Application of Multi agent Technology in the Scheduling System of Swarm of Earth Remote Sensing Satellites Procedia Computer Science 103 ( 2017 ) 396 – 402 1877 0509 © 2017 The Authors Published by El[.]
Available online at www.sciencedirect.com ScienceDirect Procedia Computer Science 103 (2017) 396 – 402 XIIth International Symposium «Intelligent Systems», INTELS’16, 5-7 October 2016, Moscow, Russia Application of multi-agent technology in the scheduling system of swarm of Earth remote sensing satellites P.O Skobeleva,b, E.V Simonova c, A.A Zhilyaevb,*, V.S Travina b a SEC Smart Solutions Ltd, Samara, Russia Department of Aircraft Designing, Samara University, 34, Moskovskoye Shosse, Samara 443086, Russia c Department of Informatics and Information Technology, Samara University, Samara, Russia Abstract The paper studies the problem of scheduling a group of Earth remote sensing satellites The following idea is proved: development of Earth remote sensing systems needs changing of approach to planning of their application Problem statement is described As criteria of efficiency, information delivery time, resolution and cost of request execution are used The sched ule has to comply with the following constraints: visibility between satellites, observation areas and data receiving points, storage capacity of the memory unit as well as coordination of operations on shooting, storing, transmitting and receiving data Review of ways of problem solution is provided Implementation of the approach has been suggested, where the sought schedule is built as dynamic balancing of interests of satellites, data receiving points and observation area agents Multi-agent planning system is developed Architecture of the system is described as well as functions of the modules it includes Dynamically occurring events are taken into account when planning, including introduction of a new task or change of task options, failure of a satellite or means of communication The experimental assessment of time spent on recovery of the damaged schedule is given In conclusion benefits of the multi-agent approach at management of swarm of Earth remote sensing satellites are provided © byby Elsevier B.V This is an open access article under the CC BY-NC-ND license © 2017 2017The TheAuthors Authors.Published Published Elsevier B.V (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the XIIth International Symposium «Intelligent Systems» Peer-review under responsibility of the scientific committee of the XIIth International Symposium “Intelligent Systems” Keywords: multi-agent technology; swarm of satellites; multisatellite; ground station; remote sensing; planning and scheduling; adaptability; multiobjective optimization * Corresponding author E-mail address: zhilyaev@smartsolutions-123.ru 1877-0509 © 2017 The Authors Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the XIIth International Symposium “Intelligent Systems” doi:10.1016/j.procs.2017.01.127 P.O Skobelev et al / Procedia Computer Science 103 (2017) 396 – 402 Introduction Modern space remote sensing tools represent complex technical systems distinguished by a great diversity of types of target equipment, its modes and application conditions as well as structure and functionality of ground infrastructure Due to increase in the number of consumers of space information, now sensing systems are being created which consist of a swarm of heterogeneous satellites and a wide network of geographically distributed data receiving and processing stations Managing such systems is closely connected with efficiency of their target application, planning of which has to take into account technical resources of satellites, external conditions, interests of operating organizations and consumers of target information as well as economic feasibility The scheduling process of a swarm of Earth remote sensing satellites is affected by the following factors: Growth of a swarm dimension and the need to switch to solution of a new type of tasks on complex management of multi-satellite space systems Change of requirements to volume, quality and efficiency of acquisition of remote sensing information, specified by consumers Updating of satellites’ action plan becomes necessary as far as new observation requests and unforeseen events occur in real time Multicriteriality Efficiency of target application of space sensing system can be assessed by various criteria, such as productivity, information capacity and efficiency of information delivery Nevertheless, most of developments in planning of Earth remote sensing are oriented to single satellite functioning in determined conditions1 As a final result, a static plan of using satellite is considered, which is updated no more than several times per 24 hours At that, no mechanisms of plan recovery are presupposed when actual conditions deviate from the expected ones in the moment of plan building In the result, the problem of finding rational methods providing acceptable locally optimal planning of actions of a swarm of satellites with an opportunity of adaptive plan changing depending on various events is becoming quite relevant Multi-satellite scheduling problem 2.1 Problem Statement Let there be a lot of satellites, stations and observation requests assigned Time of requests occurring is unpredictable, the number and characteristics of resources can also change in the process Request processing is presented as a graph where nodes are separate operations (shooting, storing, transmitting and receiving information), and bonds are relations determining the sequence of operations For each satellite and ground station there is a list of operations, cost and conditions of their execution consistency For example, each satellite is able to perform operations on shooting and data transmission without possibility of their intersection in time Sequence of one-type operations is united in a mode, in the beginning of which preparatory section is arranged There should be no inclusions of other modes between operations of one mode The duration of pause between neighboring operations should not exceed the assigned value For each satellite orbit parameters are assigned as well as technical constraints for memory capacity, duration of modes and operations For stations geographic coordinates are assigned There are models formed which allow for determining: x time when a satellite passes over visibility zones with ground stations and areas; x resolution of image in each area caught in the observation of a satellite; x volume of information recorded in the on-board memory unit while shooting; x duration of operation execution on image acquisition and information transmission Each request has the corresponding feature, for example, information delivery time, resolution in the assigned spectrum range, and cost of execution For feature a range of acceptable values can be specified as well as optimal value On the basis of separate features, components of efficiency of request distribution are formed, which are outlined in Figure Efficiency of request distribution is determined via linear convolution of all components with the assigned weighting coefficients 397 398 P.O Skobelev et al / Procedia Computer Science 103 (2017) 396 – 402 Figure General outline of efficiency component One has to form a plan of request processing by a swarm of satellites, made according to maximization criteria of distribution efficiency of all requests The acquired schedule has to comply with the following constraints: x Visibility between satellite and observation area while shooting x Visibility between satellite and ground station while transmitting information x Enough storage available in the on-board memory unit of a satellite x Coordination between time of shooting, information transmission and reception x The assigned requirements on impossibility of merging of modes of different equipment types An essential peculiarity of the problem is the need of accounting dynamically occurring events, which include occurring of a new event, appearance, failure or change of parameters of a satellite and ground tools of information reception and processing 2.2 State of the Art Methods of scheduling actions of Earth remote sensing satellites are developing in the following directions: x Methods of traditional optimization and linear programming, improving exact methods of problem solution, such as branch and bound, dynamic programming algorithm, constraint programming2; x Greedy algorithms, local search algorithms based on heuristic rules of the domain3; x Artificial intelligence methods, use of neural networks and fuzzy logic4; x Metaheuristics: genetic algorithms, tabu search; x Imitation and bio-inspired methods: ant colony optimization 5, swarm optimization, simulated annealing, squeaky wheel optimization, iterated sampling algorithms6; x Methods of distributed problem solving of resource planning using multi-agent technology7 The literature shows that the principle of adaptive satellite resource scheduling with the use of heuristic methods is very efficient The paper considers the possibility of implementing this principle with the help of multi-agent technology, which showed good results when solving traditional tasks of resource planning and allocation Problem is suggested to be solved via creation of autonomous smart agents; each of them has its own target and constraints Agents can interact with each other via negotiations which provide coordination of solutions for all participants (agents) and ensure KPI maximization for the whole system9 At the same time, there appears possibility of adaptive change of the previously made plan (without its complete rearrangement) Multi-agent system description 3.1 System architecture To solve the stated problem, multi-agent system for scheduling of actions of a swarm of Earth remote sensing satellites has been developed Architecture of the system is presented in Figure as a total of modules and links between them Performance of the system is described as interaction of the following modules: x control module – provides routing and transformation of data flow; x data store – object-oriented information data base; P.O Skobelev et al / Procedia Computer Science 103 (2017) 396 – 402 x planning module – provides formation of a plan of equipment functioning of a satellite and a network of ground stations; provides the environment for agents’ performance; contains samples of all agents of the system; defines the order and algorithm of their functioning; x ballistics module – forms intervals of satellites’ passing through visibility zones with observation areas and ground stations; x visualization module – allows for presenting structure and parameters of a space system, modeling the process of execution of Earth remote sensing task, provides interaction of user and system and formation of reporting documents according to planning results Figure System architecture 3.2 Planning mechanism To implement the multi-agent approach a concept of demand and resource network is used, where any plan is built as a flexible (configurable depending on events) network of relations between request agents (demands) and resource agents10 In the problem considered, resources are agents of satellites and ground stations, requests are agents of observation area shooting requests Request agent wants to be executed with the best parameters of efficiency, resolution and cost, regarding the assigned weighting coefficients The target of satellite and ground station agents is increase of their profit in the considered time horizon Comparison of target function values of different agents is made according to interests of the whole system, which are presented by weighted sum of target functions of separate agents The scheduling task is being solved by way of iterations, due to improvement of satisfaction function values of separate agents, which result in growth of additive function of the system11 When conflicts occur, a compensation mechanism is used: augmentation of satisfaction function of one agent can be spent on compensation of expenses of others At that, the basic principle consists in the following: if one agent asks another to worsen its state, the former has to compensate all the losses This allows for balancing of agents’ opposite interests12 In other words, each agent tries to achieve its goal in a completely “selfish” way, but when its interests clash with those of others, it solves the conflicts to the benefit of the whole system At the first stage, agents use greedy strategy of search for allocation variants and choose only vacant places without conflicts Thus, a locally optimal solution is being built which is further corrected adaptively in a sliding mode in the considered time horizon At the next stage, shooting request agents try to improve the value of their target function and offer the conflicting request to find other allocation intervals via time shift or transition to another resource Resource agents try to increase target function values by redistribution of the operations executed At that, a satellite or a station can offer requests to execute operations connected with them for lower cost by grouping them in one mode and distributing expenses on preparing of shooting or information transmission Iterative improvement of the plan is made by all types of agents until the state of “dynamic stop” occurs, that is agents try to improve their state but improvement of target function value does not happen It will mean that a 399 400 P.O Skobelev et al / Procedia Computer Science 103 (2017) 396 – 402 consensus is achieved in the result of negotiations and there is a possibility of providing a ready solution In this case, schedule is brought to acceptable form and then improved by the mechanism stated above 3.3 Scheduling visualization Information about scheduling process and results is displayed on several screens; each of them represents a certain aspect of the system functioning Screen “Physical world” in Earth 3D model contains observation areas and ground stations There are orbits and positions of satellites around the globe which correspond to the current model time (Figure 3) Screen “Grouping plan” represents a plan of actions of a selected satellite Here visibility zones of satellites are shown with all areas and ground stations as well as a plan of executing operations on shooting and data transmission Screen “Efficiency indicators” contains diagrams demonstrating how the quality of schedule is changing in the process of scheduling Here a graph of changing of the system target function is located as well as diagrams reflecting the change of a number of images made by each satellite and received by each station (Figure 4) Figure Fragment of the screen “Physical world” Figure Fragment of the screen “Efficiency indicators” For input of events terminals are used which allow for changing structure and parameters of a swarm of satellites and a network of ground stations as well as for editing requests for observation area shooting Simulation results The main advantage of the planning system is the ability to parry external events that change the conditions of the problem solved The experiment is conducted to evaluate the time spent on restoration of a damaged schedule due to removal of one of the satellites A swarm of three satellites, seven stations and 400 observation areas is used Later, one of the most loaded satellites is removed Figure shows the change in the value of the system target function in the process of building a schedule The event which happened on the 57th second (removal of one of the satellites) has led to a decrease in the value of the target function by 0.2 However, over the next 30 seconds, the planning P.O Skobelev et al / Procedia Computer Science 103 (2017) 396 – 402 system is able to parry this event and rebuild the schedule, increasing the value of the target function of the system by 0.14 Figure Change in the target function value when an event occurs Conclusion In this paper, we present a new approach to solving the Earth sensing satellites scheduling problem using multiagent technology Experimental results show that our proposed approach is robust and able to find local-optimal solution within reasonable computational time The key advantage is the ability to dynamically change the parameters of the problem, followed by an adaptation of the current schedule to the changes Using the multi-agent technology gives the possibility to: x increase the efficiency of request fulfillment by dynamically adjusting of the schedules "on the fly" with the occurrence of unforeseen events; x take into account the individual characteristics of consumers by providing a flexible mechanism for setting constraints and criteria, depending on the situation; x improve information value of services through integration of heterogeneous satellites, each of which is equipped with a certain type of equipment; x improve the scalability and openness of the space-based systems through dynamically connecting new components (satellites, ground stations) without stopping and restarting the whole system The developed system is most effective when working in real time, when the quality and efficiency of decisions depends on the very moment of decision-making Here introduce the paper, and put a nomenclature if necessary, in a box with the same font size as the rest of the paper The paragraphs continue from here and are only separated by headings, subheadings, images and formulae The section headings are arranged by numbers, bold and 10 pt Here follows further instructions for authors Acknowledgment This work was carried out in SEC "Smart Solutions" Ltd with the financial support of the Ministry of Education and Science of the Russian Federation References Sollogub A, Anshakov G and Danilov V 2009 Spacecraft systems for sensing of the Earth's surface (Moscow: Mechanical Engineering) 401 402 P.O Skobelev et al / Procedia Computer Science 103 (2017) 396 – 402 Lemtre M, Verfaillie G, Jouhaud F, Lachiver J M and Bataille N 2002 Selecting and scheduling observations of agile satellites Aerospace Science and Technology pp 367–381 Tangpattanakul P, Jozefowiez N and Lopez P 2015 A multi-objective local search heuristic for scheduling Earth observations taken by an agile satellite European Journal of Operational Research 245 pp 542–554 Liu S, Bai G and Chen Y 2015 Prediction method for imaging task schedulability of earth observation network Journal of Astronautics 36 pp 583–588 Iacopino C, Palmer P, Policella N, Donati A and Brewer A 2014 How ants can manage your satellites Acta Futura pp 57–70 Globus A, Crawford J, Lohn J and Pryor A 2004 Application of techniques for scheduling earth-observing satellites Proc of the 16th Conf on Innovative Applications of Artificial Intelligence pp 836–843 Bonnet J, Gleizes M P, Kaddoum E, Rainjonneau S and Flandin G 2015 Multi-satellite Mission Planning Using a Self-Adaptive Multi-agent System IEEE 9th International Conf on Self-Adaptive and Self-Organizing Systems pp 11–20 Skobelev P 2011 Multi-Agent Systems for Real Time Resource Allocation, Scheduling, Optimization and Controlling: Industrial Application 10-th International Conf on Industrial Applications of Holonic and Multi-Agent Systems pp 1–14 Wooldridge M 2009 An Introduction to Multiagent Systems (London: John Wiley&Sons) 10 Rzevski G and Skobelev P 2014 Managing complexity (London-Boston: WIT Press) 11 Belokonov I, Skobelev P, Simonova E, Travin V and Zhilyaev A 2015 Multiagent planning of the network traffic between nanosatellites and ground stations Procedia Engineering: Scientific and Technological Experiments on Automatic Space Vehicles and Small Satellites pp 118–130 12 Skobelev P, Simonova E, Zhilyaev A and Travin V 2015 Multi-Agent Planning of Spacecraft Group for Earth Remote Sensing Proc of the 5th International Workshop on Service Orientation in Holonic and Multi-Agent Manufacturing (SOHOMA15) 640 pp 309–317 ... processing 2.2 State of the Art Methods of scheduling actions of Earth remote sensing satellites are developing in the following directions: x Methods of traditional optimization and linear programming,... architecture To solve the stated problem, multi-agent system for scheduling of actions of a swarm of Earth remote sensing satellites has been developed Architecture of the system is presented in Figure as... shooting and data transmission Screen “Efficiency indicators” contains diagrams demonstrating how the quality of schedule is changing in the process of scheduling Here a graph of changing of the system