幂激励前向神经网络改进下的瓦斯涌出量预测

Prediction of gas emission quantity based on improvement of power-activation feed-forward neural network

结合SPSS软件的最大方差旋转的因子分析法,设计出依据较少数据进行扩充丰富的随机调和算法,改进了双输入幂激励前向神经网络.该算法有效地解决了幂激励前向神经网络在采样数据较少情况下预测精度偏低的问题,改进的双输入幂激励前向神经网络需要利用权值直接确定法和最优结构法确定最优结构,然后利用随机调和算法在有限采样数据下生成大量训练数据,随之确定最终网络的最优权值,最后在给定次数的循环下确定验证数据的预测值.数值仿真结果表明该算法具有较高的预测精度.

Combined with the SPSS of the maximum rotation factor analysis of variance, a random harmonic al- gorithm was constructed for few data to rich, which improved double input power-activation feed-forward neural network. The algorithm effectively solved the power-activation feed-forward neural network in sampling data under less problem of low prediction accuracy. The optimal structure was determined based on the direct- weight-determination and optimal structure of double input power-activation improved feed-forward neural net- work. Then used the random harmonic algorithm to generate a large number of training data in limited sam- pling data, and determined the optimal weights of final network. Finally, the forecast verification data value in the given number of cycles was received. The results of numerical simulation showed that the algorithm has higher prediction accuracy.