Research On Key Technologies Of Highway Landslide Monitoring And Early Warning Based On BDS

During the construction and operation of highways,if landslides occur in sections near mountains,it is easy to cause damage to the highways and casualties.Against the backdrop of the rapid development of Beidou satellite navigation systems,monitoring landslides by placing Beidou monitoring equipment in areas prone to landslides is an important way to manage landslides.Due to the influence of natural and human factors on landslides in real environments,there are still issues with difficult detection of outliers and insufficient accuracy of prediction models in monitoring data.In response to the above issues,the main research work of this article is as follows:(1)A modified Anomaly Transformer time series unsupervised anomaly detection model is proposed to address the issue of difficulty in extracting feature relationships between landslide data points due to multiple factors such as precipitation and possible reservoir water levels along the way.The model is divided into two parts: one is a prior correlation module,which is based on a learnable Gaussian kernel function.The unimodal nature of the Gaussian kernel allows for more attention to the connections between adjacent points;The other part is the sequence correlation module,which has a built-in self attention mechanism that can effectively extract the connections between each point in the sequence and the entire sequence.By setting a sparsization function,the interference of noise in the calculation process is reduced,the computational complexity is reduced,and relative position representation is introduced in the self attention calculation process to strengthen the model’s attention to the relative relationships between points,improving detection efficiency and accuracy.(2)Aiming at the problems that the traditional static depth learning model does not fully exploit the time series characteristics,ignores the redundancy between features when selecting environmental characteristics,and lacks objectivity when manually selecting the hyperparameter of short-term memory artificial neural network(LSTM)in landslide periodic term displacement prediction,a coupled landslide displacement prediction model is proposed.Firstly,the maximum correlation minimum redundancy algorithm is used to filter out environmental features that have the greatest correlation with periodic term displacement and the least redundancy between features;Then,it is input into the LSTM network,and the optimization algorithm of LSTM hyperparameter,the Tianshu Search Algorithm(BAS),is improved.In the search process,a feedback mechanism is introduced to avoid the problem of being far from the optimal solution in the original algorithm,and the fixed decline factor in the algorithm is changed to a dynamic decline factor to improve the early global and late local optimization capabilities;Finally,the improved BAS algorithm is used to optimize the hyperparameter of LSTM network to obtain the best combination of network parameters and improve the accuracy of network prediction.(3)Based on the above anomaly detection methods and landslide displacement prediction model,a Beidou landslide monitoring and warning system has been designed and implemented.Through requirement analysis of the system’s business,functionality,and performance,as well as research on landslide warning thresholds,a Beidou landslide monitoring and warning system was developed based on the Lay UI front-end framework and the Spring MVC+Spring+Mybatis back-end framework,achieving functions such as data point monitoring,data analysis,and warning release.The system has the advantages of simple operation and strong stability,and has certain practical value.