基于EMD-GWO-SVR组合模型的短期风速预测

风速预测对风电场进行调度与控制具有重大意义。针对风速序列的随机性与间歇性，文中提出了EMD-GWO-SVR组合预测模型。先对原始序列进行经验模态分解，并应用GWO算法对支持向量回归模型的参数进行寻优。随后将寻优得到的最佳参数代入支持向量回归模型，并对分解后的本征模函数及残差项分别进行预测，将得到的各预测结果相加从而对风速进行预测。以甘肃省酒泉市的历史气象数据为例，建立BP神经网络、SVR、PSO-SVR、GWO-SVR、EMD-PSO-SVR和EMD-GWO-SVR6种预测模型，对该地的风速进行预测。仿真结果表明，文中提出的EMD-GWO-SVR模型预测精度相比SVR提高了61.759 8%,且其MAE、MAPE和RMSE等误差指标评价值显著低于其它5种模型。     

应用SVM的PM2.5未来一小时浓度动态预报模型

目前现有的PM2.5模式预报值偏离实况观测值较大。针对上述问题,从上海浦东气象局获得2012年11月~2013年11月的PM2.5实况观测浓度、PM2.5模式预报（WRF-CHEM）浓度和主要气象影响因子的模式预报数据资料,在PM2.5模式预报数据的基础上,加入另外5个主要气象影响因子的模式预报数据,应用支持向量机（SVM）建立动态预报模型,提高PM2.5未来一小时浓度预报的精度,并且与径向基神经网络（RBFNN）、多元线性回归法（MLR）、WRF-CHEM作对比。实验结果表明:该算法较大提高了PM2.5未来一小时浓度预报的精度,预报精度优于RBFNN、MLR和WRF-CHEM,并且对PM2.5浓度变化剧烈的情况具有较好地预报能力。     

基于向量自回归模型的旅游热门景点预测方法研究

常规的预测方法变量之间的皮尔逊相关值较低,造成预测的结果出现偏差,为此,设计基于向量自回归模型的旅游热门景点预测方法。综合不同的搜索引擎数据,计算旅游景点关键词网络搜索指数,对其进行预处理并筛选出与旅游景点热度相关性较强的关键词,利用向量自回归模型对变量进行均值化处理,确定影响最大的网络搜索指数,实现对旅游热门景点的预测。实验结果表明:与常规的灰度预测方法和SVR模型预测方法相比,基于向量自回归模型的预测方法的皮尔逊相关值能够保持在0.8～1.0之间,变量之间具有极强的相干性,适合应用在旅游热门景点预测中。     

基于Adam 注意力机制的PM2:5浓度预测方法

大气 PM2:5 浓度是一种具有较强时序特征的数据, 故目前关于 PM2:5 浓度的预测多选择 RNN、 LSTM 等 序列模型进行。但由于 RNN、 LSTM 等模型对不同时刻输入的数据都采用相同的权重进行计算, 不符合类脑设 计, 造成 PM2:5 浓度预报准确率较低。针对以上问题, 提出一种基于 Adam 注意力机制的 PM2:5 预测方法 (AT-RNN 和AT-LSTM), 该方法首先通过 Adam 算法寻找 RNN 或 LSTM 的最优参数并在 Encoder 阶段引入注意力机制, 将 注意力权重分配给具有时间序列特征的输入, 再进行 Decoder 解析和预测。通过实验, 对比了 BP、 RNN、 LSTM 和AT-RNN、 AT-LSTM 预测合肥市 PM2:5 浓度的效果。结果表明, 基于 Adam 注意力模型的预测方法准确率优于其它 方法, 证明该方法在污染物预测中的有效性。     

基于INTLBO-SVR的低压断路器热脱扣时间预测

热脱扣时间是低压断路器的关键指标,利用断路器生产过程中可检测数据可以实现热脱扣时间的预测。针对支持向量回归(SVR)进行热脱扣时间预测,参数的选择对预测的精度和泛化性能影响较大问题,提出一种基于隔离小生境教学算法(Isolated Niche Teaching-Learning-Based Optimization Algorithm,INTLBO)优化支持向量回归的热脱扣时间预测方法。该方法针对教学算法易陷入局部最优的缺点,采用隔离机制的小生境技术对其进行改进,然后利用INTLBO寻优找到最优的SVR参数。根据低压断路器生产历史数据,建立基于INTLBO-SVR的热脱扣时间预测模型。仿真结果表明,与TLBOSVR和常规SVR等方法相比,INTLBO-SVR模型具有较好的预测性能。     

ACPSO—SVR结合的非线性建模预测算法研究

提出一种基于自适应混沌粒子群优化和支持向量机结合的非线性预测建模算法（ACPSO—SVR）,引入ACPSO启发式寻优机制对SVR模型的超参数进行自动选取,在超参数取值范围变化较大的情况下,效果明显优于网格式搜索算法。选取UCI机器学习数据库中的Forestfires标准数据集进行测试,实验结果表明该方法具有较高的精度和良好的泛化能力．对于解决多变量的回归预测问题是一种有效的方法。最后给出了混合算法在碳一多相催化领域的两种典型应用．在反应动力学模型未知的情况下建立催化剂组份模型和操作条件模型,以及基于混合算法的最优催化剂设计框架。     

基于随机森林回归分析的PM2.5浓度预测模型

针对神经网络算法在当前PM2.5浓度预测领域存在的易过拟合、网络结构复杂、学习效率低等问题,引入RFR(random forest regression,随机森林回归)算法,分析气象条件、大气污染物浓度和季节所包含的22项特征因素,通过调整参数的最优组合,设计出一种新的PM2.5浓度预测模型—RFRP模型.同时,收集了西安市2013-2016年的历史气象数据,进行模型的有效性实验分析.实验结果表明,RFRP模型不仅能有效预测PM2.5浓度,还能在不影响预测精度的同时,较好地提升模型的运行效率,其平均运行时间为0.281 s,约为BP-NN(back propagation neural network,BP神经网络)预测模型的5.88％.     

基于SVR-LUR模型的城市道路PM10空间浓度分布模拟

针对传统土地利用回归模型(Land use regression, LUR)未考虑影响因子与大气污染物之间非线性复杂关系 和易出现多重共线性的问题。以PM10为例,采用支持向量回归机(Support vector machine regression, SVR) 改进土地利用回归模型的建模方法构建SVR-LUR模型,对上海市南浦大桥周边区域PM10空 间分布进行模拟。研究结果表明: 1)研究区域PM10浓度与100 m缓冲区内的空地面积, 150 m缓冲区内的建筑工地面积、 空地面积、河流面积, 200 m缓冲区内的绿地面积和河流面积,以及湿度、交通流量和背景浓度相关性较高。 2) SVR-LUR模型可较好地对研究区PM10浓度进行空间分布预测。SVR-LUR模型与LUR模型相比, SVR-LUR模型预测精 度较高,其测试集比LUR模型测试集的平均绝对误差(Mean absolute error, MAE)及均方根误差(Root mean squares error, RMSE)分别减小了22.92\%、33.51\%,拟合指数(Index of agreement, IA)值增加了13.20\%。 相较于普通克里金插值模型所得到的单一梯度空间分布预测结果, SVR-LUR 模型能够更有效揭示小范围内的空间差异。 3)研究区PM10浓度空间分布呈现出西高东低的总格局,在建筑物和路网密集的地方浓度较高,而在靠近江面和空地 的区域浓度相对较低。模拟结果与实际情况相符。     

基于粒子群算法的BP 神经网络在大气 NO2 浓度预测中的应用研究

NO2 是主要的大气污染气体之一, 在大气光化学过程中起着重要作用。研究 NO2 浓度的时空演变, 预测其浓 度变化趋势, 对政府出台改善环境措施具有重要意义。提出利用粒子群算法 (PSO) 的反向传播 (BP) 神经网络对大气 NO2 浓度进行预测。以合肥地区 2017 年 1 月 1 日至 2019 年 12 月 31 日的大气污染数据和气象数据为基础, 结合逐步 回归方法筛选出与 NO2 浓度相关性较大的影响因子作为输入样本。构建 PSO-BP 神经网络预测模型, 利用 PSO 找出 BP 神经网络最优的初始权值和阈值。对比 BP 神经网络、遗传算法改进的 BP 神经网络和 PSO 改进的 BP 神经网络 三种模型的预测结果, 发现 PSO-BP 模型能够较为准确地预测出 NO2 浓度的动态变化规律, 并且预测精度高、模式简 单, 有望广泛应用于大气污染物浓度预测等方面的研究。     

多因素影响下的忙时话务量预测方法的研究

为了提高节假日移动话务量的预测精度,提出了一种基于多因素影响的忙时话务量预测方法。首先对忙时话务量进行相关性分析,得到4个影响话务量的重要因子,然后把4个因子与忙时话务量训练数据一起作为输入变量,最后用改进的具有非线性惯性权重的粒子群算法优化的最小二乘支持向量机模型进行预测。实验结果表明该预测模型有更高的预测精度和较强的泛化能力。     

周期为4v(几乎)平衡理想二进制序列构造研究

具有理想自相关特性的序列在无线通信、雷达以及密码学中具有重要的作用.因此为了扩展更多可应用于通信系统的理想序列,该文基于2阶分圆类和中国剩余定理,提出3类新的周期为T=4v(v是奇素数)平衡或几乎平衡理想二进制序列构造方法.构造所得序列的周期自相关函数满足:当v≡3(mod4)时,序列的周期自相关函数旁瓣值取值集合为{...     

合肥市霾日气象要素特征分析及预报方法研究

利用合肥国家气象观测站建站以来历史观测资料(1952$\sim$2016年)及2004$\sim$2016年逐日地面观测资料和高空观测资料, 分析了合肥霾日及相关气象要素特征;并结合2004$\sim$2016年合肥空气质量数据,对合肥市霾的预报方法进行了研究, 初步得到了霾日和霾日能见度的预报方程;最后利用2017年的气象资料和空气质量资料对预报结果进行了检验。结果表明: 合肥出现的霾以轻微霾和轻度霾为主;全市霾日逐月分布不均,其中1月和12月最多。霾的生消有比较明显的日变化, 霾出现的时间主要集中在每日的11:00$\sim$20:00。霾出现的有利气象条件是:风速$\leq$3 m/s,地面主导风向为 偏东风到偏南风,有负变压,正变温,相对湿度在50$\sim$70\%之间。低能见度的霾需要相对较高的相对湿度和负变压。 建立的霾日及霾日能见度方程预报准确度较高,可以为合肥市霾的预报预警提供一定参考。     

边缘网络中一种VNF需求预测方法

针对当前虚拟网络功能（Virtualization Network Functions,VNF）需求预测方法准确率较低且不适用于边缘网络的问题,提出了一种在边缘网络中基于支持向量回归（Support Vector Regression,SVR）与门控循环单元（Gated Recurrent Unit,GRU）神经网络模型结合的VNF需求预测方法。考虑到网络边缘流量具有突发性、自相似性及长相关性等特点,结合SVR和GRU两种模型的优点,利用计算复杂度较低的SVR和GRU模型分别提取网络服务历史时序数据的短期特征和长期特征,以提高VNF需求预测准确率,实现边缘网络中VNF的提前放置。实验表明,所提出的预测方法在边缘网络中针对不同网络服务的预测较于传统方法、循环神经网络（Recurrent Neural Networks,RNN）、长短期记忆网络（Long Short-Term Memory,LSTM）模型能够降低20%~30%的误差,有更佳的预测效果。     

Optimal Randomizer Efficiency in the Bounded-Storage Model

In the bounded-storage model for information-theoretically secure encryption and key-agreement one can prove the security of a cipher based on the sole assumption that the adversarys storage capacity is bounded, say by $s$ bits, even if her computational power is unlimited. Assume that a random $t$-bit string $R$ is either publicly available (e.g., the signal of a deep-space radio source) or broadcast by one of the legitimate parties. If $s < t$, the adversary can store only partial information about $R$. The legitimate sender Alice and receiver Bob, sharing a short secret key $K$ initially, can therefore potentially generate a very long $n$-bit one-time pad $X$ with $n\gg|K|$ about which the adversary has essentially no information. All \looseness = –1 previous results in the bounded-storage model were partial or far from optimal, for one of the following reasons: either the secret key $K$ had to be longer than the derived one-time pad ($n < |K|$), or $t$ had to be extremely large ($t > ns$), or the adversary was assumed to be able to store only $s$ actual bits of $R$ rather than arbitrary $s$ bits of information about $R$, or the adversary received a non-negligible amount of information about $X$. In this paper we prove the first non-restricted security result in the bounded-storage model: $K$ is short, $X$ is very long, and $t$ needs to be only moderately larger than $s + n$. In fact, $s/t$ can be arbitrarily close to $1$ and hence the storage bound is essentially optimal. The security can be proved also if $R$ is not uniformly random, provided that the min-entropy of $R$ is sufficiently greater than $s$.     

Asymptotic Throughput Capacity Analysis of Multi-Channel, Multi-Interface Wireless Mesh Networks

Research into the analytical solutions for the capacity of the infrastructure wireless mesh networks (InfWMN) is highly interesting. An InfWMN is a hierarchical network consisting of mesh clients, mesh routers and gateways. The mesh routers form a wireless mesh infrastructure to which the mesh clients are connected through the use of star topology. The previous analytical solutions have only investigated the asymptotic per-client throughput capacity of either single-channel InfWMNs or multi-channel InfWMNs under conditions in which each infrastructure node (i.e. wireless routers and gateways), has a dedicated interface per-channel. The results of previous analytical studies show that there are quite few studies that have addressed the more practical cases where the number of interfaces per-node is less than the number of channels. In this paper, we derive an original analysis of the asymptotic per-client throughput capacity of multi-channel InfWMNs in which the number of interfaces per-infrastructure node, denoted by m, is less than or equal to the number of channels, denoted by c. Our analysis reveals that the asymptotic per-client throughput capacity of multi-channel InfWMNs has different bounds, which depend on the ratio between c and m. In addition, in the case that m < c, there is a reduction in the capacity of the InfWMN compared to the case in which c = m. Our analytical solutions also prove that when ${\frac{\text{c}}{\text{m}}=\Omega\left({\frac{{\rm N}_g^2}{{\rm N}_{\rm r}}}\right)}$ , where N_g and N_r denote the number of gateways and mesh routers respectively, gateways cannot effectively increase the throughput capacity of the multi-channel InfWMNs.     

Critical Layer Thickness in Exponentially Graded Heteroepitaxial Layers

Exponentially graded semiconductor layers are of interest for use as buffers in heteroepitaxial devices because of their tapered dislocation density and strain profiles. Here we have calculated the critical layer thickness for the onset of lattice relaxation in exponentially graded In _x Ga_1?x As/GaAs (001) heteroepitaxial layers. Upwardly convex grading with \( x = x_{\infty } \left( {1 - {\rm e}^{ - \gamma /y} } \right) \) was considered, where y is the distance from the GaAs interface, γ is a grading length constant, and x _∞ is the limiting mole fraction of In. For these structures the critical layer thickness was determined by an energy-minimization approach and also by consideration of force balance on grown-in dislocations. The force balance calculations underestimate the critical layer thickness unless one accounts for the fact that the first misfit dislocations are introduced at a finite distance above the interface. The critical layer thickness determined by energy minimization, or by a detailed force balance model, is approximately \( h_{\rm{c}} \approx <Exponentially graded semiconductor layers are of interest for use as buffers in heteroepitaxial devices because of their tapered dislocation density and strain profiles. Here we have calculated the critical layer thickness for the onset of lattice relaxation in exponentially graded In _x Ga_1−x As/GaAs (001) heteroepitaxial layers. Upwardly convex grading with x = x_￥ ( 1 - e ^{- g/y} ) x = x_{\infty } \left( {1 - {\rm e}^{ - \gamma /y} } \right) was considered, where y is the distance from the GaAs interface, γ is a grading length constant, and x _∞ is the limiting mole fraction of In. For these structures the critical layer thickness was determined by an energy-minimization approach and also by consideration of force balance on grown-in dislocations. The force balance calculations underestimate the critical layer thickness unless one accounts for the fact that the first misfit dislocations are introduced at a finite distance above the interface. The critical layer thickness determined by energy minimization, or by a detailed force balance model, is approximately h_c ? < h_{\rm{c}} \approx < Although these results were developed for exponentially graded In _x Ga_1−x As/GaAs (001), they may be generalized to other material systems for application to the design of exponentially graded buffer layers in metamorphic device structures such as modulation-doped field-effect transistors and light-emitting diodes.     

Silicon field-effect transistors as radiation detectors for the Sub-THz range

The nonresonance response of silicon metal-oxide-semiconductor field-effect transistors (Si-MOSFETs) with a long channel (1–20 μm) to radiation in the frequency range 43–135 GHz is studied. The transistors are fabricated by the standard CMOS technology with 1-μm design rules. The volt-watt sensitivity and the noise equivalent power (NEP) for such detectors are estimated with the calculated effective area of the detecting element taken into account. It is shown that such transistors can operate at room temperature as broadband direct detectors of sub-THz radiation. In the 4–5 mm range of wavelengths, the volt-watt sensitivity can be as high as tens of kV/W and the NEP can amount to ${{10^{ - 11} - 10^{ - 12} W} \mathord{\left/ {\vphantom {{10^{ - 11} - 10^{ - 12} W} {\sqrt {Hz} }}} \right. \kern-\nulldelimiterspace} {\sqrt {Hz} }}${{10^{ - 11} - 10^{ - 12} W} \mathord{\left/ {\vphantom {{10^{ - 11} - 10^{ - 12} W} {\sqrt {Hz} }}} \right. \kern-\nulldelimiterspace} {\sqrt {Hz} }}. The parameters of detectors under study can be improved by the optimization of planar antennas.