一种面向系统状态参数预报的过程神经网络模型及其算法

A process neural network model oriented to condition parameter prediction and its algorithm

为解决难以用确定机理模型描述的非线性系统状态预报问题,提出一种基于过程神经网络的预报模型及其算法.利用过程神经网络对动态系统的非线性映射机制和直接辨识建模能力,面向系统状态参数预测,建立一种反映系统过程模态特征变化的过程神经网络模型,分析模型的预测机制,给出相应学习算法.为弥补实际采样数据不足及提高数据信息利用率,利用相空间重构方法构造过程神经网络训练函数样本集.以油田开发井组采油速度状态变化预报为例,通过实验验证模型和算法的有效性.

Aiming at condition prediction problems of nonlinear system that are difficult to be described with definite mechanism model, a prediction model and method based on process neural network (PNN) is proposed in the paper. Using the nonlinear mapping mechanism to dynamic system and direct identification modeling ability of PNN, oriented to system state parameter prediction, a process neural network model is built that can reflect modal characteristic change of system process, its prediction mechanism is analyzed, and the learning algorithm is given. At the same time, in order to make up for the deficiency of real sampling data and improve information use ratio, the training sample set of PNN is constructed using phase space reconstruction. Taking state change prediction of well group oil recovery rate in oil field development as example, experiment results proved the effectiveness of the model and the algorithm.