Flatness Prediction of Cold Rolled Strip Based on Deep Neural Network with Improved Activation Function

With the improvement of industrial requirements for the quality of cold rolled strips, flatness has become one of the most important indicators for measuring the quality of cold rolled strips. In this paper, the strip production data of a 1250 mm tandem cold mill in a steel plant is modeled by an improved deep neural network (the improved DNN) to improve the accuracy of strip shape prediction. Firstly, the type of activation function is analyzed, and the monotonicity of the activation function is deemed independent of the convexity of the loss function in the deep network. Regardless of whether the activation function is monotonic, the loss function is not strictly convex. Secondly, the non-convex optimization of the loss functionextended from the deep linear network to the deep nonlinear network, is discussed, and the critical point of the deep nonlinear network is identified as the global minimum point. Finally, an improved Swish activation function based on batch normalization is proposed, and its performance is evaluated on the MNIST dataset. The experimental results show that the loss of an improved Swish function is lower than that of other activation functions. The prediction accuracy of a deep neural network (DNN) with an improved Swish function is 0.38% more than that of a deep neural network (DNN) with a regular Swish function. For the DNN with the improved Swish function, the mean square error of the prediction for the flatness of cold rolled strip is reduced to 65% of the regular DNN. The accuracy of the improved DNN is up to and higher than the industrial requirements. The shape prediction of the improved DNN will assist and guide the industrial production process, reducing the scrap yield and industrial cost.

Network representation learning method embedding linear and nonlinear network structures

With the rapid development of neural networks, much attention has been focused on network embedding for complex network data, which aims to learn low-dimensional embedding of nodes in the network and how to effectively apply learned network representations to various graph-based analytical tasks. Two typical models exist namely the shallow random walk network representation method and deep learning models such as graph convolution networks (GCNs). The former one can be used to capture the linear structure of the network using depth-first search (DFS) and width-first search (BFS), whereas Hierarchical GCN (HGCN) is an unsupervised graph embedding that can be used to describe the global nonlinear structure of the network via aggregating node information. However, the two existing kinds of models cannot simultaneously capture the nonlinear and linear structure information of nodes. Thus, the nodal characteristics of nonlinear and linear structures are explored in this paper, and an unsupervised representation method based on HGCN that joins learning of shallow and deep models is proposed. Experiments on node classification and dimension reduction visualization are carried out on citation, language, and traffic networks. The results show that, compared with the existing shallow network representation model and deep network model, the proposed model achieves better performances in terms of micro-F1, macro-F1 and accuracy scores.

Bio-inspired computing by nonlinear network dynamics - a brief introduction

The field of bio-inspired computing has established a new frontier for conceptualizing information processing, aggregating knowledge from disciplines as different as neuroscience, physics, computer science and dynamical systems theory. The study of the animal brain has shown that no single neuron or neural circuit motif is responsible for intelligence or other higher-order capabilities. Instead, complex functions are created through a broad variety of circuits, each exhibiting an equally varied repertoire of emergent dynamics. How collective dynamics may contribute to computations still is not fully understood to date, even on the most elementary level. Here we provide a concise introduction to bio-inspired computing via nonlinear dynamical systems. We first provide a coarse overview of how the study of biological systems has catalyzed the development of artificial systems in several broad directions. Second, we discuss how understanding the collective dynamics of spiking neural circuits and model classes thereof, may contribute to and inspire new forms of "bio-inspired" computational paradigms. Finally, as a specific set of examples, we analyze in more detail bio-inspired approaches to computing discrete decisions based on multi-dimensional analogue input signals, via $k$-winners-take-all functions. This article may thus serve as a brief introduction to the qualitative variety and richness of dynamical bio-inspired computing models, starting broadly and focusing on a general example of computation from current research. We believe that understanding basic aspects of the variety of bio-inspired approaches to computation on the coarse level of first principles (instead of details about specific simulation models) and how they relate to each other, may provide an important step towards catalyzing novel approaches to autonomous and computing machines in general.

Characteristics and Challenges in Prosumer-Dominated Thermal Networks

Prosumer-dominated thermal networks interconnect distributed prosumers. These networks form the infrastructure that allows to execute the trading of thermal energy as desired in the context of local energy markets. However, the significantly different behavior of such networks compared to conventional district heating and cooling networks has not yet been comprehensively investigated. This paper provides a compilation of instrinsic characteristics of prosumer-dominated thermal networks and discusses challenges that arise from these characteristics. As a basis for the investigations an underlying reference concept for the considered type of networks is described. Simulative case studies are combined with scientific deduction and literature references to gain new insights on the design and operation of this type of networks. It is found that due to the variability in these networks, the definition of a design case is a challenge for the dimensioning of concrete network implementations. To anticipate the strong coupling between prosumers and the nonlinear network behavior, it is concluded that centralized control combined with a model of the physical network behavior is necessary. The discussion of characteristics and challenges in prosumer-dominated thermal networks indicates open points in this field and thus provides a starting point for consecutive studies.

Pattern Formation in a RD-MCNN with Locally Active Memristors

This chapter presents the mathematical investigation of the emergence of static patterns in a Reaction–Diffusion Memristor Cellular Nonlinear Network (RD-MCNN) structure via the application of the theory of local activity. The proposed RD-MCNN has a planar grid structure, which consists of identical memristive cells, and the couplings are established in a purely resistive fashion. The single cell has a compact design being composed of a locally active memristor in parallel with a capacitor, besides the bias circuitry, namely a DC voltage source and its series resistor. We first introduce the mathematical model of the locally active memristor and then study the main characteristics of its AC equivalent circuit. Later on, we perform a stability analysis to obtain the stability criteria for the single cell. Consequently, we apply the theory of local activity to extract the parameter space associated with locally active, edge-of-chaos, and sharp-edge-of-chaos domains, performing all the necessary calculations parametrically. The corresponding parameter space domains are represented in terms of intrinsic cell characteristics such as the DC operating point, the capacitance, and the coupling resistance. Finally, we simulate the proposed RD-MCNN structure where we demonstrate the emergence of pattern formation for various values of the design parameters.