Bridge construction investment is huge and the service cycle is long. During the service cycle, the bridge structure not only beared the load effect caused by fatigue damage, but also effected by the natural environment and human damage. Beam bridge is the most kind of bridge built on the highway and had a long-term service in China. The main beam of beam bridge is the main load-bearing component. Real-time evaluation of main beam's health degree will greatly improve the safety of highway transportation. Through the rapid assessment of the main beam of the bridge, it can not only directly reflect whether the deflection of the main beam is beyond the dangerous range and the overall condition of the main beam, but also observe the long-term variation rule of the main beam. The current assessment algorithm only stays in the monitoring of whether the deflection of the main beam is beyond the dangerous range, without a complete assessment combined with massive historical data. Based on the theory of Gray Correlation Analysis and combined with the realtime data and historical data of bridge monitoring, we calculate the statistical indicator and morphological indicator of the main beam quickly, and evaluate the comprehensive health indicator of the bridge according to the technical specification in this paper.

In the face of massive information, batch processing of files is an important way of information transmission and storage, and the application is quite common. With the increasing demand for batch files processing reliability and speed, and the problem of low storage efficiency for current batch file processing, the paper proposes a storage method that combine distributed storage system HDFS file storage advantage and Redis cache technology to form a rapid batch merge files. The files that meet the conditions are merged into the Sequence File and stored in the HDFS. The multiple linear regression analysis method is used to determine the load factor, so that the load balancing is adjusted and the Redis cache hash data is used to ensure the efficiency. Through experiments on the corresponding file platform for file upload, query, delete and memory usage, we analysis batch processing method and non-batch method comparatively. It can be concluded that compared with the non-batch direct upload file to HDFS way, improved batch file processing method can process files more faster and ensure the stability and reliability of the file at the same time.

Mining user's learning preference is one of the key issues in the personalized online learning system, which is of great significance technology for modern educational. In this paper, using the hierarchical characteristics of the knowledge points in the course domain, we defined the equivalence relation and equivalence of knowledge points, and defined the structure of the knowledge points quotient space. Then, the functions of support, pheromone concentration and preference were defined on various levels, and an improved ant colony optimization was proposed to handle the multi granularity data structure of quotient space. An algorithm of multi-granularity Learning Preference Mining based on Ant Colony Optimization (ACO-LPM) was proposed to address the problems about too many learning knowledge points and too few user's test data in the online personalized learning system. The pheromone has the characteristic of dynamic evaporation, so, the preference patterns mined by ACO-LPM can be changed with the change of user interest in real time. The experimental results show that the algorithm can mining the user's learning preferences in online learning system effectively and efficiently.