基于多策略改进神经网络的力学性能近似估算

Approximate estimation of mechanical properties based on multi-strategy back propagation algorithm

介绍了应用于Q235焊接力学性能的神经网络近似分析方法,开展了焊接力学性能的样本数据对神经网络近似估算的实验研究,为提高估算的准确度,提出了一种新的基于融合多策略混合粒子群优化BP算法(MSBPA)应用于焊接性能数据的非线性映射处理,建立了估算模型。将估算模型应用于Q235焊接力学性能估算分析。为验证算法的有效性,分别运用了融合多策略混合粒子群优化BP算法、遗传算法优化BP算法和传统BP算法对焊接性能近似估算问题进行对比仿真分析。结果表明:基于融合多策略混合粒子群优化BP估算模型对Q235焊接力学性能有较好的非线性拟合能力,估算值与实验值间最大相对误差仅为3. 5%,具有较优的估算准确性。

The approximate analysis method of neural network was applied for the mechanical properties of Q235 welding,and the experiments was carried out to estimate the sample data of welding on the mechanical properties of the neural network approximation.In order to improve the accuracy of estimation,a new multi-strategy hybrid particle swarm optimization back-propagation(BP)algorithm(MSBPA)for nonlinear mapping process of the welding performance data was proposed.The estimation model was applied to estimate the mechanical properties of Q235 welding.In order to verify the effectiveness of the algorithm,the fusion strategy,the hybrid particle swarm optimization BP algorithm,the genetic algorithm optimization BP algorithm and the traditional BP algorithm were used to analyze and compare the approximate estimation of the welding performance.Results showed that the multi-strategy fusion hybrid particle swarm optimization BP estimation model of Q235 welding mechanical properties had a good non-linear fitting capability.The maximum relative error between the estimated values and the experimental ones was only 3.5%,and the proposed method had better estimation accuracy.