Wi-Fi indoor localization based on long short-term memory neural network model of genetic algorithm
Aiguo Zhang,Zhen Chen,Xinping Jin,Jiahao Hu
DOI: https://doi.org/10.1109/Agro-Geoinformatics59224.2023.10233581
2023-07-25
Abstract:With the continuous improvement of the time needed for people to carry out indoor activities in modern society, the demand for indoor location services is also increasing. The satellite positioning system is one of the most widely used positioning technologies, but in the indoor environment, satellite signals are easily hindered by buildings, GPS positioning error is large, then it is necessary to use other technologies to achieve indoor positioning. For indoor positioning technology, there have been a lot of research results, such as Bluetooth, ultra-wideband, geomagnetic, Wi-Fi (Wireless Fidelity), Zigbee, and PDR (Pedestrian Dead Reckoning) technology. Wi-Fi devices are very popular in many indoor environments today, because most mobile devices can receive Wi-Fi signals, and Wi-Fi Access points (AP) is widely deployed indoors and has the advantages of low cost and easy access to signals. The indoor positioning technology based on Wi-Fi signals has become a hot topic in indoor positioning research. Wi-Fi positioning technology based on location fingerprint is the most widely used Wi-Fi indoor positioning technology. However, due to the complex and changeable indoor environment, signal propagation is affected by many obstacles, which leads to the low positioning accuracy of Wi-Fi-based fingerprint positioning technology. Aiming at Wi-Fi location fingerprint indoor positioning technology requirements, this paper proposes an indoor positioning method based on LSTM (Long Short-Term Memory), which uses the time domain information in the positioning process to improve positioning accuracy. The network consists of a fully connected layer and an LSTM layer, whose input is a time series of RSSI, and the output is an estimate of the position at the last moment in the input sequence. However, using only this method may produce some phenomena such as unbalanced prediction, and fall into local optimal solutions. Due to the existence of the above problems, GA (Genetic Algorithm) is used in the hyperparameter optimization of the LSTM model in this paper. The genetic algorithm takes the maximization of the fitness value as the goal and continuously iterates the population to find the optimal time window width and the optimal hidden layer neural unit. The optimized parameters are substituted into the long short-term memory neural network model, and the conclusion that the LSTM model optimized by the genetic algorithm is more accurate than the LSTM model is obtained. DNN, CNN and other models are also used for comparison, which shows that the prediction effect of the LSTM model is better than that of tother models.
Engineering,Computer Science