基于启发式Johnson算法优化BP神经网络的水产养殖产量预测模型

针对水产养殖产量预测难的现状,提出一种基于启发式Johnson算法优化的反向传播神经网络(BPNN)的产量预测模型。该模型在传统BP神经网络的基础上,针对网络训练时间长、易陷入局部最优的问题,通过启发式Johnson算法降低输入神经元维度,再结合试凑法确定神经网络隐层个数,构建启发式Johnson反向传播神经网络(HJA-BPNN)学习预测模型。实验结果表明,该模型在山东省对虾海水养殖产量预测中,预测的均方根误差小于传统BP神经网络和GM(1,1),且学习效率相比传统BP神经网络有所提升。研究表明,该学习预测模型在大量历史数据的模型构造上有更大的优势,能够缩短建模时间,同时获得良好的预测效果,为水产养殖产量预测提供了一种可行的新方法。     

Developing a reliable strategy to infer the effective soil hydraulic properties from field evaporation experiments for agro-hydrological models

The Richards equation has been widely used for simulating soil water movement. However, the take-up of agro-hydrological models using the basic theory of soil water flow for optimizing irrigation, fertilizer and pesticide practices is still low. This is partly due to the difficulties in obtaining accurate values for soil hydraulic properties at a field scale. Here, we use an inverse technique to deduce the effective soil hydraulic properties, based on measuring the changes in the distribution of soil water with depth in a fallow field over a long period, subject to natural rainfall and evaporation using a robust micro Genetic Algorithm. A new optimized function was constructed from the soil water contents at different depths, and the soil water at field capacity. The deduced soil water retention curve was approximately parallel but higher than that derived from published pedo-tranfer functions for a given soil pressure head. The water contents calculated from the deduced soil hydraulic properties were in good agreement with the measured values. The reliability of the deduced soil hydraulic properties was tested in reproducing data measured from an independent experiment on the same soil cropped with leek. The calculation of root water uptake took account for both soil water potential and root density distribution. Results show that the predictions of soil water contents at various depths agree fairly well with the measurements, indicating that the inverse analysis is an effective and reliable approach to estimate soil hydraulic properties, and thus permits the simulation of soil water dynamics in both cropped and fallow soils in the field accurately.     

基于遗传算法的投影寻踪回归模型在参考作物滕发量预测中的应用

根据朝阳气象站的实测气象数据(降水量、温度、湿度、风速、蒸发量、日照时数),提出基于加速遗传算法的投影寻踪回归模型用于预测朝阳地区参考作物滕发量.用自相关分析技术确定预测因子(气象数据),构造出新的投影指标,采用可变阶的正交Hermite多项式拟合岭函数,并用基于实数编码的加速遗传算法优化投影指标函数,进而构建投影寻踪...     

Inverse Modeling to Quantify Irrigation System Characteristics and Operational Management

Remotely sensed (RS) data is a major source to obtain spatialdata required for hydrological models. The challenge for thefuture is to obtain besides the more direct observable data(landcover, leaf area index, digital elevation model andevapotranspiration), non-visible data such as soilcharacteristics, groundwater depth and irrigation practices.In this study we have explore the option of using inversemodeling to obtain these non-RS-visible data. For a commandarea in Haryana, India, we applied for the 2000–2001 rabiseason a RS-GIS-combined inverse modeling approach to derivenon-RS-visible data required in the regional application ofhydrological models. A Genetic Algorithm loaded stochasticphysically based soil-water-atmosphere-plant model (SWAP) wasdeveloped for the inverse problem and used in the study. Theresults showed good agreement with the inventoried data suchas soil hydraulic properties, sowing dates, groundwaterdepths, irrigation practices and water quality. The deriveddata could be used to predict the state of the system at anytime in the cropping season, which can be used to evaluateoperational management strategies.     

遗传算法在离心泵叶片优化设计中的应用

对遗传算法在流体机械优化设计中的应用现状作了较为全面的总结；提出了将遗传算法应用于离心泵叶片优化设计中，分析了遗传算法在离心泵优化设计中应用的可行性以及存在的问题；对离心泵优化设计的发展提出了一些新建议和尝试。     

车辆半主动悬架新型控制算法的研究

车辆1/4悬架模型不能反映车辆的俯仰情况,且一般控制算法多以簧载质量加速度作为反馈,具有一定的局限性。为更好地改善悬架性能,本文建立了车辆1/2半主动悬架4自由度模型,针对悬架为一非线性、有时滞、不确定系统,提出了一种新型控制算法(在模糊控制的基础上引入PID控制),且在MATLAB中分别进行了时域和频域仿真,并就簧载质量加速度、悬架动挠度、轮胎动载荷与被动悬架、模糊控制半主动悬架进行比较。结果表明,该算法对簧载质量加速度、悬架动挠度、轮胎动载荷的改善非常有效,对车辆悬架的控制和开发具有较大的参考价值。     

分类规则挖掘算法综述

分类规则挖掘是数据挖掘中一个重要的研究领域.通过介绍当前数据挖掘中具有代表性的分类算法,总结了各种算法的优缺点,给出了分类算法的应用以及分类算法面临的挑战,并对分类算法的发展方向进行了展望,为使用者选择算法或研究者改进算法提供了借鉴.     

记忆指导模拟退火算法的研究

提出了改进模拟退火算法的记忆指导模拟退火算法。该算法能够利用已知信息对搜索过程进行指导，提高算法的搜索效率，该法对回火退火算法的计算量减少５０％。     

逐步判别分析法的计算机实现

用汉字Ｆｏｘｂａｓｅ开发一个操作界面好、通用性强的逐步判别分析计算程序．     

原条量材设计有向图及其解的研究

原条量材设计是在原条造材前对原条各部位用途进行优化设计的过程。本文在国内外有关专家的研究基础上,提出用图论中的有向图理论,解决原条量材设计中的优化问题。用锯口线为点x_d,两个锯口线间的木段为弧y_n,描绘出由点集X和弧集Y组成的原条量材设计有向图T=(X,Y)。由于弧的权ω_(ij)是由木段的售价赋给,且保证ω_(ij)≥0,故用Dij kstra算法,求出最佳木材利用率下原条售价最高的路。