Application of Correlated Long-short Term Memory Algorithms for Intelligent Management of Sensors (IMSLSTM)

This study presents an example of connecting relative humidity and temperature using three different Long Short-Term Memory (LSTM) algorithms that are learned and utilized to forecast WSN nodes data.The example predicts relative humidity values as a function of temperature using Stochastic Gradient Descent with Momentum (SGDM), Root Mean Square Propagation (RMSProp), and Adaptive Moment Estimation (ADAM).The results obtained from combining and averaging the three methods are utilized to provide the best possible predicted data.The Root Mean Squared Error (RMSE), which is utilized to enable verifying the correctness of predicted, is evaluated using two learning rate schedules.The two mathematical formulas that represent the expected and intended relationship between temperature and relative humidity are given.With a tiny intercept difference of 0.9, the predicted form function for the three associated LSTM algorithms demonstrated correct representation.This discrepancy is proportionate to the RMSE results that were achieved, allowing for the fine-tuning of the trained algorithms and the verification of the reduction in mathematical expressions that describe the behaviour of Wireless Sensor Network (WSN) nodes.Additionally, a mathematical model of the correlation between RMSE and epochs (number of iterations) is provided to allow for training cycle optimization for improved performance.The strategy of utilizing three distinct LSTM algorithms and comparing and correlating their outputs was demonstrated to be an effective way to lower prediction error and get the best solution.The general correlation mathematical model, which includes weights for each procedure before averaging to get the final value, is described in the discussion.Future advances could use this to incorporate each algorithm's strongest elements.