220 likes | 338 Views
The use of Heuristics in the Design of GPS Networks. Peter Dare and Hussain Saleh School of Surveying University of East London Longbridge Road Dagenham, Essex, England Email: Peter@uel.ac.uk. Topics. Aim GPS Sessions and Schedule Problem description
E N D
The use of Heuristics in the Design of GPS Networks Peter Dare and Hussain Saleh School of Surveying University of East London Longbridge Road Dagenham, Essex, England Email: Peter@uel.ac.uk
Topics • Aim • GPS Sessions and Schedule • Problem description • Formulation as a Travelling Salesman Problem • Examples • Simulated Annealing • Recommendations and conclusions
Aim • To develop a method to determine the cheapest schedule given the sessions to be observed.
GPS Session • For a GPS session 2 or more receivers observe simultaneously. • For a network we have a number of sessions. • With 2 receivers, 6 sessions required for this network. • List of sessions is a schedule.
GPS Session • Sessions Required • A B • C B • C D • A D • A C • B D
Sessions and Schedule Schedule: ab-ac-dc
Problem Description • Given the list of sessions required, what is the optimum order of the sessions? • Need to define cost. • Cost can be defined, for example, by time of travel or shortest distance. • As optimum sought we aim to minimise the total cost incurred.
One receiver problem • Classic Travelling Salesman Problem (TSP) of Operational Research (OR). • Solved using Branch-and-Bound algorithm in Turbo Pascal to make use of pointers. • Limitations: Only one receiver; starts and ends at a point. • Developments: 2 or more receivers; start and end at non-survey point; allow for more than one observing day.
Example with one receiver Cost to move between B and C Cost Matrix: A B C D A 0 5 6 3 B 5 0 4 1 C 6 4 0 3 D 3 1 3 0 Least-cost Solution: A-D-B-C-A Cost: 14 units
Two Receiver Problem • For 2 receivers, cost is maximum of individual movements if time is criteria. • For example, cost of changing from session AC to BD is: • A to B: 5 units C to D: 3 units • Total cost: 5 units. • If distance is criteria, sum costs (e.g., total 8 units).
Two Receiver Problem • Need to allow reversal of sessions e.g., AC to DB. Cost is: • A to D: 3 units C to B: 4 units Total cost: 4 units. • However, now need to prevent receiver swaps. • For example, AC to CA. • Prevented by setting cost to infinity.
Two receiver problem: example • Four sessions: AB-BC-CD-DA
Solution to Two Receiver Problem • Modifications needed to standard TSP algorithm. • Solution (costing 9 units) is: • Rec. 1 A A D B A • Rec. 2 B D C C B • However, first and last sessions are duplicates! • Concept of base station needed.
Further Developments • To incorporate base, introduce dummy point. • To allow observations over more than one working day: • Extra dummy points. • Connect dummy points.
Example Survey - 1 • Cost matrix: 20*20 400 elements not shown here! • Observed schedule: • Rec. 1 Day 1: 2 2 1 Day 2: 2 3 5 6 6 • Rec. 2 Day 1: 3 4 4 Day 2: 1 4 4 5 3 • Total time: 180 minutes.
Example Survey - 2 • Optimal schedule: • Rec. 1 Day 1: 1 1 2 2 Day 2: 3 4 6 6 • Rec. 2 Day 1: 2 4 4 3 Day 2: 4 5 5 3 • Total time: 173 minutes. • But large cost matrix needed: 20*20. • To work with larger networks, approximate solutions (heuristics) needed.
Heuristics • Heuristics belong to the field of OR. • A Heuristic attempts to find near-optimal solutions in a reasonable amount of time. • The solution may be optimal but no guarantee. • Popular heuristics are: • Simulated annealing • Tabu search
Simulated Annealing (SA) • ‘Annealing’ - the cooling of material in a heat bath. • Solid material • Heated past melting point • Cooled back to a solid • Structure of new solid depends upon cooling rate
Application to Schedule Design - 1 No SA: • ‘Guess’ a schedule. • Change schedule to reduce cost. • Stop when no more improvements can be made. • Problem - local optimum often found - need global optimum.
Local and global optimum Cost Start Global optimum Local optimum Iterations
Application to Schedule Design - 2 With SA: • ‘Guess’ a schedule. • Change schedule to reduce cost. • Allow some ‘uphill’ moves climb out of local optimum. • Stop when no more improvements can be made global optimum (hopefully!)
Recommendations and conclusions • Optimal solution obtainable for small networks. Heuristics for large networks. • Further development of non-optimal solutions: • simulated annealing; tabu search; genetic algorithms. • Incorporate with other aspects of network design.