过程神经网络模型结构复杂,正交基展开后学习参数多,传统梯度下降存在对初值敏感、计算复杂等问题,将过程神经网络进行正交基展开化简,在结构上转化为统神经网络,利用极限学习作为过程神经元网络的学习算法.学习过程中摒弃梯度下降算法的迭代调整策略,采用 Moore-Penrose 广义逆计算输出权值,同时为弥补该算法由于随机赋值造成的模型稳定性差的缺点,提出一种双链结构的量子粒子群算法,优化极限学习过程中随机赋值参数.二者结合使用,使模型在稳定性、训练误差方面都得到了一定程度的提高.仿真实验以Mackey-Glass时间序列和太阳黑子预测为例验证了算法的有效性.“,”The process neural network model structure is complex and has many learning parameters after the orthogonal basis expansion. The traditional gradient descent algorithm is sensitive to initial values, complicated in computation. The process neural network model is simplified through orthogonal basis expansion in this paper. The process neural network is converted into the traditional neural network. A new process neural network training algorithm based on extreme learning machine is presented in this paper. The iterative adjustment strategy is rejected in the training process and Moore-Penrose is used to calculate the output weight. In order to cover the poor stability produced by the random assignment, the quantum particle swarm algorithm is taken and the model parameters are optimized with its global search ability. The model and algorithm are applied to the application examples of Mackey-Glass chaotic time series and sunspot prediction. The simulation results confirm the validity and feasibility of the learning algorithm.