Prediction of Chlorophyll-a Content Base on Multi-module One Dimensional Convolutional Neural Network.

Aiming at the problem of red tide control in Bohai Bay, a multi-module one dimensional convolutional neural network (M-1DCNN) model was proposed to predict the content of chlorophyll-a in multi-step. Firstly, the dynamic standardization method was used to extract the data waveform characteristics, and then a general module structure suitable for the prediction of chlorophyll-a content in each step was designed, after that through the improved deep deterministic policy gradient (DDPG) algorithm divided the prediction task into multiple modules, meanwhile optimized the parameters of each module, and finally combines all modules to complete the training. The results show that the improved DDPG algorithm can complete the parameter search efficiently and stably, and the prediction results of M-1DCNN in each step are stronger than the comparison model. M-1DCNN has excellent short-term and medium-term prediction capabilities for chlorophyll-a content, which can provide a reference for early warning of eutrophication and red tide phenomena in the Bohai Bay waters.