基于集成特征选择的森林火灾风险评估

摘 要：采用集成学习的思想,提出了一种基于集成特征选择的森林火灾风险评估方法。以特征选择方法的多样性和独立性为考量,选择了15种特征选择器并利用差异度进行筛选,获得异质选择器集合,进而得到特征子集集合。其次,利用各特征子集分别构建基于BP神经网络的森林火灾风险评估模型,并依据模型准确度筛选林火重要影响因子,构建最优森林火灾风险评估模型。结果表明,该算法准确度为85.96%,具有良好的泛化能力,可实现对森林火灾风险的有效评估。     

基于PNN算法的多传感器火灾探测技术研究

从传统算法和人工神经网络算法两方面,总结火灾探测中的多特征信息融合算法。以火焰、烟雾、CO和CO_2四种火灾特征的组合为例,基于MATLAB,运用PNN对实验采集到的数据进行训练和仿真测试。测试结果表明,采用PNN将多传感器信息融合后对火灾探测的准确度远高于单一种类火灾探测器;当扩展系数取0.3时,PNN对测试数据进行模式识别的准确度可高达98.95%。训练后的PNN可以更好地用于火灾的探测。     

基于自注意力的隧道视频火灾识别技术研究

针对公路隧道火灾样本量少、深度学习效果不理想的问题，研究一种小样本学习技术，以提高对隧道火灾样本的利用率，并在此基础上利用成熟的机器学习方法，提出一种基于自注意力的隧道视频火灾识别技术。该技术采用自注意力机制结合SVM分类器搭建火焰识别模型，该模型针对各项特征对火焰识别的重要性分配不同的注意力权重，形成注意力矩阵，并将权重矩阵与特征向量的值相加权，通过SVM的Hinge Loss进行线性支持向量机分类，对公路隧道火灾进行识别和预警。在火灾识别训练过程中，通过对火焰疑似区域进行检测，并利用数据增强技术达到样本扩增的目的，随后采用多通道融合的特征提取方式构建特征向量，输入设计的自注意力火焰识别模型中，通过梯度下降优化器进行小批量模型训练，降低迭代次数，最终获得最优特征权重参数，得到最佳识别模型。试验结果表明，该方法在模型训练时收敛较快，在火焰识别时相比未使用小样本学习的传统SVM算法，准确率提高了5%，因此能在小样本环境下有效提高火灾识别的准确度。     

锅炉热水供热系统能耗机器学习诊断模型

供热系统能耗诊断是一个难点。提出一种基于机器学习算法的能耗诊断标准模型结构,通过聚类或分类算法,从采集的诊断数据中筛选出节能特性较好的运行数据,基于回归模型建立能耗诊断模型对被诊断数据进行诊断。研究发现:1)经K-means聚类筛选数据并基于贝叶斯正则化训练的人工神经网络建立拟合模型,R值分别达到了0.975 6、0.970 5、0.921 4和0.910 1,模型拟合度较高;2)模型经过3个被诊断数据集验证,经过筛选的数据建立诊断模型,节能率指标分别10.7%、17%和4%,累积误差指标达到了-149 498.67、-86 526、-4 052.27kW,诊断效果优于未经聚类的数据建立的模型;3)对诊断结果进行分析,发现供热系统二次换热端热水流量变频节能控制措施节能效率较低,一次供热端热水流量人工调节是造成能耗过高的主要原因。这种数据建模诊断的方式是基于输入、输出变量之间的物理响应关系而不受数据时间特性的影响。     

基于贝叶斯算法的火灾报警系统研究

针对传统的建筑火灾报警系统探测信号单一、功能简单等问题,提出了以单片机作为下位机、LabVIEW作为上位机,采用多种传感器信息融合的贝叶斯算法的火灾报警系统。该系统将贝叶斯融合算法应用于火灾报警探测中,将收集到的火灾数据进行预处理,同时利用MATLAB软件对贝叶斯网络算法的数据融合进行仿真分析。仿真结果表明,该系统能够实时监测建筑内火灾事故相关的环境因素,3个特征量的融合提高了对火灾特征信号响应的均衡度和准确度,进而极大地提高了系统可靠性,为建筑火灾报警系统的设计与应用提供参考。     

基于BP神经网络的早期火灾图像识别软件

为实现对早期火灾的快速监测,设计了基于BP神经网络的早期火灾图像识别软件。该软件基于MATLAB平台,通过图像处理算法对火焰特征进行提取,并选取其中识别效果较好的特征指标对BP神经网络模型进行训练,从而实现了对火灾图像的快速识别。软件运行结果表明,该软件对火焰图像的识别准确度较高,能够达到识别监测早期火灾、完善火灾报警系统功能的设计目的。     

超声波技术处理含油污泥研究进展

以Indian Pine数据集为研究对象,利用等角特征映射对其进行特征提取,然后选取BP网络传统的梯度下降训练方法和正切拟牛顿法、Polak-Ribiere共轭梯度法、Levenberg-Marquart法3种数值优化的训练方法对其进行分类. 对分类结果进行对比分析,结果表明:基于数值分析的训练方法训练网络的耗时均比梯度下降法长,但收敛效果更好;总体分类精度均比梯度下降法至少提高6%;各类别制图精度都较高,且较为稳定,而梯度下降法只对易分的类别精度高;3种数值分析训练方法中,正切拟牛顿法和Polak-Ribiere共轭梯度法的收敛效率和分类精度比 Levenberg-Marquart法高.     

基于领域相似性迁移学习的行为识别

提出了1种基于领域相似性的迁移学习算法,利用其他领域中的相关数据帮助完成当前领域的行为识别任务。首先通过典型相关性分析,得到领域间相似性的约束并与目标分类模型相联系,以充分利用相关领域中的有效信息;然后学习1种具有重建性、判别性、域适应性的跨域字典对,将不同领域的数据特征映射到同一空间;最后根据映射特征和分类模型对行为进行识别。利用网络中的大量图像,在UCF Sports Action数据集上的实验结果表明了算法的有效性。     

基于LSTM网络的船舶机舱火灾多特征融合探测

针对船舶机舱火灾高效准确探测的需求,建立基于LSTM-ID3 判决的船舶火灾探测方法。首先确定采集船舶火灾特征的三类传感器,然后完成 LSTM 神经网络模型的构建、参数的优化,将 LSTM 神经网络输出的明火、阴燃火、无火的概率值与烟雾持续时间作为决策树的输入量,输出火灾探测结果。利用国家标准火典型数据进行训练,并开展相关试验,对船舶机舱火灾进行探测。试验结果表明,与其他算法进行对比,探测准确率达到97%以上,该方案能对机舱火灾做出有效探测,为船舶安全提供科学依据。     

基于机器学习的火灾事故等级分类研究

为了探究火灾的严重程度与消防队响应时间、救援人数、火灾地点、火灾探测器以及自喷系统等之间的关系,采用随机森林、人工神经网络、支持向量机和极限学习机4 种机器学习算法对美国旧金山市的历史火灾数据进行了挖掘分析。运用模糊理论将消防队响应时间和救援人数转换为三角模糊数,提出了一种基于模糊推理的火灾因素分类方法。研究发现4 种算法的准确率均超过90％,并通过交叉验证的方法证明了这些算法的可靠性。在4 种算法中,随机森林算法的准确率和Kappa 值均高于其他算法,但是其计算得到的第三类火灾的AUC 值最小。     

A novel machine learning‐based algorithm to detect damage in high‐rise building structures

A novel model is presented for global health monitoring of large structures such as high‐rise building structures through adroit integration of 2 signal processing techniques, synchrosqueezed wavelet transform and fast Fourier transform, an unsupervised machine learning technique, the restricted Boltzmann machine, and a recently developed supervised classification algorithm called neural dynamics classification (NDC) algorithm. The model extracts hidden features in the frequency domain of the denoised measured response signals recorded by sensors on different elevations or floors of a structure. The extracted features are used as an input of the NDC to detect and classify the global health of the structure into categories such as healthy, light damage, moderate damage, severe damage, and near collapse. The proposed model is validated using the data obtained from a 3D 1:20 scaled 38‐story reinforced concrete building structure. The results are compared with 3 other supervised classification algorithms: k‐nearest neighbor (KNN), probabilistic neural networks (PNN), and enhanced PNN (EPNN). NDC, EPNN, PNN, and KNN yield maximum average accuracies of 96%, 94%, 92%, and 82%, respectively.     

A multiple kernel learning-based decision support model for contractor pre-qualification

Due to the complex nature of the contractor pre-qualification such as subjectivity, non-linearity and multi-criteria, advanced model should be required for achieving a high accuracy of this decision-making process. Previous studies have been conducted to build up quantitative decision models for contractor pre-qualification, among them artificial neural network (ANN) and support vector machine (SVM) have been proved to be desirable in solving the pre-qualification problem with regards to their higher accuracy and efficiency for solving the non-linear problem of classification. Based on the algorithm of SVM, multiple kernel learning (MKL) method was developed and it has been proved to perform better than SVM in other areas. Hence, MKL is proposed in this research, the capability of MKL was compared with SVM through a case study. From the result, it has been proved that both SVM and MKL perform well in classification, and MKL is more preferable than SVM, with a proper parameter setting. Therefore, MKL can enhance the decision making of contractor pre-qualification.     

PSO优化ELM在火灾探测中的应用

为提高火灾探测精度,避免标准ELM陷入局部最优,本文基于火灾特征值CO浓度、烟雾浓度、温度建构了一种基于粒子群(PSO)优化极限学习机(ELM)的火灾探测模型,通过PSO优化ELM输入层与隐含层权值以及偏置,利用最优值进行极限学习机网络训练,将训练好的网络对测试样本进行预测并验证方法有效性.研究显示,PSO-ELM的均...     

基于机器学习的风景园林智能化分析应用研究

机器学习使实现数据的智能化处理及充分利用数据中蕴含的知识与价值成为可能。探索基于机器学习在风景园林领域智能化分析应用的途径,开展3个实验。其中2个与数据分析研究相关,提出基于调研图像色彩聚类分析的城市色彩印象和基于图像识别技术的景观视觉质量评估与网络应用平台部署实验。最后1个实验与数字化设计创作相关,提出用于设计方案遴选的地形生成方法,包括2个子项目:应用深度学习生成对抗网络(GAN)的地形生成和建立遮罩、预测未知区域的高程。3个实验应用到机器学习中分类、聚类和回归3个主要方向中的算法以及深度学习的生成对抗网络,对传统的研究问题提出了基于机器学习新的研究方法。因此,在应用机器学习风景园林领域,可以有效地从多源数据中学习相互增强的知识,发现问题,并提出解决问题的新方法。     

Predicting the presence of E. coli in tap water using machine learning in Nepal

Within developing countries, a multitude of problems that affect the water supply process can result in the contamination of water taps. While machine learning applications have become popular for attaining efficient water quality predictions, acquiring the necessary data for modelling for developing countries is challenging. This study constructs water quality prediction models by machine learning with a pseudo-pipeline network to complement the missing data of the water supply process. Using both water source and water tap quality information measured by the Government of Nepal, we apply the three machine learning models: support vector machine (SVM), random forest (RF) and LightGBM. Furthermore, we also apply a traditional statistical method—logistic regression (LR)—to the prediction of the Escherichia coli (E. coli) contamination in water taps. With some input variables (such as the length from the nearest sources) obtained from the pseudo-pipeline network, the results show that SVM has stable and high accuracy for both the 26 cities (70%) and for the 25 cities except for Kathmandu (79%). LR performed a significantly lower accuracy for all cities (61%) than for 25 cities (79%). Additionally, we show that our method can be applied to other regions where a water quality survey has not yet been conducted.     

基于循环神经网络的盾构隧道#br# 引发地面最大沉降预测

伴随着计算机技术的快速发展,机器学习等新兴算法正在被越来越多地运用于预测隧道掘进引发的地面最大沉降。在隧道施工过程中,由盾构机和地面监测点位采集的数据具有很强的序列化特征,而传统的机器学习算法对序列数据的处理存在一定的局限性。循环神经网络(RNN)具有极强的对时序型数据的处理能力,在视频识别、语音翻译等领域有着广泛的应用。采用两种RNN模型(LSTM、GRU)和传统的BP神经网络模型,以地质参数、几何参数和盾构机参数作为输入,对隧道施工过程中引发的地面最大沉降进行预测分析。结果显示,RNN对隧道沉降的预测结果优于传统的BP神经网络模型,并且RNN在连续未知区段的预测结果比BPNN更加稳定。     

SPT based determination of undrained shear strength: Regression models and machine learning

The purpose of this study is the accurate prediction of undrained shear strength using Standard Penetration Test results and soil consistency indices, such as water content and Atterberg limits. With this study, along with the conventional methods of simple and multiple linear regression models, three machine learning algorithms, random forest, gradient boosting and stacked models, are developed for prediction of undrained shear strength. These models are employed on a relatively large data set from different projects around Turkey covering 230 observations. As an improvement over the available studies in literature, this study utilizes correct statistical analyses techniques on a relatively large database, such as using a train/test split on the data set to avoid overfitting of the developed models. Furthermore, the validity and consistency of the prediction results are ensured with the correct use of statistical measures like p-value and cross-validation which were missing in previous studies. To compare the performances of the models developed in this study with the prior ones existing in literature, all models were applied on the test data set and their performances are evaluated in terms of the resulting root mean squared error (RMSE) values and coefficient of determination (R²). Accordingly, the models developed in this study demonstrate superior prediction capabilities compared to all of the prior studies. Moreover, to facilitate the use of machine learning algorithms for prediction purposes, entire source code prepared for this study and the collected data set are provided as supplements of this study.     

基于深度学习的VGG16 图像型火灾探测方法研究

为了快速、有效地检测不同场景下的火灾信息,基于深度迁移学习设计了一种改进VGG16 的图像型火灾检测方法。搜集不同场景下的照片,使用离线数据增强技术增加样本数量,对VGG16 进行改进,并使用迁移学习的方法训练火灾识别模型。结果表明：改进的VGG16 网络对于火灾现场的图片分类识别准确率为98.7%,优于Resnet50 网络和Densenet121 网络,可快速、准确地检测到火灾信息。