一种基于BP算法的融合神经网络

针对水电仿真系统水机温度建模中存在非线性动态数学模型问题,提出了一种采用融合神经网络的温度模型·并且为消除应用中神经网络训练速度慢、容易陷入局部极值的影响,采用了可变学习速度的VLBP算法作为更新网络梯度和权值的算法·在该模型的实际应用中,首先设置多个传感器采集温度参数,然后使用采集数据对神经网络进行离线训练,而后使用训练完成的网络对水机温度参数进行实时在线预测·通过现场数据和网络预测数据的对比分析,证明该模型的实际准确率可达96 5%,可以满足实际仿真的要求·

A Fused Neural Network Based on BP Algorithm

A temperature model, as a nonlinear dynamic one, was set up on a basis of fused neural network for the hydroelectric plants. The VLBP (Variable Learning-rate Back Propagation) algorithm was utilized to update network gradients and weight values with the aim of eliminating the slowness in the drill application of the neural network which is easy to get into local extremum. In applications of the model, several temperature-acquisition parameters were got for sensors, then use such parameters to drill off-line the fused neural network. Thus, the real-time on-line forecasts will be available to the temperature parameters of hydroelectric power generator if using the drilled network. Actual accuracy of the model can be up to 96.5%. So, the model can be regarded as meeting the requirements of the simulation.