Abstract: In view of the difficulty in forecasting aquaculture production, a forecasting model of aquaculture production based on Heuristic Johnson Algorithm Optimization and BP Neural Network(BPNN)is proposed in this paper, which is based on the traditional BP Neural Network, with the intention to solve the problems of long time network training and easily being trapped into local optimal solution. It uses Hheuristic Johnson Algorithm to reduce input neuron, and the cut-and-try method to determine the number of hidden layers, thus to construct the Heuristic Johnson Back Propagation Neural Network (HJA-BPNN) Forecasting Model of high precision and high efficiency. The forecasting results of shrimp production in Shandong province, by means of the forecasting model showed that the root of mean square error is smaller than that that by means of traditional BP Neural Network and GM (1,1) forecasting method, and the learning efficiency is improved by comparing with the traditional BP neural network. The study showed that this forecasting model has more advantages in the model construction of a large number of historical data, which can shorten the modeling time and achieve good forecasting results so as to provide a new feasible method for forecasting aquaculture production.

Key words: aquaculture production, forecasting model, BP Neural Network, Johnson Algorithm

摘要： 针对水产养殖产量预测难的现状，提出一种基于启发式Johnson算法优化的反向传播神经网络（BPNN）的产量预测模型。该模型在传统BP神经网络的基础上，针对网络训练时间长、易陷入局部最优的问题，通过启发式Johnson算法降低输入神经元维度，再结合试凑法确定神经网络隐层个数，构建启发式Johnson反向传播神经网络（HJA-BPNN）学习预测模型。实验结果表明，该模型在山东省对虾海水养殖产量预测中，预测的均方根误差小于传统BP神经网络和GM(1,1)，且学习效率相比传统BP神经网络有所提升。研究表明，该学习预测模型在大量历史数据的模型构造上有更大的优势，能够缩短建模时间，同时获得良好的预测效果，为水产养殖产量预测提供了一种可行的新方法。

关键词: 水产养殖产量, 预测模型, BP神经网络, Johnson算法