MODELLING REACTIVE AND PROACTIVE BEHAVIOUR IN SIMULATION MAZLINA ABDUL MAJID PROF UWE AICKELIN DR PEER-OLAF SIEBERS School of Computer Science University of Nottingham, UK mva@cs.nott.ac.uk 1
CONTENT Introduction Types of Simulation Motivation & Research Questions Research Approach Case Study Conceptual Model Model Implementation Experimentation Conclusion Future Work 2
INTRODUCTION Types of Simulation Motivation & Research Questions 3
Simulation Continuous : INTRODUCTION System Dynamic (SD): Modelling an aggregate level of a system using stock & flow or differential equations. Discrete : Discrete Event Simulation (DES): Modelling a sequence of events represented by entities; flowchart diagram. Agent Based Simulation (ABS): Modelling an individual approach behaviour represented by an agent ; statechart diagram. Why DES & ABS? Able to model an aggregate at low abstraction level e.g. individual behaviour. Able to model heterogeneous individuals who may behave differently Why exclude SD?? High suitability for modelling the aggregate behaviour of population. 4 We used combined DES/ABS ABS only is not represents our investigation which is more to process-oriented system.
INTRODUCTION Motivation Found randomness in selecting the simulation approach when modelling human behaviour Some researches used DES or ABS-selecting based on experience Knowledge about choosing the right simulation approach to used is important especially for the new simulation user. Found a disparity of work comparing SD and ABS or SD and DES contrasted to the about of work comparing DES and ABS. modelling human behavioural and where the focus is simulation output performance. Research Questions What kind of human behaviour can be modelled with standard model designs in DES and ABS? How similar are the results when one model the same human centric system with the two different approaches? 5
RESEARCH APPROACH Case Study Conceptual Model Model Implementation Experimentation : Output Performance Sensitivity analysis experiment Reactive vs. Proactive experiment 6
CASE STUDY Application Area Retail; Department Store( In UK top ten retailers ); Womenswear Fitting Room Problem Need to identify the potential impact for fitting room performance when having different numbers of fitting room cubicle and customers arrival rates. Human Behaviour investigation Reactive behaviour Staff responding to the customer when being available and requested. Proactive behaviour Staff take control /self started behaviour 7
CASE STUDY Proactive Behaviour Speed up the serving time Reactive Behaviour Reactive Behaviour Job 1 : counting no of clothes & give card Job 2 : providing help Job 3: receiving card & unwanted clothes An illustration for a fitting room operation at a department store 8
DEM CONCEPTUAL MODEL 9
DEM/ABM CONCEPTUAL MODEL 10
MODEL IMPLEMENTATION Simulation Software Anylogic TM 6.2 : Multi-paradigm software Similar programming language to develop DES and ABS models Model Description Consists of an arrival process (customer), three single queues (entry queue, return queue and help queue), resources(sales staff, fitting room) Run Length : One day ( 9.00 am -5.00 pm) Replications : 100 runs Both models used same inputs Model scenario : One staff does 3 reactive tasks ( counting garments on entry, fulfils request and counting garments on exit ) with oneproactive behaviour (faster service time). 11
MODEL IMPLEMENTATION Reactive Behaviour Setup One member does all reactive jobs following FIFO serving order. Proactive Behaviour Setup Condition: Fitting room cubicles are available or to get served by the staff Staff changes the service time from normal to fast (reduced 20%) Decision making -set of selection rules (using decision tree solutions) and probabilistic distribution. 12
Model Implementation :DEM Process presented in Flowchart State Information chart represents on inputs the and different outputs arrival of the rate simulation in a daymodels 13
Model Implementation :DES/ABS Simulation State Information chart animation represents inputs represents the and different the states outputs arrival changes of the rate simulation of in customers a daymodel 14
Model Implementation: DES/ABS Customer Agent Staff Agent Fitting Room Agent Communication through message passing 15
SENSITIVITY ANALYSIS EXPERIMENTATION Sensitivity Analysis Used sensitivity analysis for validation purpose Test based one two parameters varied the customer arrival rates (we increased it by 30% each time) varied the fitting room cubicles (8-10) Observed the effect on the system performance measures customer waiting times from three queues staff utilisation number of service time changes cubicle utilisation number customer not been served 16
SENSITIVITY ANALYSIS Part 1 : 8 Fitting Cubicles Customer waiting time 100 80 Staff utilisation 60 is much higher in DES 40 DEM 20 0 300 400 500 700 900 Number of customers DEM/ ABM Staff utilisation 100 80 60 40 20 0 DEM DEM/ABM 300 400 500 700 900 Number of customers Cabin utilisation 150 100 DEM 50 DEM/ ABM Number of service 0 time changes is much 300 400 500 700 900 higher in DES/ABS Number of customers Number of service time change 100 80 60 40 20 0 DEM DEM/ ABM 300 400 500 700 900 Num ber of customers Number customer not served 500 400 300 200 100 0 DEM DEM/ABM 300 400 500 700 900 Number of customers 17
SENSITIVITY ANALYSIS Part 2 : 10 Fitting Cubicles 100 80 Staff utilisation has 60 reduced in DES/ABS : 40 DEM frequently change 20 service time DEM/ ABM 0 300 400 500 700 900 Customer waiting time Num ber of customer Staff utilisation 100 80 60 40 20 0 300 400 500 700 900 Number of customer DEM DEM/ ABM Cabin utilisation 150 100 Number of service time changes drastically 50 DEM increased in DES/ABS DEM/ABM 0 300 400 500 700 900 Num ber of customers Number of service change 100 80 60 40 20 0 DEM DEM/ ABM 300 400 500 700 900 Number of customers Number customer not served has reduced in DES/ABS Number of customer not served 500 400 300 200 100 0 DEM DEM/ ABM 300 400 500 700 900 Num ber of customers 18
VALIDATION EXPERIMENTATION Sensitivity Analysis Conclusion proactive staff behaviour has an impact on both the performance measures of the DES model and the DES/ABS model the impact on the DES/ABS model is much stronger. surprisingly the addition of two fitting room cubicles did not drastically reduced the customer waiting time. we will investigate this issue further by adding another proactive behaviour the staff calling for help from another staff when condition met. 19
REACTIVE AND PROACTIVE EXPERIMENTATION Comparing the Impact of Reactive and Mixed Reactive and Proactive Behaviour Compare the previous reactive experiment work (Majidet al, 2009) with current mixed reactive and proactive experiment results (Majid et al,2010). Compare using statistical method Mann Whitney Purpose of comparison To establish similarities and differences between both models inoutput performance. Experiment Name: Reactive experiment = Experiment A (Majidet al, 2009) Mixed reactive and proactive experiment = ExperimentB (Majidet al, 2010) 20
REACTIVE AND PROACTIVE EXPERIMENTATION Similar model scenario for both simulation models One member of staff that does three jobs ((1) counting garments on entry, (2) providing help, and (3) counting garments on exit) A fixed number of customers arriving per day (300) A fixed number of fitting room cubicles (8) The proactive feature is the demand driven change in service times. Customer waiting time and staff utilisation as performance measures for our statistical comparison. 21
REACTIVE AND PROACTIVE EXPERIMENTATION The hypotheses : HoA = The average customer waiting times resulting from our DES model are not significantly different in Experiment A and B. HoB = The average customer waiting times resulting from our DES/ABS model are not significantly different in Experiment A and B. HoC = The staff utilisation values resulting from our DES modelare not significantly different in Experiment A and B. HoD = The staff utilisation values resulting from our DES/ABS model are not significantlydifferent in Experiment A and B. 22
REACTIVE AND PROACTIVE EXPERIMENTATION The Results : Performance Measures DES (P-value) Experiment A vs. B DES/ABS (P-Value) Waiting time Staff utilisation 0.6761 >0.05 =accept HoA 0.3019 >0.05 = accept HoC 0.015 <0.05 =reject HoB 0.000 <0.05 =reject HoD The above results supported our findings in Experiment B DES/ABS produced higher number of service time changes compared to DES which then has produced abigger impact on customer waiting times and staff utilisation. 23
CONCLUSION AND FUTURE WORK Findings the advantages and disadvantages of implementing the reactive and proactive service behaviour in DES and combined DES/ABS with regards to output performance. Previous paper (Majidet al, 2009) dealt with reactive behaviour. Current paper (Majid et al 2010) investigates on mixed reactive and proactive behaviour. Conducted sensitivity analysis model validation experiment. DES/ABS shows much stronger impact on the performance measures. 24
CONCLUSION AND FUTURE WORK Compared the reactive behaviour with mixed reactive and proactive behaviour No significant difference found in DES but significant difference in DES/ABS. Overall we found differences in output performance when modelling proactive behaviour in two difference approaches( DES vs. DES/ABS) Overall conclusion: The magnitude of impact to model proactive behaviour depends very much on the chosen simulation approach. This knowledge is an important result to support justification for related research in this area. Future work : add other forms of proactive behaviour and case study - to generalise our findings. 25
THANK YOU. Question? 26