时滞recurrent神经网络模型的全局渐近稳定性

讨论了时滞recurrent神经网络模型的全局渐近稳定性,通过构造适当的Lyapuov函数,利用线性矩阵不等式,给出了一类常时滞recurrent神经网络的新的充分条件,所获的稳定性条件是时滞相关的,稳定性判别条件更宽松.最后通过一个实例说明方法的可行性.     

基于人工神经网络的金属切削刀具磨损的识别与控制

根据金属切削刀具磨损与切削力的关系，建立了基于人工神经网络的切削刀具磨损的识别与控制系统，初步实现了用人工神经网络对金属切削刀具磨损的动态识别与控制。     

基于小波神经网络的穆棱河作物生育期径流预测

为提高研究精度,在基本的小波神经网络基础上改变隐含层节点数选取方法,降低基本方法随机性和盲目性。将改进的小波网络与人工神经网络进行对比分析。结果表明,小波神经网络改进方法更适合穆棱河流域径流预测。     

城市输配水管网水质变化模型及其数值方法

论述了水质在管网中变化的稳态和动态数学模型，分别给出了求解稳态模型的方法和用拉格朗日法求解动态模型的典型数值方法。     

基于布谷鸟搜索算法优化LSTM的大坝变形预测

对大坝变形进行合理分析和准确预测是确保大坝安全运行的重要手段.大坝变形监测数据具有趋势性、周期性、随机性和非线性等特性,现有的机器学习模型大都基于大坝变形监测数据的非线性特点进行构建,而忽略了监测数据还具有趋势性和周期性的线性特征.提出了一种大坝变形预测模型,通过采用布谷鸟搜索算法(CS)对长短期记忆人工神经网络(LSTM)进行优化,再基于物联网传感器的实时监测数据,使用局部加权回归的周期趋势分解方法(STL)将数据分解成趋势分量、周期分量和余项分量,采用优化后的LSTM模型对趋势分量和余项分量分别进行预测,并通过简单周期估计方法进行计算,将3个分量的预测结果求和后得到最终变形预测结果.试验选取浙江利山水库开展变形预测研究,结果表明：STL-CS-LSTM模型的水平和沉降变形预测精度都高于其他模型,水平位移预测精度由高到低依次为LSTM模型、支持向量回归模型SVR和人工神经网络模型ANN,沉降预测精度由高到低依次为ANN模型、LSTM模型、SVR模型.     

Stimulation of acetylcholinesterase activity by triiodothyronine in the brain of Singi fish,Heteropneustes fossilis (Bloch)

Three consecutive days of injections of triiodothyronine (T₃)(0.038, 0.075, 0.15 and 1.54 nmoles/g) significantly elevated the acetylcholinesterase (AchE) activity in the brain of Singi fish, Heteropneustes fossilis (Bloch). The higher doses of 0.075, 0.15 and 1.54 nmoles of T₃/g induced a greater increase in enzyme activity than 0.038 nmoles/g. A T₃ dose of 0.019 nmoles/g was found to be ineffective. The T₃ action on AchE activity was blocked by cycloheximide. Thiourea treatment for 30 days decreased the AchE activity below the control level. This reduced level of the enzyme activity was brought back even above the control level by T₃ injections. It is, therefore, suggested that thyroid hormone is involved in the sustenance of AchE activity in fish brain.     

Research on immunization strategies based on independent sets of complex networks

According to the concept of independent set, this paper proposes a new immunization strategy of complex networks which immunizes the nodes in an independent set with maximum degree, called targeted immunization in the independent set. When the total degree of immune nodes in independent set is equal to the total degree of targeted immunization nodes in the whole network, the immunization strategy of independent set is more effective than the targeted immunization strategy of the whole network. Based on the network structure, this paper explains the reason. Comparing the random immunization of independent sets with the random immunization in the whole network, test results indicate that more nodes do not be removed with high degree for the random immunization of independent sets. For classic susceptible-infected(SI) model, all nodes in the network are only two states: susceptible or infected. In the study of immunization strategies, comparing SI model with SIR, SIS models. The use of SI model will be more favorable.     

Automatic Fingerprint Classification by GA-Based Neural Network

Fingerprint classification can provide an important indexing mechanism in a fingerprint database. An accurate and consistent classification can greatly reduce fingerprint matching time for large database. In the paper, by combining genetic algorithm and neural network is presented a fingerprint classification algorithm which is able to achieve an accurate classification. By inputting the global feature represented by directional image to three layer neural network trained by genetic algorithm, the fingerprints were classified into six categories: whorl, right loop, left loop, arch, double loop and undiscerning type successfully.     

Artificial Neural Networks Models for Flow Stress during High Temperature Plastic Deformation of Al-Li Alloy

Using gained experimental data to develop the models of stable flow stresses at high temperature plastic deformation by statistical methods for alloy materials, precision of the models is poor and at the same time the processes of modeling are complicated with great workload. On the basis of the data obtained on Gleeble-1500 Thermal Simulator,the predicting models for the relation between stable flow stress during high temperature plastic deformation and deformation strain, strain rate and temperature for 1420 Al-Li alloy have been developed with BP Artificial Neural Network method. The results show that the model on basis of BPNN is practical and it reflects the real feature of the deforming process. It states that the difference between the real value and the output of the model is in order of 5 percent.