人工神经网络在水产科学中的应用

人工神经网络（ANN）是一种动态信息（处理）系统，它具有联想记忆、自组织、自适应、自学习和容错等优异的特性而得到广泛应用。ANN已广泛地应用于诸如模式识别、拟合、分类、决策和预测等领域，而水产科学有很多涉及上述技术的问题。本文在简述ANN结构和工作原理的基础上，讨论分析了利用BP神经网络模型、自组织特性神经网络或Kohonen神经网络模型进行分类、模式识别、图像处理和鉴别、预测与评价、系统模拟以及最优化和多目标决策等方面的应用实例。从神经网络模型建模和数据预处理原理研究了应用人工神经网络技术建模的局限性和缺陷。并明确指出：若不采用检验样本监控学习过程，对于一定数量的样本数据，过大的神经网络结构将不可避免地引起对样本数据的过拟合，从而得到了不能正确反映样本数据结构和内在特性和神经网络模型，而可能是对样本数据的噪声的反映。本文最后探讨了人工神经网络技术与模糊数学、逻辑控制和拓扑学以及非确定性原理相结合的应用趋势。

Application of artificial neural networks to fisheries sciences

The topology of artificial neural network ensure its powerful capacities such as association memory, self organization, auto adjustment, self learning and error tolerability. Properly trained ANN model possessed the flexibility as well as high accuracy. ANN, widely used as dynamic information system, was used in various fields such as pattern recognition, fitting,classification and decision making and prediction, so on. There are various issues related to above discussed fields in fisheries sciences. The principle and structure of ANN was discussed in this paper. Some examples such as applying BP (Error Back Propagation algorithm) and SOFM (Self Organization Feature Mapping) or Kohonen model to classification and pattern recognition, image process and identification, prediction and assessment, system simulation as well as optimization and multi objective decision were also introduced. Some restrictions or shortcomings of ANN modeling were also discussed according to the principle of modeling and data preprocessing. Furthermore, without cross verification, a network with a large number of weights and a modest amount of training data can overfit the training data, learning the noise present in the data rather than the underlying structure. The trend of ANN applications was put forward according to combination of fuzzy mathematics, mathematical logic, topology sciences as well as uncertainty.