Real-Time Optimized Prediction Model for Dissolved Oxygen in Crab Aquaculture Ponds Using Back Propagation Neural Network

摘要

Dissolved oxygen concentration forecast and control is a key point in the process of crab breeding. As the dissolved oxygen in outdoor ponds has low controllability and scalability, this paper proposes a real-time prediction model for dissolved oxygen using the BP neural network. Based on the BP neural network, this model was optimized in the following aspects: model structure, data pretreatment, selection of learning parameters, and removal of zero value influence. These modifications greatly improved the precision and speed of the prediction model. This model was the first to put forward dissolved oxygen prediction in crab aquaculture ponds in real-time; the model has been embedded successfully in the monitoring platform to allow real-time on-line viewing of dissolved oxygen. In addition, this model has been tested and found to be accurate for dissolved oxygen prediction in the demonstration base of breeding crabs.