基于全球导航卫星系统（GNSS）的定位服务在城市中得到广泛应用，但不同信号遮挡场景下定位精度差异较大，准确快速的场景识别是开发自适应GNSS定位算法的关键。为提高场景识别准确率和鲁棒性，提出一种基于并联长短期记忆网络（PC-LSTM）的GNSS场景识别方法。首先，将城市静态定位环境细分为五类遮挡场景；然后，以原始观测数据和站星几何关系构建特征向量并分析各特征在不同场景识别中的表现；最后，提出一种基于PC-LSTM的GNSS定位场景识别模型。实验结果表明，该模型识别准确率可达99.27%；同时，每轮次训练时间和每样本预测时间低至0.68 s和2.07 ms，在丢失部分观测数据的情况下识别准确率仍可达到最大98.87%，具有良好的实时性和鲁棒性。

Positioning services based on global navigation satellite system(GNSS) are widely applied in cities, but positioning accuracy varies greatly under different signal obstruction scenes. Accurate and fast scene recognition is the key to developing adaptive GNSS positioning algorithms. In order to improve the accuracy and robustness of GNSS scene recognition model, we proposed a new method based on parallel-connected long short-term memory network(PC-LSTM). Firstly, we subdivided the urban static positioning environment into five types of occlusion scenes. Then, we constructed the feature vector based on the original observation data and the station-satellite geometric relationship, and analyzed the performance of each feature in different scenes in detail. Finally, we proposed a GNSS positioning scene recognition model based on PC-LSTM. Experimental results show that the recognition accuracy of this model can reach 99.27%. At the same time, the training time of each epoch and the prediction time of each sample are as low as 0.68 s and 2.07 ms, and the recognition accuracy can still reach a maximum of 98.87% even if part of the observation data is lost, which shows that the proposed model has good real-time performance and robustness.