RETRACTED ARTICLE: Rainfall forecast and computer data algorithm optimization in coastal areas based on improved neural network

• Published in: Arabian journal of geosciences 2021, Vol.14 (15), Article 1469
• Main Authors: Wu, Jingzhu, Xing, Xiuzhi
• Format: Article
• Language: English
• Subjects: Algorithms; Big Data; Coastal zone; Coasts; Complexity; Data; Decision making; Disasters; Earth and Environmental Science; Earth science; Earth Sciences; Emergency warning programs; Environment and Low Carbon Transportation; Feature maps; Hurricanes; Load distribution; Mathematical models; Model testing; Natural disasters; Network management systems; Neural networks; Original Paper; Parallel processing; Parameters; Prediction models; Rain; Rainfall; Temporal distribution; Training; Typhoons; Weather
• ISSN: 1866-7511; 1866-7538
• DOI: 10.1007/s12517-021-07579-1

Typhoon weather is often accompanied by heavy rainfall. This phenomenon often brings great inconvenience to people’s lives and also causes natural disasters. According to the basic principles of BP neural network, by organically integrating the characteristics of typhoon weather and rainfall, a BP neural network prediction model for typhoon rainfall can be constructed. The model uses the rainfall recorded by the 201509 typhoon at 88 rainfall observation stations in a specific area for 6 consecutive hours and related characteristic parameters to verify the model. Model testing shows that the relative error of the 6-h rainfall forecast is less than 30%, accounting for 71.1% of the data set. This detection result provides a useful attempt to quickly predict the spatial and temporal distribution of typhoon rainfall, and has important practicability for typhoon disaster warning, loss assessment, and emergency decision-making in coastal areas. At the same time, in order to solve the problems of excessive network redundancy parameters, slow convergence speed, and low parallel efficiency of the parallel DCN algorithm in the big data environment, a parallel deep convolutional neural network optimization algorithm PDCNNO is proposed. First, the algorithm designs a feature map-based pruning strategy (PFM), pre-training the network, and obtains the compressed network, which effectively reduces the redundant parameters and reduces the DCNN training time and space complexity; the control is proposed in the reduce stage. The load balancing strategy (LBRLA) of the load rate obtains the global classification results and realizes the fast and uniform data grouping, thereby improving the speed-up ratio of the parallel system. Experiments show that the algorithm not only reduces the time and space complexity of DCNN training in a big data environment, but also improves the parallelization performance of parallel systems.