基于WT-CNN-LSTM的溶解氧含量预测模型

溶解氧（Dissolved oxygen, DO）含量是影响水产养殖产量的重要因素之一,具有时序性、不稳定性和非线性等特点,且其影响因子过多、存在复杂的耦合关系,难以实现精准预测。针对传统长短时记忆神经网络（Long short term memory, LSTM）预测模型易引入冗余数据,且在训练过长序列时会出现梯度消失现象,从而不能捕捉因子间长期的依赖性问题,提出了基于小波变换（Wavelet transform, WT）、卷积神经网络（Convolutional neural network, CNN）和LSTM的溶解氧含量预测模型。首先,使用WT降低数据噪声;然后,使用CNN深度挖掘各变量之间的潜在关系;最后,利用LSTM的时序性预测2h后的水产养殖溶解氧含量。结果表明,本文提出的WT-CNN-LSTM模型预测效果良好,其平均绝对误差、均方根误差和决定系数分别为0.138、0.229和0.954,比传统LSTM模型分别优化了28.87%、21.03%和4.61%。     

基于SVM的RBF网络在农机总动力预测中的应用

径向基神经网络是一种单隐层的三层前向网络,具有结构简洁、学习速度快等优点。为此,分析了径向基神经网络采用传统聚类方法确定基函数中心存在的问题,提出了一种基于支持向量机聚类确定径向基网络基函数中心的方法。该方法以最大间隔原理和结构风险最小化原则为前提,利用核方法把输入空间的样本映射到高维特征空间完成聚类工作来确定基函数的数量。采用改进的方法训练的径向基神经网络对黑龙江省农机总动力进行非线性时间序列预测,结果表明:改进的网络在确定网络结构、学习速度和提高网络预测精度方面都有较好的效果。     

不同采样密度的土壤水分特征参数预测

利用不同取样精度的土壤,将土壤质地(砂土、淤泥、粘土含量)和容重作为输入值,探讨了使用基于土壤转换函数的BP神经网络模型来预测0~20 cm表层土壤水分特征曲线参数,用甘肃省称钩河流域小流域的土样进行预测并进行了误差分析。结果表明,使用线性回归能够减小预测误差与实测值差距;使用BP神经网络来预测饱和体积含水量,其准确性比使用BP神经网络预测剩余体积含水量和田间持水量要高。为了进一步提高预测精度,还应尽可能地包括土壤结构、有机质含量等信息。     

Ancillary data supply strategies for improvement of temperature-based ETo ANN models

The development of new and more precise models for reference evapotranspiration (ET_o) estimation from minimum climatic data is mandatory, since the application of existing methods that provide acceptable results is limited to those places where large amounts of reliable climatic data are available. The performance quality of empirical equations and their local calibrations is to be questioned in a large variety of climatic contexts. As an alternative to traditional techniques, artificial neural networks (ANNs) are highly appropriate for the modelling of non-linear processes, which is the case of evapotranspiration. Nevertheless, temperature-based ANN models do not always provide accurate enough ET_o estimations and their performance depends highly on the specific relationships temperature-ET_o of the studied continental context. This paper describes the performance improvement of temperature-based ANN models through the consideration of exogenous ET_o records as ancillary inputs in different continental contexts of the autonomous Valencia region, on the Spanish Mediterranean coast. The influence on the model performance of the number of considered ancillary stations and the corresponding number of training patterns is also analysed. Finally, this performance is compared with existing empirical and ANN temperature-based models. The proposed models can be used with high accuracy not only for infilling purposes, but also for estimating ET_o outside the training station. Concerning models which demand scant climatic inputs, the proposed model provides performances with lower associated errors than the currently existing temperature-based models, which only consider local data. The local performance of the model presents 0.084 of average absolute relative error (AARE). The external performance of the model presents 0.1072 of AARE.     

基于GARBF神经网络的耕地土壤有效磷空间变异分析

为了调整耕地管理措施、合理施用磷肥、减少磷素流失、降低水体非点源污染,该研究以高州市为例,在全市各区镇共采集了664个耕作层(0～20cm)土样,利用遗传算法优化的径向基函数(radial basis function network optimized by geneti calgorithm,GARBF)神经网络和普通克里金法(Ordinary Kriging)等方法,分析了县域耕地土壤有效磷在不同采样尺度下的空间变异特征及其空间分布格局与成因。结果表明,高州市耕地表层土壤有效磷存在半方差结构,半方差函数曲线与指数和球状模型曲线拟合较好;5种采样尺度下(训练样点数分别为100、200、300、400和500)耕地表层土壤有效磷均表现出弱的结构空间相关,在较大范围内空间自相关性较差。GARBF神经网络空间插值能力在整体上要有优于基于邻近点RBF神经网络和普通克里金法。300样本下GARBF神经网络空间插值结果表明,高州市耕地表层土壤有效磷的盈余现象比较严重,并且盈余有效磷的流失对该地区水环境会产生严重的威胁。该研究结果可以为土壤属性空间估测、合理施肥以及降低水体非点源污染提供理论依据和技术支持。     

低压宽温状态下气体流量的动态测量方法

针对气体绝对压力小于40kPa的低压、宽温(-70～+80℃)状态下流量的测量,目前仍无简便有效的质量流量测量手段的问题,该文基于动态平衡原理提出了一种流量测量方法,设计了一套流量标定装置,其原理为低压、宽温状态下,流经调节阀的流量取决于阀前、后压力、阀门开度、气流温度4个变量。通过对由流量标定装置采集的原始数据样本的分析,利用BP神经网络方法建立了流量与阀前、后压力、阀门开度、气流温度的关系表达式。试验结果表明模型计算值与实测流量偏差小于1.8%,验证了该文所提出的流量测量方法有效性。     

补偿模糊神经网络水果形状分级器分级误差

针对神经网络对水果进行分级时精度有待提高的问题,分析了补偿模糊神经网络椪柑形状分级器的分级误差。将椪柑图像前4个傅里叶描述子按期望输出模糊变量值大小排列成单调递增、单调递减、钟形分布和锯齿形分布4种训练样本,分别训练同一补偿模糊神经网络水果形状分级器,用递减排序后的同一测试样本检验分级器性能,试验表明,单调递减顺序训练样本所训练的分级器分级误差最小为1.875%,钟形分布、单调递增顺序和锯齿形分布训练样本所训练的分级器分级误差依次增大,分别为15%、63.125%、75%。分析分级误差与样本间顺序的对应关系,建立分级误差模型,结果表明,同顺序的测试样本与训练样本间相关系数大,分级误差小;不同顺序的测试样本与训练样本间相关系数小,分级误差大。因此,测试样本与训练样本按水果同一品质特征同序排列,提高样本间的相关程度,将大幅度降低神经网络类分级器分级误差,提高正确识别率。     

雷公藤提取农药用总生物碱工艺的优化(简报)

为杀虫活性物质雷公藤生物碱的产业化开发利用奠定基础,以雷公藤根皮中总生物碱提取率为评价指标,综合考察pH值、提取时间、乙醇浓度、固液比和提取次数5个关键因素的影响,以L16(45)正交试验设计优选了雷公藤总生物碱的提取工艺条件,并利用人工神经网络进行仿真、评估和优化,最终获得了雷公藤根皮中总生物碱的提取优化工艺:根皮粉细度30目,pH2.5,固液比1︰10,乙醇浓度65%,提取时间2h,提取2次。该优化工艺条件不仅可减少成本,降低能耗,而且使雷公藤总生物碱提取率略有提高,在实际生产中具有较好的应用价值。     

可视化智能机械故障诊断系统的研究

运用模糊神经网络推理技术,结合传统的振动信号谱分析方法,设计了可视化智能机械故障诊断系统软件。采用面向对象技术,在Ｗｉｎｄｏｗｓ环境下使用面向对象的编程语言Ｂｏｒｌａｎｄ Ｃ＋＋４．５开发了系统软件。具体介绍了该系统的组成,功能衣特点,给出了齿轮箱诊断实例。     

Prioritisation of conservation areas in the Western Ghats, India

Areas of high conservation value were identified in the Western Ghats using a systematic conservation planning approach. Surrogates were chosen and assessed for effectiveness on the basis of spatial congruence using Pearson’s correlations and Mantel’s tests. The surrogates were, threatened and endemic plant and vertebrate species, unfragmented forest areas, dry forests, sub-regionally rare vegetation types, and a remotely sensed surrogate for unique evergreen ecosystems. At the scale of this analysis, amphibian richness was most highly correlated with overall threatened and endemic species richness, whereas mammals, especially wide-ranging species, were better at capturing overall animal and habitat diversity. There was a significant relationship between a remote sensing based habitat surrogate and endemic tree diversity and composition. None of the taxa or habitats served as a complete surrogate for the others. Sites were prioritised on the basis of their irreplaceability value using all five surrogates. Two alternative reserve networks are presented, one with minimal representation of surrogates, and the second with 3 occurrences of each species and 25% of each habitat type. These networks cover 8% and 29% of the region respectively. Seventy percent of the completely irreplaceable sites are outside the current protected area network. While the existing protected area network meets the minimal representation target for 88% of the species chosen in this study and all of the habitat surrogates, it is not representative with regard to amphibians, endemic tree species and small mammals. Much of the prioritised unprotected area is under reserve forests and can thus be incorporated into a wider network of conservation areas.