Environment Feature and Obstacle Position Prediction Using Long Short-Term Memory

Abstract

Congestion management and procedural knowledge require obstacle prediction of network based on large amounts of dataset. Traditional time series forecasting approaches struggle to create effective prediction models since time series analysis in prediction of network traffic is very unstable time parameter and is also non linear in nature, which may cause a very low forecast accuracy. Hence the define usage of LSTM ie. Long Short Term Memory Recurrent Neural Network has been developed as very important alternative for Neural Network (NN) efficiency. The paper proposed to develop an efficient method in combination with genetic algorithm (GA) and of Long Short Term Memory Recurrent Neural Network (LSTMs) in prediction of environment features and positions of obstacles. The combination of both will be comprises of two sections, one with LSTM which is used for feature extraction and GA is used to enhance hyper parameters extracted for the LSTMs networks. The method assumes the higher prediction accuracy as compared to previous research study with decrease in prediction of errors, with categorization of complex changes with considered data; this is done by comparing the ARIMA i.e. Auto Regressive Integrated Moving Average and LSMTs.