基于粒子群算法优化BP神经网络漏钢预报的研究

Breakout Prediction Based on BP Neural Network with PSO Algorithms in Continuous Casting Process

针对标准BP神经网络中收敛速度慢以及易陷入局部最优解等问题,利用粒子群算法的全局搜索性,将粒子群算法应用到BP神经网络训练中建立了PSO-BP神经网络模型,结果表明改进模型不仅可以克服传统BP网络收敛速度慢和易陷入局部权值的局限问题,而且很大程度地提高了结果精度和BP网络学习能力,将此模型应用到结晶器漏钢预报系统中,并用某钢厂采集到的历史数据对该模型进行训练与测试,与标准BP神经网络测试结果进行分析与比较,实验表明PSO-BP网络模型预报更加实时、准确,具有很好的应用前景。

For the standard BP neural network has the limitations of slow convergence and local extreme values, using the global property of PSO, PSO algorithm optimization was used in BP neural network. So a PSO--BP neural network was established. The result shows that the improved algorithm can not only overcome the limitations in hoth the slow convergence and the local extreme values of traditional BP algorithm, but also improve the precision of the result and the learning ability greatly, and then it was introduced into the breakout prediction system. This model was trained and tested with the historical data collected from a steel plant, analysis and comparison of the two models. The results show that PSO--BP neural network predictes more real--time, exact and it has good antieipant application.