Soft sensing of water depth in combined sewers using LSTM neural networks with missing observations

Information and communication technologies combined with in-situ sensors are increasingly being used in the management of urban drainage systems. The large amount of data collected in these systems can be used to train a data-driven soft sensor, which can supplement the physical sensor. Artificial Neural Networks have long been used for time series forecasting given their ability to recognize patterns in the data. Long Short-Term Memory (LSTM) neural networks are equipped with memory gates to help them learn time dependencies in a data series and have been proven to outperform other type of networks in predicting water levels in urban drainage systems. When used for soft sensing, neural networks typically receive antecedent observations as input, as these are good predictors of the current value. However, the antecedent observations may be missing due to transmission errors or deemed anomalous due to errors that are not easily explained. This study quantifies and compares the predictive accuracy of LSTM networks in scenarios of limited or missing antecedent observations. We applied these scenarios to an 11-month observation series from a combined sewer overflow chamber in Copenhagen, Denmark. We observed that i) LSTM predictions generally displayed large variability across training runs, which may be reduced by improving the selection of hyperparameters (non-trainable parameters); ii) when the most recent observations were known, adding information on the past did not improve the prediction accuracy; iii) when gaps were introduced in the antecedent water depth observations, LSTM networks were capable of compensating for the missing information with the other available input features (time of the day and rainfall intensity); iv) LSTM networks trained without antecedent water depth observations yielded larger prediction errors, but still comparable with other scenarios and captured both dry and wet weather behaviors. Therefore, we concluded that LSTM neural network may be trained to act as soft sensors in urban drainage systems even when observations from the physical sensors are missing.     

Electronic synaptic characteristics and simulation application of Ag/CeO2/Pt memristor

With the development of science and technology, artificial neural networks (ANNs) have become a research spot. Furthermore, two-terminal oxide memristors with adjustable resistance have attracted extensive attention due to their simple structure, low power consumption, and easy integration, among other attractive features. Additionally, among many oxides, ceria has exhibited good performance, such as longer retention and better stability in resistive devices; however, it was currently rarely used in artificial neural synapses. In this work, a self-designed Ag/CeO₂/Pt memristor was found to realize the slow transition between the high-resistance state (HRS) and low-resistance state (LRS) at a very small working voltage. It was also found to exhibit very good retention performance and cyclic characteristics. The conductivity of the device was analyzed by the current-voltage (I–V) characteristics curve. Furthermore, its artificial synaptic function was explored, and a series of neuromorphic systems simulations were carried out. Additionally, the relationships between the pulse sequence parameters and the resistance state of the device were explored, and an electrical signal simulation of Pavlov's dog experiment was designed. The findings demonstrated that the device could be used to realize the application of artificial neural synapse simulation.     

Modeling dynamic construction work template from existing scheduling records via sequential machine learning

Today, construction planning and scheduling is almost always performed manually, by experienced practitioners. The knowledge of those individuals is materialized, maintained, and propagated through master schedules and look-ahead plans. While historical project schedules are available, manually mining their embedded knowledge to create generic work templates for future projects or revising look-ahead schedules is very difficult, time-consuming and error-prone. The rigid work templates from prior research are also not scalable to cover the inter and intra-class variability in historical schedule activities. This paper aims at fulfilling these needs via a new method to automatically learn construction knowledge from historical project planning and scheduling records and digitize such knowledge in a flexible and generalizable data schema. Specifically, we present Dynamic Process Templates (DPTs) based on a novel vector representation for construction activities where the sequencing knowledge is modeled with generative Long Short-Term Memory Recurrent Neural Networks (LSTM-RNNs). Our machine learning models are exhaustively tested and validated on a diverse dataset of 32 schedules obtained from real-world projects. The experimental results show our method is capable of learning planning and sequencing knowledge at high accuracy across different projects. The benefits for automated project planning and scheduling, schedule quality control, and automated generation of project look-aheads are discussed in detail.     

Geometry optimization and pressure analysis of a proton exchange membrane fuel cell stack

For proton exchange membrane fuel cells (PEMFCs), the distribution of reactant streams in the reactor is critical to their efficiency. This study aims to investigate the optimal design of the inlet/outlet flow channel in the fuel cell stack with different geometric dimensions of the tube and intermediate zones (IZ). The tube-to-IZ length ratio, the IZ width, and the tube diameter are adjusted to optimize the geometric dimensions for the highest pressure uniformity. Four different methods, including the Taguchi method, analysis of variance (ANOVA), neural network (NN), and multiple adaptive regression splines (MARS), are used in the analyses. The results indicate the tube diameter is the most impactive one among the three factors to improve the pressure uniformity. The analysis suggests that the optimal geometric design is the tube-to-IZ length ratio of 9, the IZ width of 14 mm, and the tube diameter of 9 mm with the pressure uniformity of 0.529. The relative errors of the predicted pressure uniformity values by NN and MARS under the optimal design are 1.62% and 3.89%, respectively. This reveals that NN and MARS can accurately predict the pressure uniformity, and are promising tools for the design of PEMFCs.     

Multi-Level Fusion Net for hand pose estimation in hand-object interaction

This work is about solving a challenging problem of estimating the full 3D hand pose when a hand interacts with an unknown object. Compared to isolated single hand pose estimation, occlusion and interference induced by the manipulated object and the clutter background bring more difficulties for this task. Our proposed Multi-Level Fusion Net focuses on extracting more effective features to overcome these disadvantages by multi-level fusion design with a new end-to-end Convolutional Neural Network (CNN) framework. It takes cropped RGBD data from a single RGBD camera at free viewpoint as input without requiring additional hand–object pre-segmentation and object or hand pre-modeling. Through extensive evaluations on public hand–object interaction dataset, we demonstrate the state-of-the-art performance of our method.     

3D PdPb nanochain networks with efficient catalytic performance for ethylene glycol electrooxidation

The noble metal anodic catalysts with three-dimension (3D) chain-like network structure have been researched thoroughly due to their unique morphological characteristics. Herein, a novel scheme has been designed rationally to synthesize 3D PdPb nanochain networks (PdPb NCNs). Numerous nanochains were interlaced and stacked to form the nanonetworks, which were contributed to improving electrocatalytic performance. Abundant steps and kinks existed on the nanochain networks, which provided plentiful active sites to improve the electrocatalytic activity. In the subsequent electrochemical tests, the mass activity of Pd₆₅Pb₃₅ NCNs was 4.47 A mg _pd^?1, higher than other catalysts. Moreover, in the chronoamperometry and consecutive CV measurements, Pd₆₅Pb₃₅ NCNs exhibited the best stability than other prepared samples. This work explored the rational synthesis of PdPb nanochain networks, and confirmed the excellent electrocatalytic performance in EGOR.     

Growing topology representing network

This paper describes a method for finding the topology of a data distribution online using a new growing graph network architecture. Many growing neural networks for finding the topology of data online, such as the Growing Neural Gas, depend on the order and number of input data. For this reason, conventional methods have certain drawbacks: weakness to noise, generating redundant nodes, requiring a great deal of input data, and so on. The proposed method is robust with respect to these issues since it has been developed from the viewpoint of a generative model. This paper presents both the theory and an algorithm in this paper. Moreover, the effectiveness of the proposed method is shown by experiments comparing the proposed method with various growing graph networks.     

Effect of chain length distribution on thermal characteristics of model polytetrahydrofuran (PTHF) networks

A series of model polytetrahydrofuran (PTHF) networks were synthesized via end-linking reactions of α, ω-allyl PTHF oligomers with a stoichiometric tetrafunctional crosslinker. The telechelic PTHF oligomers were synthesized by living cationic ring-opening polymerization of tetrahydrofuran followed by a termination reaction with allyl alcohol. Networks thus prepared have well-controlled architecture in terms of the inter-crosslink chain length (M_c) and chain length distribution: resulting in unimodal, bimodal and clustered structures. Unimodal network was prepared by using polymer chains of same molecular weight, bimodal networks were synthesized by using two groups of polymer chains with different average molecular weights, and the clusters are prepared by incorporating clusters of networks with small molecular weight chains in a network matrix made of longer chains. Thermal characteristics of these model networks were investigated as a function of crosslink density, as well as inhomogeneities of crosslink distribution using DSC. We demonstrate that glass transition temperature (T_g) and crystallization behavior (melting temperature and crystallinity) of the networks are both strongly influenced by crosslink density (M_c). By comparing the unimodal, bimodal and clustered networks with similar average M_c, the effects of inhomogeneities in the crosslink distribution on the thermal properties were also investigated. Results show that inhomogeneities have trivial influence on T_g, but strongly affects the crystallization behavior. Moreover, the effects of the content ratio and length ratio between long and short chains, and the effects of cluster size and size distribution on the thermal characteristics were also studied.     

基于WTD-AR谱和MEA-BPNN的轴承故障诊断方法

针对滚动轴承故障诊断模型在噪声干扰下鲁棒性能差的问题,提出一种基于小波阈值去噪（WTD）、AR谱和思维进化算法（MEA）优化反向传播神经网络（BPNN）的轴承故障诊断方法。以原始振动信号为输入,采用小波方法分解重构原始信号滤除高频噪声,然后采用Burg算法估计AR模型参数提取降噪信号功率谱特征,最后将特征向量与对应标签分别作为MEA-BPNN神经网络的输入、输出进行训练,最终实现诊断。将该方法与一些先进的人工神经网络诊断方法作比较,测试该诊断模型的性能。研究结果表明：WTD-AR谱-MEA-BPNN诊断模型能够有效降低轴承振动信号的噪声干扰,实现特征增强,分辨率更高;相较于传统神经网络训练速度更快,在更短时间内甄别故障类型且识别率高。     

密炼机液压上顶栓压力人工神经网络控制系统算法研究

阐述了BP神经网络控制算法原理;论证了人工神经网络相对于经典PID控制在上顶栓压力闭环控制上优越性;建立了系统的机电液耦合模型,通过MATLAB仿真实验,研究BP算法在该系统中的动态特性,为神经网络在该控制系统中的应用提供实验依据;最后仿真结果表明,引入BP算法可以使系统超调量明显下降,响应时间显著得到提高,静差得到很好的消除。