A CS-LSTM Model-Based Spoofing Prediction Method for GNSS-Based Train Positioning

In modern intelligent railway systems, train positioning based on Global Navigation Satellite System (GNSS) has become an important technical direction, especially for the novel train control systems. Considering the vulnerability of GNSS under the possible spoofing conditions with great threat to the GNSS-based train positioning, the solution to enhance the active protection capability against the spoofing attack is investigated. In this paper, a novel data-driven Long Short-Term Memory (LSTM) model-based prediction solution is proposed to predict the possible GNSS spoofing attack trend, where enhancement by the Cuckoo Search (CS) strategy is introduced into the LSTM scheme to optimize the determination of the hyperparameters. The proposed CS-LSTM solution is used to establish a spoofing prediction model using field GNSS positioning data through a spoofing-injection test platform under different frequency sine wave spoofing attacks. Synchronous prediction of multiple feature entities is carried out to reflect the trend of a possible spoofing attack. The results demonstrate that the proposed CS-LSTM-based solution improves the ability of off-line model training to search for hyperparameters, based on which this solution achieves an enhanced performance level over other methods. The prediction and comparison illustrate the great potentials in identifying the developing effects of GNSS spoofing at an early stage and triggering reasonable countermeasures.