Optimizing the Performance of Kalman Filter and Alpha-Beta Filter Algorithms through Neural Network

In this paper, we have developed two neural network-based algorithms: the neural network-based Kalman filter (KF) algorithm and the neural network-based alpha-beta (α-β) filter algorithm. These algorithms incorporate a neural network to improve prediction accuracy and performance. Both algorithms consist of three input layers (temperature sensor, humidity sensor, and sensor readings), 15 hidden layers, and different output layers. The KF algorithm has a single output layer, while the alpha-beta filter algorithm has two output layers. These output layers dynamically interact with the KF algorithm and α-β filter to predict the final output values. For the KF algorithm, we consider two factors: R computation and the noise factor F. To evaluate the performance of these algorithms, we utilize the root mean square error (RMSE). The sensor readings for both algorithms are relatively high, specifically 5.215. Through the neural network-based alpha-beta filter, we achieved a minimum error of 3.21. In the case of the neural network based Kalman filter, we obtained the best-case result of 2.41 with R=14 and F=0.01. The proposed neural network-based system yields improved results compared to the simple Kalman filter and alpha-beta filter algorithms.