基于神经网络的建筑结构安全评估模型研究

为实现建筑结构安全的快速评估,提出基于神经网络的建筑结构安全评估方法。基于《民用建筑可靠性鉴定标准》的调查与检测要求并考虑数据易获取性,选择45个涵盖承载力、耐久性、历史记录和环境情况等变量作为输入参数,以《民用建筑可靠性鉴定标准》中的安全等级作为输出参数,采用深度置信网络学习输入参数与输出参数间的非线性映射关系。对输入参数的选择、样本缺值问题、小样本问题和神经网络评估的可靠性进行探讨和验证。结果表明:在无法准确判断输入参数与输出参数相关性的前提下,采用全部输入参数的评估模型具有更高的鲁棒性; 迷失森林算法相较其他常用的缺值插补算法有更好的插补性能; 采用变分自编码器扩充训练样本集能有效提高神经网络的泛化能力和分类精度; 对深度置信网络引入加权交叉熵损失函数加以改进可增加训练时对不安全类别的敏感性,牺牲少量不安全类别的查准率可以大幅提高其查全率; 基于神经网络的结构安全评估模型能较好地预测结构的安全等级,具有快速且大批量运算的优势,是实现大范围建筑群结构安全监测的有效手段。     

Artificial neural network based modelling of desiccant wheel

In this article, the concept of artificial neural network and goal oriented design have been used to propose a computer design tool that can help designers to evaluate performance of desiccant cooling system and behaviour of the desiccant wheel. Based on the experimental observations on desiccant wheel, a neural network model has been developed using a neural network toolbox of MATLAB^® with feed forward back propagation method. The model has been validated against experimental data sets. A number of training algorithms with feed forward back propagation method have been used for the modelling of desiccant wheel to identify a training algorithm with least mean square error (MSE). The performance of all training algorithms has been analyzed and training algorithm trainlm (Levenberg-Marquardt back propagation) is found most suitable for the prediction of outputs which have least mean square error of 0.064462 and 0.007575 for specific humidity and temperatures respectively. The proposed model can predict the specific humidity and temperature at the outlet of desiccant wheel within the range of experimental values.     

Neural Networks for Evaluating CPT Calibration Chamber Test Data

Abstract: The feasibility of using neural network models for evaluating CPT calibration chamber test data is investigated. The backpropagation neural network algorithm was used to analyze the data. After learning from a set of randomly selected patterns, the neural network model was able to produce reasonably accurate predictions for patterns not included in the training set. The neural network performance was found to be simpler and more effective than regression analysis for modeling the CPT test data. Correlations between the cone measurements and the engineering properties of sand can be developed using the generalization capabilities of the neural network.     

Artificial neural network optimized by differential evolution for predicting diameters of jet grouted columns

A novel and effective artificial neural network (ANN) optimized using differential evolution (DE) is first introduced to provide a robust and reliable forecasting of jet grouted column diameters. The proposed computational method adopts the DE algorithm to tackle the difficulties in the training and performance of neural networks and optimize the four quintessential hyper-parameters (i.e. the epoch size, the number of neurons in a hidden layer, the number of hidden layers, and the regularization parameter) that govern the neural network efficacy. This approach is further enhanced by a stochastic gradient optimization algorithm to allow ‘expensive’ computation efforts. The ANN-DE is first trained using a prepared jet grouting dataset, then verified and compared with the prevalent machine learning tools, i.e. neural networks and support vector machine (SVM). The results show that, the ANN-DE outperforms the existing methods for predicting the diameter of jet grouting columns since it well balances training efficiency and model performance. Specifically, the ANN-DE achieved root mean square error (RMSE) values of 0.90603 and 0.92813 for the training and testing phases, respectively. The corresponding values were 0.8905 and 0.9006 for the optimized ANN, then, 0.87569 and 0.89968 for the optimized SVM, respectively. The proposed paradigm is bound to be useful for solving various geotechnical engineering problems regardless of multi-dimension and nonlinearity.     

Deep learning of rock microscopic images for intelligent lithology identification: Neural network comparison and selection

An intelligent lithology identification method is proposed based on deep learning of the rock microscopic images. Based on the characteristics of rock images in the dataset, we used Xception, MobileNet_v2, Inception_ResNet_v2, Inception_v3, Densenet121, ResNet101_v2, and ResNet-101 to develop microscopic image classification models, and then the network structures of seven different convolutional neural networks (CNNs) were compared. It shows that the multi-layer representation of rock features can be represented through convolution structures, thus better feature robustness can be achieved. For the loss function, cross-entropy is used to back propagate the weight parameters layer by layer, and the accuracy of the network is improved by frequent iterative training. We expanded a self-built dataset by using transfer learning and data augmentation. Next, accuracy (acc) and frames per second (fps) were used as the evaluation indexes to assess the accuracy and speed of model identification. The results show that the Xception-based model has the optimum performance, with an accuracy of 97.66% in the training dataset and 98.65% in the testing dataset. Furthermore, the fps of the model is 50.76, and the model is feasible to deploy under different hardware conditions and meets the requirements of rapid lithology identification. This proposed method is proved to be robust and versatile in generalization performance, and it is suitable for both geologists and engineers to identify lithology quickly.     

Automated detection of sewer pipe defects in closed-circuit television images using deep learning techniques

Sanitary sewer systems are designed to collect and transport sanitary wastewater and stormwater. Pipe inspection is important in identifying both the type and location of pipe defects to maintain the normal sewer operations. Closed-circuit television (CCTV) has been commonly utilized for sewer pipe inspection. Currently, interpretation of the CCTV images is mostly conducted manually to identify the defect type and location, which is time-consuming, labor-intensive and inaccurate. Conventional computer vision techniques are explored for automated interpretation of CCTV images, but such process requires large amount of image pre-processing and the design of complex feature extractor for certain cases. In this study, an automated approach is developed for detecting sewer pipe defects based on a deep learning technique namely faster region-based convolutional neural network (faster R-CNN). The detection model is trained using 3000 images collected from CCTV inspection videos of sewer pipes. After training, the model is evaluated in terms of detection accuracy and computation cost using mean average precision (mAP), missing rate, detection speed and training time. The proposed approach is demonstrated to be applicable for detecting sewer pipe defects accurately with high accuracy and fast speed. In addition, a new model is constructed and several hyper-parameters are adjusted to study the influential factors of the proposed approach. The experiment results demonstrate that dataset size, initialization network type and training mode, and network hyper-parameters have influence on model performance. Specifically, the increase of dataset size and convolutional layers can improve the model accuracy. The adjustment of hyper-parameters such as filter dimensions or stride values contributes to higher detection accuracy, achieving an mAP of 83%. The study lays the foundation for applying deep learning techniques in sewer pipe defect detection as well as addressing similar issues for construction and facility management.     

Combination of neural network and wavelet transform for islanding detection of distributed generation in a small-scale network

In this paper, an optimum and intelligent method is proposed for islanding detection using wavelet transform. The suggested relay is based on neural network (NN) in which different heuristic algorithms are used for training the NN. In the proposed method, the appropriate signals for detection procedure as well as mother wavelet are selected optimally, based on the mean square error (MSE) concept. Lately, the desired relay is trained by the optimally selected signals using four different algorithms and the optimum condition of the fault detector is identified. Simulation results approved that non detection zone (NDZ) has a significant reduction utilising the proposed intelligent technique. The contributions of the proposed method include presenting an appropriate signal selection method based on MSE, selecting optimum number of relay input signals using the proposed technique, fast training of intelligent relay by using least information, solving threshold selection problem and reduction of NDZ approximately to zero.     

Tide Prediction Using Neural Networks

Prediction of tides at a subordinate station located in the interior of an estuary or a bay is normally done by applying an empirical correction factor to observations at some standard or reference station. This paper presents an objective way to do so with the help of the neural network technique. In complex field conditions this approach may look more attractive to apply. Prediction of high water and low water levels as well as that of continuous tidal curves is made at three different locations. The networks involved are trained using alternative training algorithms. Testing of the networks indicated satisfactory reproduction of actual observations. This was further confirmed by a high value of the accompanying correlation coefficient. Such a correlation was better than the one obtained through use of the statistical linear regression model. The training algorithm of cascade correlation involved the lowest training time and hence is found to be more suitable for adaptive training purpose.     

Enhancing Neural Network Traffic Incident-Detection Algorithms Using Wavelets

Researchers have presented freeway traffic incident-detection algorithms by combining the adaptive learning capability of neural networks with imprecision modeling capability of fuzzy logic. In this article it is shown that the performance of a fuzzy neural network algorithm can be improved through preprocessing of data using a wavelet-based feature-extraction model. In particular, the discrete wavelet transform (DWT) denoising and feature-extraction model proposed by Samant and Adeli (2000) is combined with the fuzzy neural network approach presented by Hsiao et al. (1994). It is shown that substantial improvement can be achieved using the data filtered by DWT. Use of the wavelet theory to denoise the traffic data increases the incident-detection rate, reduces the false-alarm rate and the incident-detection time, and improves the convergence of the neural network training algorithm substantially.     

基于神经网络的空调系统负荷预测研究

以北京某建筑的空调系统作为实例研究对象,在所采集的3年实测数据基础上,主要探讨了基于ELMAN神经网络的日冷负荷预测方法和误差。用多元线性回归方法分析了日冷负荷神经网络预测模型输入参数对输出结果的影响度。最后经实验验证,以3周以上历史数据为训练集经多次预测后取平均值,具备较高的预测精度,同样可以指导工程实际设计。     

Prediction of shield tunneling-induced ground settlement using machine learning techniques

Predicting the tunneling-induced maximum ground surface settlement is a complex problem since the settlement depends on plenty of intrinsic and extrinsic factors. This study investigates the efficiency and feasibility of six machine learning (ML) algorithms, namely, back-propagation neural network, wavelet neural network, general regression neural network (GRNN), extreme learning machine, support vector machine and random forest (RF), to predict tunneling-induced settlement. Field data sets including geological conditions, shield operational parameters, and tunnel geometry collected from four sections of tunnel with a total of 3.93 km are used to build models. Three indicators, mean absolute error, root mean absolute error, and coefficient of determination the (R²) are used to demonstrate the performance of each computational model. The results indicated that ML algorithms have great potential to predict tunneling-induced settlement, compared with the traditional multivariate linear regression method. GRNN and RF algorithms show the best performance among six ML algorithms, which accurately recognize the evolution of tunneling-induced settlement. The correlation between the input variables and settlement is also investigated by Pearson correlation coefficient.     

基于KPCA-WLSSVM的建筑能耗预测模型

为降低建筑能耗影响因素间复杂相关性对模型性能的影响,建立了一种基于KPCAWLSSVM的建筑能耗预测模型。利用核主元分析(KPCA)对输入变量进行数据压缩,消除变量之间的相关性,简化模型结构;进一步采用加权最小二乘支持向量机(WLSSVM)方法建立建筑能耗预测模型,同时结合一种新型混沌粒子群-模拟退火混合优化(CPSO-SA)算法对模型参数进行优化,以提高模型的预测性能及泛化能力。通过将KPCA-WLSSVM模型方法应用于某办公建筑能耗的预测中,并与WLSSVM、LSSVM及RBFNN模型相比,实验结果表明,KPCA-WLSSVM模型方法能有效提高建筑能耗预测精度。     

Recurrent Neural Networks for Structural Optimization

This paper presents an improvement for an artificial neural network paradigm that has shown significant potential for successful application to a class of optimization problems in structural engineering. The artificial neural network paradigm includes algorithms that belong to the class of single-layer, relaxation-type recurrent neural networks. The suggested improvement enhances the convergence performance and involves a technique that sets the values of weight parameters of the recurrent neural network algorithm. The complete procedure of solving an optimization problem with a single-layer, relaxation-type recurrent neural network is introduced. The discrete Hopfield network is employed to solve the weighted matching problem. A set of simulation experiments is performed to analyze the performance of the discrete Hopfield network. Simulation results confirm that the discrete Hopfield network locates a locally optimal solution after each relaxation once the weight parameters are specified as defined in the suggested technique.     

On-line building energy prediction using adaptive artificial neural networks

While most of the existing artificial neural networks (ANN) models for building energy prediction are static in nature, this paper evaluates the performance of adaptive ANN models that are capable of adapting themselves to unexpected pattern changes in the incoming data, and therefore can be used for the real-time on-line building energy prediction. Two adaptive ANN models are proposed and tested: accumulative training and sliding window training. The computational experiments presented in the paper use both simulated (synthetic) data and measured data. In the case of synthetic data, the accumulative training technique appears to have an almost equal performance with the sliding window training approach, in terms of training time and accuracy. In the case of real measurements, the sliding window technique gives better results, compared with the accumulative training, if the coefficient of variance is used as an indicator.     

Multivariate adaptive regression (MARS) and hinged hyperplanes (HHP) for doweled pavement performance modeling

Standard neural networks in infrastructure performance modeling cannot handle discontinuities in the input training data set, and the performance can in some cases be an issue in the presence of higher frequency and higher order non linearity in pavement condition, traffic and other environmental data. This makes the traditional neural network more of a “black box” with limited physical explanation of the results. This paper is a comparative analysis between multivariate adaptive regression and hinged hyperplanes for doweled pavement performance modeling.     

Building design optimization using a convergent pattern search algorithm with adaptive precision simulations

We propose a simulation–precision control algorithm that can be used with a family of derivative free optimization algorithms to solve optimization problems in which the cost function is defined through the solutions of a coupled system of differential algebraic equations (DAEs). Our optimization algorithms use coarse precision approximations to the solutions of the DAE system in the early iterations and progressively increase the precision as the optimization approaches a solution. Such schemes often yield a significant reduction in computation time.We assume that the cost function is smooth but that it can only be approximated numerically by approximating cost functions that are discontinuous in the design parameters. We show that this situation is typical for many building energy optimization problems. We present a new building energy and daylighting simulation program, which constructs approximations to the cost function that converge uniformly on bounded sets to a smooth function as precision is increased. We prove that for our simulation program, our optimization algorithms construct sequences of iterates with stationary accumulation points. We present numerical experiments in which we minimize the annual energy consumption of an office building for lighting, cooling and heating. In these examples, our precision control algorithm reduces the computation time up to a factor of four.     

Multiclass stand-alone and ensemble machine learning algorithms utilised to classify soils based on their physico-chemical characteristics

This study has provided an approach to classify soil using machine learning. Multiclass elements of stand-alone machine learning algorithms (i.e. logistic regression (LR) and artificial neural network (ANN)), decision tree ensembles (i.e. decision forest (DF) and decision jungle (DJ)), and meta-ensemble models (i.e. stacking ensemble (SE) and voting ensemble (VE)) were used to classify soils based on their intrinsic physico-chemical properties. Also, the multiclass prediction was carried out across multiple cross-validation (CV) methods, i.e. train validation split (TVS), k-fold cross-validation (KFCV), and Monte Carlo cross-validation (MCCV). Results indicated that the soils' clay fraction (CF) had the most influence on the multiclass prediction of natural soils' plasticity while specific surface and carbonate content (CC) possessed the least within the nature of the dataset used in this study. Stand-alone machine learning models (LR and ANN) produced relatively less accurate predictive performance (accuracy of 0.45, average precision of 0.5, and average recall of 0.44) compared to tree-based models (accuracy of 0.68, average precision of 0.71, and recall rate of 0.68), while the meta-ensembles (SE and VE) outperformed (accuracy of 0.75, average precision of 0.74, and average recall rate of 0.72) all the models utilised for multiclass classification. Sensitivity analysis of the meta-ensembles proved their capacities to discriminate between soil classes across the methods of CV considered. Machine learning training and validation using MCCV and KFCV methods enabled better prediction while also ensuring that the dataset was not overfitted by the machine learning models. Further confirmation of this phenomenon was depicted by the continuous rise of the cumulative lift curve (LC) of the best performing models when using the MCCV technique. Overall, this study demonstrated that soil's physico-chemical properties do have a direct influence on plastic behaviour and, therefore, can be relied upon to classify soils.     

Concrete corrosion in wastewater systems: Prediction and sensitivity analysis using advanced extreme learning machine

The implementation of novel machine learning models can contribute remarkably to simulating the degradation of concrete due to environmental factors. This study considers the sulfuric acid corrosive factor in wastewater systems to simulate concrete mass loss using five machine learning models. The models include three different types of extreme learning machines, including the standard, online sequential, and kernel extreme learning machines, in addition to the artificial neural network, classification and regression tree model, and statistical multiple linear regression model. The reported values of concrete mass loss for six different types of concrete are the target values of the machine learning models. The input variability was assessed based on two scenarios prior to the application of the predictive models. For the first assessment, the machine learning models were developed using all the available cement and concrete mixture input variables; the second assessment was conducted based on the gamma test approach, which is a sensitivity analysis technique. Subsequently, the sensitivity analysis of the most effective parameters for concrete corrosion was tested using three different approaches. The adopted methodology attained optimistic and reliable modeling results. The online sequential extreme learning machine model demonstrated superior performance over the other investigated models in predicting the concrete mass loss of different types of concrete.     

基于残差网络的混凝土结构病害分类识别研究

针对混凝土结构病害识别类型单一、精度较低的现状,提出了基于残差网络和迁移学习的病害分类识别方法,通过构建多属性病害数据集,利用迁移学习优化残差网络模型,提出混凝土结构健康状态识别的多个任务。首先收集混凝土结构的病害状态图像,依次通过数据清洗、尺寸均一化、数据扩增和多人投票标注,最终得到包含6 680张图像的混凝土结构病害多属性数据集,并依据不同标注属性进行了相应训练集、验证集和测试集的划分; 然后利用迁移学习对预训练的ResNet-34网络前3个部分进行参数冻结,后续2个部分的参数进行重新训练,并在模型末端添加新的参数,基于已构建的数据集进行训练; 最后在提出的构件类别检测、剥落检测、病害检测和病害类别检测任务中,分别获得84.88%、98.56%、97.18%和85.34%的F1分数。结果表明:通过构建多属性标注的混凝土结构病害数据集训练深度学习模型,可较好地实现多场景特征下的病害识别效果; 采用迁移学习技术可从开源数据中获取较好的特征提取效果; 改进的ResNet-34网络可克服网络退化问题,并针对混凝土结构病害识别的多个任务获得较好的效果; 相对于单一的混凝土结构病害识别,进行病害部位、程度、多类别的系统性检测,可为结构状态评估提供详细信息,更贴合工程实际需要。