遗传模拟退火的BP算法在冲击地压中的应用
|
摘要
冲击地压的预测、预报的研究,大多数仍停留在简单的统计研究和单因素的预测方面,因而,结果也不十分理想。笔者采用多层前向网络对该问题进行数学建模,网络的训练算法采用基于遗传模拟退火的BP优化算法。该算法是在遗传算法中引入模拟退火机制,将其同BP算法结合,形成一个混合的优化算法。新算法既有神经网络的学习能力和鲁棒性,又有遗传算法的强全局随机搜索能力。同时,利用华丰矿冲击地压的实际监测数据,通过遗传算法的主要性能指标对新算法的参数进行了比较研究,得到优化后的一组参数。利用该参数,对冲击地压的神经网络模型的结构、权值和阈值进行了优化,得到了非全连接的优化神经网络模型。最后,利用该模型对华丰矿冲击地压进行了短期最大震级的预报。预测结果的相对误差率平均为 7.84 %,预测效果比较理想。
The study of the forecast on rockburst mostly remains at simple statistical study and the forecast of single factor, so the result is not satisfying. This thesis realizes rockburst mathematical model by adopting multilayer forward network and optimization BP algorithm based on genetic simulated annealing algorithm which has been used in the network training perceptrons. This optimization BP algorithm introduces simulated annealing mechanism to genetic algorithm and have it combined with multilayer perceptrons to form a compounding optimization perceptrons. The new perceptrons have both the learning ability of the neural network and robustness, and strong random searching ability of genetic algorithm. At the same time, this thesis makes a contrast study on the parameters of genetic simulated annealing algorithm by the principal performance index of genetic algorithm by using the practical monitoring data of Huafeng Mine rockburst and obtains a set of optimized parameters. By using those parameters, the structure weights and threshold value of the rockburst neural network model is optimized and obtains non-all-connected optimized neural network model. At last, this thesis carries out the short-term maximum magnitude of earthquake on Huafeng Mine rockburst by using this model. The relative error of the forecast is 7.84 % averagely. The forecast effect is comparatively ideal.
The study of the forecast on rockburst mostly remains at simple statistical study and the forecast of single factor, so the result is not satisfying. This thesis realizes rockburst mathematical model by adopting multilayer forward network and optimization BP algorithm based on genetic simulated annealing algorithm which has been used in the network training perceptrons. This optimization BP algorithm introduces simulated annealing mechanism to genetic algorithm and have it combined with multilayer perceptrons to form a compounding optimization perceptrons. The new perceptrons have both the learning ability of the neural network and robustness, and strong random searching ability of genetic algorithm. At the same time, this thesis makes a contrast study on the parameters of genetic simulated annealing algorithm by the principal performance index of genetic algorithm by using the practical monitoring data of Huafeng Mine rockburst and obtains a set of optimized parameters. By using those parameters, the structure weights and threshold value of the rockburst neural network model is optimized and obtains non-all-connected optimized neural network model. At last, this thesis carries out the short-term maximum magnitude of earthquake on Huafeng Mine rockburst by using this model. The relative error of the forecast is 7.84 % averagely. The forecast effect is comparatively ideal.
引文
[1]RumelhartD E,HintonWilliamsR.Learning representation by backpropagation errors[J].Nature,1986,(323):533-536.
[2]王虎栓.基于人工神经网络的峰值地震动物理参数的智能判别[J].地震学报,1993,15(2):208-216.
[3]王炜,蒋春曦,张军等.BP神经网络在地震综合预报中的应用[J].地震,1999,19(2):118-126.
[4]OctavianStan,EdwardKamen.A local linearized least squares alogorithm for training feedforward neural networks[J].Institution ofElectricalElectronics,TransationNeuralNetwork,2000,11(2):487-495.
[5]AbidS,FnaiechF,NajimM.A fast feedforward training algorithm using a modified form of the standard backpropagation algorithm[J].Institution ofElectricalElectronics,TransationNeuralNetwork,2001,12(2):424-429.
[6]陈荣.多层前向网络的研究——遗传BP算法和结构优化策略[J].自动化学报,1997,23(1):43-49.
[7]曹先彬,郑振,范磊等.融合神经网络的一种改进遗传算法[J].模式识别与人工智能,1999,12(4):486-492.
[8]黎明,严超华,刘高航.遗传算法优化前向神经网络结构和权重矢量[J].中国图象图形学报,1999,4(6):491-495.
[2]王虎栓.基于人工神经网络的峰值地震动物理参数的智能判别[J].地震学报,1993,15(2):208-216.
[3]王炜,蒋春曦,张军等.BP神经网络在地震综合预报中的应用[J].地震,1999,19(2):118-126.
[4]OctavianStan,EdwardKamen.A local linearized least squares alogorithm for training feedforward neural networks[J].Institution ofElectricalElectronics,TransationNeuralNetwork,2000,11(2):487-495.
[5]AbidS,FnaiechF,NajimM.A fast feedforward training algorithm using a modified form of the standard backpropagation algorithm[J].Institution ofElectricalElectronics,TransationNeuralNetwork,2001,12(2):424-429.
[6]陈荣.多层前向网络的研究——遗传BP算法和结构优化策略[J].自动化学报,1997,23(1):43-49.
[7]曹先彬,郑振,范磊等.融合神经网络的一种改进遗传算法[J].模式识别与人工智能,1999,12(4):486-492.
[8]黎明,严超华,刘高航.遗传算法优化前向神经网络结构和权重矢量[J].中国图象图形学报,1999,4(6):491-495.
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心