Research Progress of Parallel Control and Management

Gang Xiong, Xisong Dong, Hao Lu, and Dayong Shen

Abstract—Based on ACP (artificial systems, computational experiments, and parallel execution) methodology, parallel control and management has become a popularly systematic and complete solution for the control and management of complex systems. This paper focuses on summarizing comprehensive review of the research literature of parallel control and management achieved in the recent years including the theoretical framework,core technologies,and the application demonstration.The future research, application directions, and suggestions are also discussed.

I. INTRODUCTION

FROM Wiener’s [1] cybernetics to Tsien’s [2] engineering cybernetics,advanced control and intelligent control have become two hot research topics in modern science and technology.However,most researches carried out in advanced control and intelligent control concern only physical systems that can be expressed by mathematical equations.For example,the CPS(cyber-physical systems), proposed in 2007, has now become one of the globally hottest research topics [3], but it is still a kind of complex system dominated by engineering complexity.In recent years, the popularity of new information communication technologies (ICT), like IoT (internet of things), social media, big data, and so on, has fundamentally changed the management, control and operation mode of many systems.Beside physical elements, more and more social elements are involved in these systems and resulted in various CPSSs(cyber physical social systems) with different UDC (uncertainty,diversity and complexity) features [4]. Now, the control and management problems of those complex systems are facing the hitherto unknown challenges and development opportunities,which mainly include three scientific problems, namely, 1)how to organically unify the overall modeling and reduction analysis of complex systems? 2) how to predict and guide the complex systems with human being in the loop and including both social complexity and engineering complexity? 3) based on 1) and 2), how to realize the control and management of complex systems with AFC (agility, focus and convergence)through interactive feedback and co-evolution?

Qianet al.[5] initiated the innovative research on open complex giant system from the perspective of system science. Their work promoted the analysis of complex systems from qualitative stage to quantitative stage, and two notable achievements are the integrated scientific thought of humanmachine combination and hall for workshop of metasynthetic engineering (HWME). In 1999, the journalSciencepublished the complex system monograph [6], which is regarded as one milestone in the process of complex systems research. Wang proposed ACP methodology (artificial systems, computational experiments, parallel execution) in 2004 for more efficient management and control of the complex system without accurate models [7]. Inspired by Everett’s “multiple world”or “parallel world” explanation about Quantum mechanics[8] and Karl three worlds (physical world, mental world,and artificial world) theory [9], ACP methodology [7], [10]integrates information, psychology, simulation, decision into a whole framework, and provides new ideas and methods for the control and management of complex systems which are hardly calculated, operated and implemented using traditional methods. To our knowledge, up to now, ACP methodology is the sole systematic and complete research framework for complex systems.

Based on the ACP methodology, parallel control and management combines three kinds of scientific research methods(theoretical research, experimental method, and calculation technology), and aims to effectively deal with complex systems like CPSS. Parallel control and management can help us to improve the understanding and analysis ability of the system dynamic evolution rules and interactions of the elements in complex systems, and improve the management and control ability in dealing with all kinds of accidents in abnormal states,and thus provide an innovative and effective technical tool for the management and control of complex systems.

The paper focuses on the fruitful research progress about the theoretical framework, core technologies, and application demonstrations of parallel control and management achieved in recent years, and looks forward to the future development and suggestions.

The rest of the paper is organized as follows: Section II introduces the research progress overview of parallel control and management; Section III presents its research progress in detail; Section IV shows some application practices; Section V draws the conclusion.

II. RESEARCH PROGRESS OVERVIEW

In recent years, parallel control and management has become an important and hot topic in China, more and more researchers are joining, and they have achieved important progress in basic theories, core technologies and application practices,etc.Since 2009,the national parallel control conference and the national parallel management conference were held once a year by CAA (Chinese Association of Automation).In 2011,CAA established the Specialized Committee of Parallel Control and Management.

From the Chinese science citation database (CSCD), totally 510 journal articles are searched from Jan. 2004 to Jun. 2017.Fig.1 shows the statistical data of the number of papers according to their published year. Table I shows the main journals in the field of parallel control and management.Fig.2 shows the hotspot distribution, in which computational experiments, artificial society and parallel system are the top three hotspot keywords. The number of papers published by the research institutions and their number of citations are shown in Table II. Fei-Yue Wang from the Institute of Automation,Chinese Academy of Sciences (CASIA), Xiaogang Qiu and Bin Chen from National University of Defense Technology,Xiong Xiong, Wei Zhang, and Yongjie Zhang from Tianjin University, and Xiao Xue from Henan Polytechnic University,are important researchers in parallel control and management(Fig.3). At the same time, some foreign researchers are using the methods similar to ACP and parallel system [11], [12].

III. RESEARCH PROGRESS IN DETAIL

Before 2012,the research team of CASIA initiated research on the theoretical framework of ACP methodology and parallel control and management, and developed some core technologies and several successful application pilots in transportation systems and petrochemical production.Since 2013,with more and more related projects granted by NSFC (Natural Science Foundation of China), S&T funds of local governments and industries,more and more research units have been constantly involved.The research and development of ACP methodology focus on: the modeling, experiment and decision of complex systems,where the emphases are data,knowledge and decision respectively; big data driven artificial systems construction method for complex systems like CPSS;the scenario handlings knowledge bases for computational experiments; the parallel control and management method based on virtual-reality interaction, etc. The research framework can be summarized and shown in Fig.4.The research progress and main achievements are described in detail as follows.

TABLE I THE JOURNALS AND THE NUMBER OF PUBLISHED PAPERS

TABLE II THE MAIN RESEARCH INSTITUTES AND THE NUMBERS OF PAPERS AND RELEVANT CITATIONS

A. Artificial Systems

In the research of artificial system, agent-based modeling method is one of the most popular methods [13], [14]. Cuiet al.[15] mainly summarized some models of complex systems involving human and social complexity,mainly including the modeling of individual behaviors, organizational structure and environment.Individual behaviors’modeling methods can be broadly divided into probabilistic model, rule-based model and game based evolution model. The early modeling of organization structure is based on simple topological structures,such as square grid, regular network, complete network, etc.And, the social network modeling based on complex network can better describe the interaction between individuals. For example, after applying the complex network model into the Sznajd model [16], the statistical properties of the empirical studies can be reproduced, and the small world characteristic speeds up the information spread in society.The environmental modeling (artificial environment) can be divided into two types, the design type based on researchers hypothesis, and the analysis type based on the actual data.

Fig.2. Research point distribution graph (The deeper the color, the more number of related papers).

Fig.3. The main researchers and their cooperation relations.

Fig.4. Parallel control and management theory for CPSS.

Tanget al.[13] put forward and developed an innovative ASML (artificial society modeling language), including its meta model, multi-view model, graphical modeling language,and so on. Based on the existing multi-agent system technology and social organization theory, ASML can realize the modeling and analysis of artificial system more easily. ASML Tool can be used to support the ASML based modeling,model checking, and model transformation.

According to Qiuet al.[17]−[19], the tool development,data acquisition, model construction, resource sharing and other requirements need to be considered for the construction of artificial systems, and the development environment should include a series of good and useful tools,useful models for all elements which can be reused and combined,and the complete and reliable data for different ranges and levels.These requirements can be fulfilled by means of field simulation knowledge and engineering idea. Therefore, it is necessary to establish an artificial system construction environment of “resource +platform + artificial system”.

B. Computational Experiments

In order to solve the experimental evaluation problem of complex systems, Wang proposed computational experiments method in 2004, whose core idea is to take computer as“Laboratory” [9]. The computational experiments provide a novel tool for analyzing complex systems. Cuiet al.reviewed the research method and application progress of computational experiments [2], including model construction, experiments design, and experiments implementation, etc. A scientific and perfect computational experiment design is very important,which includes a variety of prediction/inference, online/parallel execution for different objects,and complex multi-variation and bottom-up generation for multiple worlds. In general,computational experiments are used to study the performance of systems or processes. In order to rapidly and efficiently determine the optimal experimental conditions affecting the experimental results, the existing computational experiments can be divided into the following types [19]−[22]:

1)According to the number of quantity factors of computational experiments, it can be divided into the computational experiments for single factor optimization and for multifactors optimization.The computational experiments for single factor optimization includes such optimization methods as average method,golden section method,bisection method,and Fibonacci method. The computational experiments for multifactors optimization need to determine the number of factors and the variables range,so that these factors can be measured,and then the computational experiments by using orthogonal design and uniform design method can be designed.

2)According to the different purposes,computational exper-iments can be divided into two types: indicator optimization and robust optimization.The indicator optimization is adopted for maximizing the experimental indicators, and the robust optimization is adopted for minimizing the fluctuation of the experimental indicators.

3) According to the different processes, computational experiments can be divided into two types,i.e.sequential design and integral design.The sequential design starts from a starting point, determines the experimental position according to the previous experimental results, and optimizes the experimental indicators, such as 0.618 method, dichotomy method. The integral design requires that the experimental points can be evenly distributed among all possible experimental points,and the optimal experimental conditions are determined according to the experimental results,such as orthogonal design,uniform design.

But, there also exist many problems and challenges in computational experiments such as model verification, model design and other areas,because their application fields are extensive.For example:the various models of the computational experiments are difficult to compare each other for their verification; the design methods are lacking generality, so many classical experimental design methods are not suitable for the computational environment; the increasing computational consumptions need the support of scalable and extensible software and hardware; it is hard to balance the modeling flexibility and the conclusions credibility; the commonalities between the experimental experiments need to be found for abstraction,and more general experiments design and analysis methods need to be designed.

C. Parallel Control

In the aspect of theory research,the basic concepts,methods and applications of parallel control are improved, and the differences between the virtual-reality interaction of extended parallel computing and the parallel partition method are clarified. Parallel control is the specific application of ACP methodology in the control fields, and it is a kind of data driven computational control method.Parallel control includes the modeling and representation by using artificial systems,the analysis and evaluation through computational experiments,and finally the complex systems control and management with the help of parallel execution [23]. ADP (adaptive dynamic programming) is an important method to realize parallel control. ADP breakthroughs the data-driven analysis, control and optimization method of nonlinear systems, and provides one useful solution for the realization of parallel control. The generalized value iteration ADP method with non-zero initial conditions is proposed [24]. For those discrete time nonlinear systems, the error based ADP algorithm is designed, and the generalized policy iteration ADP method is proposed to obtain the convergence condition of the performance indicator function [25]. For MIMO nonlinear discrete system, ADP method is studied for the design of tracking controller of nonlinear systems[26].For the optimal control problems of the nonlinear continuous system with unknown internal information, the observation-evaluation system structure is designed using ADP method,and the optimal controllers design method is given out [27]. For the linear control system with noise,data based controllability and observability analysis method is proposed, and the methods unbiasedness and consistency for controllability and observability matrix estimation are proven[28]. For nonlinear continuous time systems with unknown dynamics, online synchronous approximate optimal learning algorithm is put forward to solve multi-player non-zero-sum games problem [29], [30]. For continuous time nonlinear systems with uncertainties,the robust controller design method for online policy iteration based optimal control is proposed to construct the optimal robust ADP method, and get the selflearning robust control method for uncertain systems [31].The numerical adaptive learning control scheme is researched for the discrete time nonlinear system, and the discrete time iteration ADP, which is based on finite approximation error,is proposed [32], [33].

In terms of specific modeling, Chenet al.[33] proposed a self-organizing neuro-fuzzy network based on first order effect sensitivity analysis; Gaoet al.[34]−[36] proposed neural adaptive control method for uncertain chaotic systems with multiple constraints, and neural adaptive chaos control with input and output saturation [37]; Lupuet al.[38] proposed manual control of unstable systems with time delays; Chenet al.[39] proposed robust control consensus of the nonlinear multi-agent system with switching topology and bounded noise;Gaoet al.[40]proposed fuzzy dynamic surface control for uncertain nonlinear systems under input saturation; Liuet al.[41] proposed generalized policy iteration ADP for discrete-time nonlinear systems.

In the aspect of application, for the gasification system and transform furnace system, the data-driven ADP optimal controller is designed, where the gasification systems models are built,the iterative control implementation of the evaluation network is presented for approximation to the system evaluation function, and the ADP optimal control method for multi controllers is proposed to obtain the optimal control strategy effectively [42]; ADP based building energy consumption optimization method is proposed to achieve the self-learning control of power’s matching and optimization [43]−[45]; Baiet al.[46]and Songet al.[47]researched parallel robotics and parallel unmanned systems,including its framework,structure,process, platform and applications; Zhuet al.[48] applied ACP methodology in the process control of NdFeB hydrogen explosion; Zhaoet al.[49] studied intelligent management of ventilation system.

D. Parallel Management

Based on ACP methodology and big data technology,deepening the study of application technologies like data modeling,optimization analysis,and CPSS based knowledge automation method and technology, parallel management can solve the engineering complexity and social complexity problem existing in enterprise management, social management, traffic management, economic management, and so on. The main research achievements include:

1) Knowledge Automation:To meet the scientific demand on the intelligent management of enterprises, CPSS based knowledge automation theory is put forward,including human centered knowledge acquisition, knowledge storage, knowledge transfer and knowledge update process.The concept,theory, framework and application of knowledge automation are put forward to enriching the contents of parallel management.The automation of the physical process and the knowledge automation of the virtual space are the keys of enterprises to develop knowledge automation. The software defined process and system are its core technologies. And the knowledge automation level of artificial systems mainly decides AFC(agility,focus and convergence)of the organization[50]−[52].How to acquire knowledge from internet’s big data, serve the people and realize dynamic closed-loop feedback and real-time interaction among big data,knowledge and people,and finally realize the knowledge automation, are studied. The precise perception, the cross-media data acquisition and knowledge extraction technology related to the specific raw materials and products, and the big data aggregation technology of the production fields, the agile interaction techniques and tools among human, computer and physical objects, and the knowledge automation service platform for production planning of process industries, are all studied [52]. And a pilot is applied in the daily operation of Shengli Refinery, SINOPEC Qilu Petrochemical Company.

2) Block Chaining Technique:Block chain technology is an important means and basic infrastructure to realize parallel society and knowledge automation [53]. It provides a set of distributed data structure, interaction mechanism and computing model.It is effective for the distributed social systems and distributed artificial intelligence research, and it lays a solid foundation of database and credit for the realization of parallel society. Naturally, ACP methodology can be combined with block chain technology to realize parallel social management.The block chains mechanisms such as P2P network,distributed consensus cooperation and the contribution based economic incentive,are the natural ways of modeling a distributed social system, in which each node will serve as an independent and autonomous agent in a distributed system.

3) Enterprise Management:In [54], the parallel enterprise management based on deep reinforcement learning is researched, where the overall models and framework of enterprise resource planning (ERP) construction based on multiagents is proposed. The sequential game model according to ERPs’ whole process and its optimal strategy are studied to solve the complex enterprises problems like uncertainty,diversity and complexity.

For the enterprises external big data and behavior modeling,the research focuses on cyber moment organization’s (CMOs)personal and organizational behavior modeling, the demands analysis of CMOs social network, the data acquisition, distributed data storage, the intelligence analysis technology of open source information, the computing and service platform for open source intelligence, and so on. The key technologies of ACP based enterprise marketing management under the network environment includes the budget optimization of keyword bidding advertisement business,strategy planning for keyword bidding advertisement business,participants behavior model and game equilibrium analysis for advertisement realtime bidding, etc. All those lay the foundation for the enterprise management to realize their economic goals,and provide the basis of parallel evolution between actual enterprise and its equivalent artificial enterprises.

For the enterprises internal behavior modeling and big data analysis, the modeling technologies of individual and group behavior are studied. The workflow performance analysis and simulation method of multidimensional enterprise workflow are studied. Support vector machines (SVMs) based classification and modeling method are studied.The hybrid modeling methods of data driven models plus mechanism based models are studied.

4) Social Manufacturing:To meet the fast and agile trends of the enterprises internal and external resource allocation,the theory and method of social manufacturing are put forward to promote the traditional mass production modes by comprehensive use of the big data resources inside and outside the enterprises. Focusing on model building, system platform design and application problems existing in social manufacturing mode, the artificial models are built with data mining technology, and the system platform adopts the latest network and information technology. The social manufacturing mode is applied for those personalized product fields,like household appliances,clothing,footwear.The social manufacturing mode and CPSS for clothing and shoes customization are researched.Using 3D scanning technology to collect physical data, the body accurate model can be built to provide data support for personalized customization.Analytic hierarchy process(AHP)based fuzzy comprehensive evaluation method is applied as the evaluation mechanism between designers and suppliers.Parallel system for garment customization is developed by using new techniques, like cloud computing and big data technology,which provide a powerful technical support for the personalized design and intelligent manufacturing for wearable products [55]−[58].

5) Logistics Management:Xueet al.introduced ACP methodology for the analysis of cluster supply chain,which is a new enterprise network consisting of industrial clusters and clusters supply chain. The cluster supply chain is a complex dynamic system. Xue’s team has researched such topics as the enterprises collaborative complexity,the network evolution modeling method,the influence factors and the mode evolution mechanism of the collaborative procurement, so as to support the construction of service system and the application practice of cluster supply chain mode [59]−[61]. Agent technology is introduced for the construction of CPPEM (collaborative procurement pattern evolution model) of cluster supply chain,including the CPPEM environment model,CPPEM static individual model, and CPPEM dynamic collaborative model, etc.Different computational experiments schemes are proposed for the CPPEM model,including the natural evolution scheme for the cooperative procurement model,the intervention evolution scheme of the cooperative procurement under different market conditions and different service charge strategies. At the same time, parallel logistics have also been applied in railwayhighway intermodal transportation.

Liuet al.[62] proposes a heuristic algorithm for container loading of pallets with infill boxes, which breaks through the core technology of parallel logistics management. The algorithm can maximize the volume usage of container under the following constraints: the bottoms of all kinds of trays are fully supported; the packing scheme satisfies the longitudinal constraint of complete segmentation, and helps forklift truck for loading and unloading; all product package and filling can be reduced into a complete product tray.

IV. APPLICATION PRACTICES

Based on the research achievements of ACP methodology and parallel control and management,the application practices of parallel transportation,parallel enterprises,parallel logistics,parallel agriculture and other fields are executed, and a series of innovative application results are obtained.

A. Parallel Transportation

Based on parallel system theory, a comprehensive parallel transportation system covers urban traffic, public transport,static traffic, logistics and social transportation, and so on.Qingdao,China is taken as a pilot to be committed to creating a parallel transportation model, to ease the traffic congestion,to guarantee traffic safety, and to enhance traffic experience.

In aspect of theory, the model of parallel transportation system is put forward, the control theory and method of agent-based traffic network are developed,and the construction method of data-driven artificial transportation system, which can effectively analyze the engineering and social complexity of traffic control, is put forward. A cloud-based unmanned hybrid intelligent solution, which can build an unmanned parallel system to predict and guide the actual unmanned system, is proposed. Based on the ADP method, a traffic signal controller with high flow and high strain capability and adaptive cruise controller are developed. Donget al.[63] and Xiong [64]et al.researched parallel transportation management and control system for bus rapid transit using the ACP approach; Xionget al.[65], [66] studied the application of parallel transportation system in the Asian Games; Konget al.[67] studied the application of ACP methodology in traffic flow prediction; Zhuet al.[68] studied the parallel public transport system and its practice of evacuation in largescale social activities; Konget al.[69] studied a special parallel transportation management and control system for urban transport systems;Chenet al.[70]researched the application of artificial transportation system to simulate the social impact of traffic behavior; Lvet al.[71], [72] studied traffic evacuation management based on computational experiments;Yanget al.[73] researched parallel parking system.

In aspect of practice, the parallel transportation control and management system for Qingdao city (PtMS-Qingdao), including the artificial transportation system for Qingdao (ATSQingdao), is constructed, and the optimization and evaluation of traffic control and management programs are studied with the help of PtMS-Qingdao. The intelligent traffic cloud architecture is designed and implemented based on the mobile agent and cloud computing technology, to achieve the agent-based network traffic control system. PtMS-Qingdao was awarded the Best Engineering Practice Award(2014)from IEEE SMCS and the Outstanding Application Award (2015) from IEEE ITSS.

B. Parallel Enterprises

In the aspect of enterprise management and intelligent management of complex industrial systems, the parallel system,which is composed of actual system and artificial system including modeling and representation,calculation experiment,analysis and evaluation,embedded simulation and closed-loop feedback system, is proposed. Based on ADP algorithm, the parallel dynamic programming and parallel learning method are proposed.The effects of human behaviors and massive data on enterprise management,especially enterprise resource allocation, are studied quantitatively. The enterprise management framework of “scenario-response” based on ACP methodology is studied. The construction of artificial organizations utilizing multi-agent technology and the impact on the enterprise resource allocation of corporate behaviors and massive data based on computational experiments are researched. The concept and method of novel enterprise resource planning(ERP 3.0) is proposed and applied to large-scale ethylene production enterprises [74], [75]. The results are applied to the Maoming Petrochemical Ethylene Cracking Plant and Qilu Shengli Refinery Plant, which can effectively improve the handling rate of the accidents, the production efficiency of the ethylene cracker, and the level of the operator and the management of staff, getting 310 million Yuan of new economic benefit and first prize of Science and Technology Progress Award from China Petroleum and Chemical Industry Application Association. The on-line optimization blending technology of gasoline is developed,to break the key technologies of precision control of oil performance indicators. The parallel system theory is introduced to nuclear power business operations,parallel nuclear power economic optimization technology, parallel failure warning technology, parallel nuclear emergency system,and parallel information security protection system are researched and developed [76]−[78]. Based on ADP, the parallel optimization of coal methanol temperature control system is achieved.Based on parallel management approach, the enterprise market analysis and optimization methods, which can effectively improve the keyword advertising bidding mechanism and advertising strategy, are put forward.The parallel management theory is applied to the management of tire business, and the tire intelligent management platform framework composed by public opinion+management+APP is put forward, and a tire industry competitive intelligence system is developed.

C. Parallel Logistics

The framework of parallel highway and railway intermodal system based on multi-agent modeling technology, including artificial highway and railway intermodal system [78], can be seen in Fig.5. The artificial intermodal transport system involves logistics centers, warehousing management, highway transport system, railway transport system, road distribution and so on. The involved objects include carriers, shippers,goods, trucks, trains and other equipment. During the actual logistics process,many heterogeneous multi-source data about cargo transport status, and urban traffic conditions can be quickly obtained by means of equipment or systems which can perceive, collect and transmit information.

Fig.5. The framework of parallel highway and railway intermodal system.

The calculation environment of the parallel highway and railway intermodal system is composed of cloud computing,which can deal with the rapid analysis and calculation of large-scale data. The establishment of the strategy library,knowledge base and database is used to store data, rules,and knowledge of the actual intermodal transport system, the artificial system, and the results of computational experiments and parallel execution.

The rolling optimization between the actual and the artificial intermodal transport system is established by parallel execution.According to the various schemes obtained from the results of computational experiments in artificial intermodal transport system, the actual optimization plan of the actual system can be optimized, including vehicle scheduling optimization, transportation management optimization, etc. Then,the optimization results are fed backed to the actual system.Then,the results in actual system can be transported artificially to optimize their own models and calculation module. Finally,the rolling optimization process can be achieved.

Parallel highway and railway intermodal system (parallel logistics system) has been constructed in a city in northeast China as a pilot, which can achieve transparency of personnel role and physical entity system through intelligent humancomputer interaction, to achieve transparent transport information, transportation scheduling optimization, customized customer transport services, predictive analysis of systematic operation, etc., so that the intermodal transport system can be efficient, low-cost. The system was awarded as the “2017 Annual Logistics Information Platform Excellence Case” by China Federation of Logistics and Purchasing (CFLP).

D. Others

1) Parallel Emergency Management:Aiming at the two core scientific problems of dynamic scenario generation evolution and calculation experiment, the parallel management is used as a guide to study the parallel response to unconventional emergency response and emergency management theory,method and technology, and the open, scalable, customizable,visual, dynamic experimental platforms of unconventional emergencies are designed [79]−[81]. The platform can integrate people’s knowledge of the refinement, differentiation and simplification of unconventional emergencies in the form of artificial societies. The models of artificial society include individual model, environmental entity model, traffic model,public opinion model, propagation dynamics model, cluster decision behavior model and so on.Through the computational experiments and dynamic monitoring data,a variety of scenarios can be produced to refine the new knowledge of the real society,emergencies and emergency management,and visually show the whole process of the overall situation of events and emergencies to reveal the inherent mechanism.These research results can provide all-round support for emergency management, and provide experimental tools and technical reference for theoretical research and on-line monitoring, early warning and auxiliary decision-making of unconventional emergency events, to transport emergency management research into an organized and systematic experimental study. All these can break the cognitive ability and cognitive capacity of the limitations of the control and management of unconventional emergencies for the decision-makers, to solve the lack of auxiliary means and decision-making time tension.

2) Parallel Agriculture:the construction and calibration methods of functional structure model of artificial agricultural system was put forward,and the general software“Green Garden” for crop growth and structure simulation was developed.According to the requirement of computational experiment,the plant growth modeling driven by data and knowledge was put forward,which was used for the prediction and analysis of crop growth under different environments. A parallel decision method for orchard fertilization based on growth prediction is proposed.

3) Parallel Network:the research framework of parallel network was proposed. It can leverage upon software-defined networking to construct artificial networks, and then effectively optimize the network system operations via the interactions between actual and artificial networks, to allocate the network resources more effectively, improve the management and utilization of resources [82].

4)Parallel Robotics and Parallel Unmanned Systems:Combining ACP-based parallel system with the field of robotics to form a combination of hardware and software framework for the UAV, unmanned vehicles, unmanned vehicles in complex environments, the framework to incorporate robotics and software-defined surrogates is proposed. Based on the basic concept of parallel robot, the basic framework of parallel unmanned system is put forward, and the basic functions and implementation methods of each module are introduced, and the key technology, which revolves around unmanned aerial vehicles, unmanned vehicles, unmanned vehicles is discussed.

Parallel simulation is an emerging simulation technology in the research area of system modeling and simulation.Geet al.[83]proposed the basic theoretical issues of equipment parallel simulation technology.Chenet al.[84]introduced the parallel simulation of complex evacuation scenarios with adaptive agent models. Geet al.[85] presented the theoretical framework for equipment parallel simulation, including concept,technology characteristics, constituent elements, comparison of relevant paradigms, technology classification, model evolution method and formalized description of running process for parallel simulation system. Maoet al.[86] introduced parallel simulation systems for command and decision support and the key supporting technologies. Geet al.[87] introduced equipment uniform resource locator(URL)prediction oriented parallel simulation technology. Douet al.[88] proposed the conception of the application of parallel simulation technology in command and control system. Geet al.[89] introduced the parallel simulation system and essential technology for equipment precision maintenance.

In addition,Wanget al.[90]studied the application of ACP methodology in the field of health care. Duanet al.[91]studied the application of the ACP methodology in public health crisis management taking H1N1 avian influenza as an example. Lyuet al.[92] studied the application of ACP methodology in the socio-economic system. Wenet al.[93]studied the evacuation strategy of the high-rise building fire based on ACP methodology.Huet al.[94]studied ACP-based research on evacuation strategies for high-rise building fire.Zhuet al.[95] studied the parallel traffic management and control system and its application in intelligent city.Denget al.[96] studied the concept, connotation and system framework of intelligent energy system based on ACP methodology and intelligent energy system. Wanget al.studied parallel intelligence system [97] and parallel intelligent command and control system [98]. Wanget al. studied parallel vision. Wanget al.also researched the generative adversarial networks(GAN) to realize the intelligent perception and understanding of complex environments by means of parallel execution[99]. Xionget al.[100] provided the parallel transportation management and control system for subways.Yanget al.[101]studied the parallel system method for satellite navigation experimental validation. Xionget al.[102] studied the cyberphysical-social system in intelligent transportation.

V. CONCLUSION

Complex system is one of the ten basic research directions in the outline of national medium and long term scientific and technological development program (2006−2020) of China,and its research is originated from many fields in national strategic development requirements. In this background, Parallel control and management method is proposed to improve the understanding and analysis ability of the dynamic evolution rules of complex systems and guarantee that complex systems can converge to the expected goals.

This paper summarized the basic theory, key technologies and application development of parallel control and management, looked forward to its future research and application directions, and put forward several development suggestions.Through the application practice in intelligent transportation,petrochemical production and many other fields, parallel control and management methods are proven to be effective to solve the modeling,optimization and control problem of complex systems which include both engineering complexities and social complexities. Using parallel control and management,we can effectively analyze the dynamic evolution of complex behavior,and thus obtain the effective management and control measures for realizing safe, reliable and efficient operation of complex systems. In this way, we can not only improve the perception, analysis and control capabilities of complex systems, but also promote the frontier researches on the basic theory of complex systems greatly.

Based on ACP methodology, the framework and theoretical foundation of parallel control and management method has been established, and the integrated modeling, experiment,analysis and decision support of complex systems have been realized. The future development trends are listed as follows:

1) The General Construction and Verification of Artificial Systems:It includes: the organical combination of traditional theoretical modeling, empirical modeling and data modeling methods; the general construction methods and steps of artificial systems; the verification methods of completion and reliability of artificial system models and the equivalence between artificial systems and real systems;the comprehensive methods of agent-based modeling, analysis and design of social factors such as personnel behavior; the universal methods of organic integration of engineering objects, social objects and environmental objects; the parallel computing based on CPU+GPU method to realize large-scale parallel computation of complex engineering system or complex social system.

2) The Design And Analysis of Computational Experiments Based on Artificial Systems:It includes: the program design and calibration methods and the sensitivity analysis and verification algorithms based on artificial systems; the target computing and event driven computing methods; the analysis and evolution methods of computational experimental results of different elements of complex system.

3) The Operating Mechanism and Key Technologies of Parallel Implementation:It includes: the interactive protocol and equivalent verification problems between the artificial and actual systems; control problems for the actual system and artificial system under the conditions of lack of information;the evaluation and verification methods of multi-objective and effective solutions of parallel control systems; the feedback mechanism, the corresponding control and adaptive algorithms,and the optimization strategy of feedback regulation in parallel control systems; the parallel optimization and control of parallel execution layers based on multi-agent control.

4) The Application Practices and Verifications of Parallel Control and Management:It includes: the universal technology architecture, functional architecture, implementation technology and software platform of parallel control and management of the actual complex system.With the increasing types of application practices in different areas, the total solution of parallel control and management will become more and more rich and mature.

ACKNOWLEDGEMENTS

We acknowledge the contribution of Fei-Yue Wang, Xiwei Liu, Fenghua Zhu and Xiuqin Shang to the paper.