人工神经网络用于铅的化学形态模拟计算

SIMULATION FOR CHEMICAL SPECUTION OF LEAD BY ARTIFICIAL AERIAL NETWORK

用前馈线性网络法求解水体系中Ｐｂ（２＋）与ＯＨ－之间的反应常数，不同训练算法对求解结果的精度、收敛速度及权值均有影响．结果表明，批处理算法的精度最好，权值不出现负值，但运算时间最长；在线算法的精度虽不如批处理算法，而比数据变换－在线算法好，权值有时会出现负值．运算时间较长；数据变换－在线算法的优点是运算时间短，但相对误差较大，权值出现负值的机会多。采用反馈网络模拟计算铅的各种化学形态的浓度．用物料核算的方法对反馈网络模型进行检验表明，此种模型用于平衡计算是可行的，详细分析了理论模拟和实验曲线的差异的原因，温度的影响最小，在４＜ｐＨ＜９时，ＣＯ有重要的影响．在国代检验时，ｎ值取整所引入的误差的影响亦不可忽视。从本文的结果可以看到，采用前馈网络和反馈网络相结合的方法考察水体中的化学形态是可行的．从而为解决这一类问题提供了一种可能的途径．

Linear feed forward Artificial Neural Network(LANN)is used to calculate reaction coefficient between Pb2+and OH-in aqueous system.In the training process,the different training algedthms(batch data training algorithm,on-line data training algorithm and data preprocession-on-line algor thin directly influence on the predsion of results,convergent time of ANN,size and property of weight.The experimental results show that the best precision can be obtained by batch data training algorithm and weights are always positive while a long convergence time is needed.A better precision can be gotten when on-line algorithm is used.but sometimes negative weights appear and the convergence time is shorter than that by batch data training algorithm,but still longer.When data preprocess-on-line algorithm is adopted,convergence time is the shortest among these algorithms while precision is worse than other algorithms and the negative weight appearS more frequently.The various chemical speciation s of lead can be calculated based on a theoretical madal of Feedback Neural Network(FNN).The model is tested by materials balance and shows that it is possible to use this model to calculate various speeiahons of lead in the condition of chemical equilibrium.Some factors which result in the differences between theoretically simulated and experimental curve are discussed in detaill.It shows that temperature has the least effect on the calculated result.The affect on the simulation result can not be omitted when n becomes an integer.CO in aqueous system is an important factor ranging 4<pH>9.It is possible to calculate chemical spication in equeous system by feedforward combined with feedback neural netwouk and to provide a way to solve this kind of problem.