带种族遗传算法的神经网络辨识器及其应用

为了解决油田动态系统建立数字模型难和模型训练难的问题,以多层动态前向网络为模型框架,给出了一种新的油田系统神经网络辩识器,它对上述系统具有通用性。为了训练该辩识器和避免极值问题,模拟生物种族进化机制,设计了新的种族遗传算法。该算法是由改进的带共享机制的遗传算法、启发式山峰聚类算法和新的Powell算法有机结合构成的新型混合算法。应用表明,种族遗传算法具有极强的全局和局部搜索能力,这确保了上述神经网络辩识器具有很高的预测精度,平均相对误差在1%以内。

Neural Network Identifier with Species Genetic Algorithms and Its Application

It is difficult to build digital model and train the model of the oil field dynamic systems.Using multilayer dynamic forward networks as model frame,a new neural network identifier is given.It has universal property to oil field systems.To train the identifier and overcome its local minimum problem,simulating living species evolution principle,a new species genetic algorithm is designed.It consists of improving genetic algorithm with sharing,heuristic mountain clustering algorithm and new Powell algorithm.The application result shows that the species genetic algorithm designed has great global and local search ability,which assures that the identifier has high prediction precision.Its average relative errors are less than 1 %.