跨越—侧抑制神经网络序列学习BOD特征建模

Biochemical Oxygen Demand Characteristic Modelling Based on Span-lateral Inhibition Neural Network Sequential Learning

摘要: 通过对跨越—侧抑制神经网络（S-LINN）的权值连接分析，设计一种跨越输出权值——隐含层前馈连接权值分别训练的序列学习（SFSL）方法，并结合S-LINN的BOD（biochemical oxygen demand）特征建模方法，实现其在线预测．权值序列学习方法能够实现网络权值的快速收敛，进一步提高网络的学习性能．实验研究表明，S-LINN的特征建模能够实现BOD的准确预测，并且SFSL对网络的学习精度和泛化能力均有明显提升．

Abstract: Based on the analysis of the weight connection of the span-lateral inhibition neural network (S-LINN), a sequence learning approach (SFSL) is proposed on the basis of separate training of the span-output weights and the hidden-layer-feedforward weights during the learning process. By combining the intelligent characteristic biochemical oxygen demand (BOD) modeling of the S-LINN, the method can realize forecasting values online. The new learning strategy not only accelerates the convergence of weights but also improves the performance of the S-LINN. Experiment results show that the proposed S-LINN sequential learning characteristic modeling approach can achieve high-accuracy BOD prediction and that the SFSL improves the approximation and generalization abilities of the S-LINN.