected fuel running out from BTS is common once the reservoir tank is filled. It exposes the telecom service providers to network interruption when the generator finish fuel. After sensors detect severity levels knowing the magnitude is very important,The severity level of the fuel in the reservoir indicates the effect of unexpected fuel runout and BTS generator shut down. Predicting the stage of fuel level severity is mandatory. Refueling the critical cell towers before it goes to shut down and then can visit the less severed sites after. Different studies have been done on monitoring the fuel level of a BTS, but it is critical to understand the severity state. Fuel level monitoring has been proposed in the past, but not the severity level threshold. Based on a dataset obtained from five BTS fuel tanks, this work proposes BTS fuel level severity prediction which has 500,1000,2000,3375 & 5000 litter volume capacity with various manual gage levels on it,which defines the amount & severity level of fuel in the tank. Data is collected using hardware tools HC-SR04,Arduinomega2560,PS,& data logger and software which integrate the hardware.The datasets were labelled to suitable severity levels using a clustering machine-learning method. The algorithms are then applied to five datasets, namely called BTS1, BTS2, BTS3, BTS4 & BTS5 which has 4884,5208,5208,5112 & 5112 rows of datasets respectively to construct a model and the performance of each algorithm is measured and compared to select a superior model. Based on the accuracy and f measure, the result shows that the RF model outperformed the DT, KNN, and SVM models on each dataset with a maximum testing accuracy of 92-97.4 per cent and f-measure 92 -97 per cent scored. Furthermore, RF performed better in all datasets than DT, KNN and SVM. The study's findings suggest that supervised machine learning can predict BTS fuel level severity stage. Therefore RF model is recommended for BTS fuel level severity prediction.