Accurate timing is one of the key features of the Global Positioning System (GPS), which is employed in many critical infrastructures. Any imprecise time measurement in GPS-based structures, such as smart power grids, economic activities, and communication towers, can lead to disastrous results. The vulnerability of the stationary GPS receivers to the time synchronization attacks (TSAs) jeopardizes the GPS timing precision and trust level. In the past few years, studies suggested the adoption of estimators to follow the authentic trend of the clock offset information under attack conditions. However, the estimators would lose track of the authentic signal without proper knowledge of the signal characteristics. Therefore, a multi-layer perceptron neural network (MLP NN) is proposed to follow the trend of the data. The main difference between the proposed method and typical estimators is the reliance of the network on the training information consisting of signal features. The proposed MLP NN performance has been evaluated through two real-world datasets and two well-known types of TSA. The root mean square error results exhibit an improvement of at least six times compared to other conventional and state-of-art methods.

准确的时间安排是全球定位系统（GPS）的关键功能之一，该系统已在许多关键基础架构中使用。基于GPS的结构（例如智能电网，经济活动和通讯塔）中任何不精确的时间测量都可能导致灾难性的结果。固定GPS接收器容易受到时间同步攻击（TSA）的影响，从而危害了GPS的计时精度和信任度。在过去的几年中，研究建议采用估计器来跟踪攻击条件下时钟偏移信息的真实趋势。但是，在没有适当了解信号特征的情况下，估计器会丢失对真实信号的跟踪。因此，提出了一种多层感知器神经网络（MLP NN）来跟踪数据趋势。所提出的方法与典型估计器之间的主要区别是网络对包含信号特征的训练信息的依赖。拟议的MLP NN性能已通过两个真实世界的数据集和两种众所周知的TSA类型进行了评估。与其他传统和最新方法相比，均方根误差结果显示至少提高了六倍。