Current Issue

20 June 2020, Volume 2 Issue 2

Previous Issue    Next Issue

Review Intelligence

Reinforcement learning:toward action-knowledge merged intelligent mechanisms and algorithms

Fei-Yue WANG,Dongpu CAO,Qinglai WEI

2020, 2(2):  101-106.  doi:10.11959/j.issn.2096-6652.202011

Asbtract ( 1059 )   HTML ( 223)   PDF (1260KB) ( 1463 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

This article discusses briefly the history,the state of the art and the future of reinforcement learning,and outlines a roadmap of evolution from learning by doing,doing with planning to parallel intelligence that combining learning virtually in artificial systems and acting accordingly in actual systems.

Surveys and Prospectives

The application of deep learning in data-driven modeling of process industries

Xiaofeng YUAN,Yalin WANG,Chunhua YANG,Weihua GUI

2020, 2(2):  107-115.  doi:10.11959/j.issn.2096-6652.202012

Asbtract ( 880 )   HTML ( 143)   PDF (1147KB) ( 1572 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Deep learning is an artificial intelligence technique developed in recent years.Compared with traditional shallow models,deep learning has a strong ability of feature representation and function fitting.It can extract hierarchical features from massive data,which has great potential for data-driven modeling in process industries.Firstly,the history of deep learning was introduced.Then,four widely used deep networks and their applications were introduced in data-driven modeling of process industries.At last,the conclusions about deep learning and its applications in process in dustries were given.

Brain abnormalities in depression based on multimodal imaging

Shan LI,Yongchao LI,Ying ZOU,Lin YANG,Yin WANG,Zhijun YAO,Ubin H

2020, 2(2):  116-125.  doi:10.11959/j.issn.2096-6652.202013

Asbtract ( 599 )   HTML ( 67)   PDF (4051KB) ( 873 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

In recent years,structural magnetic resonance imaging (sMRI) and functional magnetic resonance imaging (fMRI) are widely used in depression research.From the perspectives of morphology,structural network and functional network,the brain abnormalities of depression were explored to understand the pathogenesis,and to assist doctors in clinical diagnosis,treatment and prognosis.A large number of researches have found that the hippocampus and amygdala of depression showed different degrees of atrophy,and the connection strength of brain network and graph theory attributes showed significant abnormal.Moreover,the abnormal brain areas were related to emotional regulation,attention and cognitive control,and the degree of abnormalities were highly correlated with the severity of depression.The research actuality of depression from different perspectives was reviewed,and some suggestions for future research were put forward.

Regular Papers

An intelligent optimization scheduling method for community patrolling and investigation in epidemic situations

Xin CHEN,Jiayu WU,Xue WU,Minxia ZHANG,Yujun ZHENG

2020, 2(2):  126-134.  doi:10.11959/j.issn.2096-6652.202014

Asbtract ( 227 )   HTML ( 34)   PDF (1191KB) ( 348 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Community plays an important role in epidemic prevention and control in the society.Based on the practical experience of community prevention and control of the novel corona virus pneumonia,a community patrolling and investigation scheduling problem in epidemic situations was presented,which estimated a risk value for each high-risk household based on epidemic-related information,and then scheduled the community staffs to patrol and investigate the high-risk households as efficiently as possible.To solve this problem,a hybrid intelligent optimization scheduling algorithm was proposed,which explored the solution space based on the search strategy of the water wave optimization (WWO) met heuristic and improved solution accuracies using two local search strategies.Whenever an exceptional case was detected during the patrolling and investigation,the solution was dynamically adapted to the changed situation.Computational results on the real-world cases of community patrolling and investigation in Zhejiang Province,China demonstrate the effectiveness and efficiency of the proposed method.

A novel blockchain-based surveillance and early-warning technology for infectious diseases

Liwei OUYANG,Yong YUAN,Xinhu ZHENG,Jun ZHANG,Fei-Yue WANG

2020, 2(2):  135-143.  doi:10.11959/j.issn.2096-6652.202015

Asbtract ( 573 )   HTML ( 65)   PDF (1285KB) ( 1347 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

The key of early prevention and control of infectious diseases is to use the early warning technology and system to monitor the abnormal occurrence and trend of infectious diseases.There are still some problems in the existing infectious diseases automated-alert and response systems,such as lack of intelligence,poor efficiency in exchange of key information,as well as difficulties in distributed decision-making.Based on the distributed blockchain architecture,a novel blockchain-based surveillance and early-warning technology for infectious diseases was proposed,leveraging emerging information technologies including artificial intelligence,big data and smart contract.The technology could aggregate monitoring forces from multiple parties efficiently,integrate various early warning technologies flexibly,and establish a distributed and collaborative monitoring environment for knowledge integration and intelligent interconnection with guaranteed security and privacy protection.In this framework,smart contract was served as a “software-defined agent” to fuse decisions,monitor the outbreak,and issue warnings in an automatic and real-time fashion,so as to meet the key requirements of accuracy and timeliness,and avoid the decision bias of single evidence.

Fusion of discourse structural position encoding for neural machine translation

Xiaomian KANG,Chengqing ZONG

2020, 2(2):  144-152.  doi:10.11959/j.issn.2096-6652.202016

Asbtract ( 296 )   HTML ( 20)   PDF (898KB) ( 479 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Most of existing document-level neural machine translation (DocNMT) methods focus on exploring the utilization of the lexical information of context,which ignore the structural relationships among the cross-sentence discourse semantic units.Therefore,multiple discourse structural position encoding strategies were proposed to represent the positional relationships among the words in discourse units over the discourse tree based on rhetorical structure theory (RST).Experimental results show that the source-side discourse structural position information is effectively fused into the DocNMT models underlying the Transformer architecture by the position encoding,and the translation quality is improved significantly.

Research on certificate information authentication method of expressway toll based on blockchain technology

Xiaoming LIU,Chunlin SHANG,Jie ZHANG,Jie CHEN,Xiangda WEI,Guiqing ZHU

2020, 2(2):  153-160.  doi:10.11959/j.issn.2096-6652.202017

Asbtract ( 220 )   HTML ( 21)   PDF (1140KB) ( 493 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

With the continuous development of expressways convenient and non-feeling toll collection,reliable travel trajectory certification has gradually become the key support for fair and accurate toll collection.In view of the problems of high vehicle inspection delay,high time consuming of massive data retrieval,and low tampering data inspection accuracy,the multi-agent and multi-stage operation characteristics of expressway vehicles were considered,started with the dynamic certification of high-speed vehicle driving process and the safety association certification of vehicle information,the characteristics of multi-stage authentication content analysis,distributed network authentication consensus,vehicle information main chain association and vehicle trajectory side-chain tracking were considered,a reliable driving track authentication method for expressway based on heterogeneous blockchain technology was proposed,which realized the real-time inspection of vehicle information and the reliable track authentication of expressway vehicles based on track information chain association.Finally,through the introduction of typical application scenarios of the method,the prospect and value of the method in the certification of highway vehicle trajectory were further analyzed.

Research on path curvature smoothing method based on energy function for intelligent vehicles

Qiang SHI,Ming YANG

2020, 2(2):  161-168.  doi:10.11959/j.issn.2096-6652.202018

Asbtract ( 466 )   HTML ( 32)   PDF (5085KB) ( 699 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Intelligent vehicles are of great significance in social efficacy such as prevention of traffic accidents,ease traffic congestion and reducing emissions.Path tracking is an important part of intelligent vehicles function,whose controlled object is a complex vehicle-road coupling system.Path curvature smoothness affects path tracking performance.A curvature smoothing method based on energy function was proposed to optimize the smoothness of path curvature.Firstly,an energy function was constructed to characterize the path curvature smoothness.Secondly,aniteration rule was designed to reduce the energy function based on discrete path description.Finally,the constraint of road boundary was considered when the road boundary was known.The experimental results of the real vehicle platform show that the energy function smoothing method can further improve the ride comfort and control accuracy of path tracking on the premise of ensuring the path feasible.The quality of intelligent vehicles is improved.

Research on data-driven modeling for photovoltaic characteristics based on hybrid neural network

Guobin ZHANG,Xinying WANG

2020, 2(2):  169-178.  doi:10.11959/j.issn.2096-6652.202019

Asbtract ( 269 )   HTML ( 17)   PDF (7217KB) ( 477 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

A data-driven modeling approach of photovoltaic modules based on hybrid structure neural network was proposed to solve the issue that traditional physical mechanism modeling methods are difficult to be applied under some complex lighting conditions.Based on an in-depth analysis of the physical mechanism and output characteristics of photovoltaic modules,the convolutional neural network and radial basis function network were used to extract features of several environmental factors such as uneven lighting conditions,temperature and humidity.In order to improve the fitting effect of the hybrid network model,an equivalent analysis method of shadow morphology for uneven lighting conditions was proposed,a genetic coding scheme was used to optimize network parameters.Model effect was verified by the actual operation data.The model had generalized tracking capacity under various uneven lighting conditions,and the average error of simulation results was kept within 7%.

A fuzzy system optimization modeling method based on improved genetic algorithm and support degree

Hongqing DU,Dewang CHEN,Yunhu HUANG,Fenghua ZHU,Lingxi LI

2020, 2(2):  179-185.  doi:10.11959/j.issn.2096-6652.202020

Asbtract ( 341 )   HTML ( 26)   PDF (2024KB) ( 632 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Fuzzy system is a kind of artificial intelligence method with strong explanatory ability.The classical Wang-Mendel (WM) method can automatically obtain fuzzy rules from data,and it has become an important intelligent modeling method.There are many problems in this method,such as the large number of rules and the low precision.And at present,there are many problems in the improved methods,such as complex calculation and low efficiency.For this reason,a new method of fuzzy system optimization modeling based on improved genetic algorithm and rule reduction based on support degree:genetic fuzzy system (GFS) was proposed.By optimizing the structure and membership function parameters of the fuzzy system,the concrete algorithm of GFS1,GFS2 and GFS3 models were constructed by different combinations of objective functions.The results of fuzzy modeling experiments on the standard and noisy power output data sets show that:1) GFSi (i =1,2,3) model fitting accuracy is higher than WM method and the number of rules is less; 2) its anti-noise capability is significantly better than that of RBF and BP neural network; 3) the fitness function of GFS3 has the best evaluation effect,so its performance is optimal.The method proposed in this paper gives full play to the advantages of the interpretability and robustness of the fuzzy system and guarantees the accuracy at the same time.It is a potential artificial intelligence algorithm.

Dynamic optimization algorithm of cement firing system based on differential evolution

Xiaochen HAO,Yakun JI,Lizhao ZHENG,Xin SHI,Yantao ZHAO

2020, 2(2):  186-193.  doi:10.11959/j.issn.2096-6652.202021

Asbtract ( 227 )   HTML ( 16)   PDF (1579KB) ( 694 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Aiming at the problem of resource waste in the process of cement firing and the difficulty of establishing an effective mathematical mechanism model,a dynamic energy optimization method based on the cement industry firing system was proposed.The method used the convolutional neural network to construct the objective function of power consumption and coal consumption of the firing system.The differential evolution algorithm was used to solve the control parameters in reverse,and the better operating index was obtained according to the current working conditions.Since the actual production conditions will change with time,the operating indicators and power consumption and coal consumption in the future will be saved,and then input into the neural network for training,and the constraint range will be determined by the actual running index value at the current time.The optimization value can meet the actual operation index adjustment requirements.Furthermore,the goal optimization of the dynamic energy consumption state of the cement firing process was realized.It effectively reduces the energy consumption of cement firing process.

Parallel security:generative adversarial systems for intelligent security in CPSS

Yidong LI,Jun ZHANG,Yaodong TAO,Wei WANG,Yuanxiang GU,Fei-Yue WANG

2020, 2(2):  194-202.  doi:10.11959/j.issn.2096-6652.202022

Asbtract ( 421 )   HTML ( 46)   PDF (2439KB) ( 1116 )   Knowledge map   

Figures and Tables | References | Related Articles | Metrics

Developing technologies such as AI and big data are changing the landscape of cybercrime for both attackers and defenders.Traditional defense technology mainly use the passive detect-then-defend model,which is hard to meet the security protection requirements brought by new attack characteristics.A novel security framework,named parallel security was proposed,which aimed at providing a comprehensive solution with descriptive security,predictive security and prescriptive security.The main idea was based on the ACP theory of parallel intelligent,and integrated new methods such as generative adversarial leaning model,parallel blockchain.Human elements,sociology and psychology were taken as an important component in the framework,in order to enhance the immunity of the system.In addition,a parallel security based platform was design and developed for several typical scenarios in the field of industrial internet as a proof of concept.