TiBT’06

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

APTS systems dispatching intelligence public transport

Andrzej ADAMSKI1

INTELLIGENT ANTICIPATIVE DISPATCHING SYSTEM OF PUBLIC TRANSPORT The public transport reliability and its quality and punctuality play a crucial role for the transportation system of a in-city communication. The work explains some factors of the solution; an APTS system technology, based on an advanced information technologies. The integrated real-time surveillance and intelligent anticipative dispatching control actions, co-responding with the traffic status recognition was elaborated and proposed. In this approach Model Predictive Control integrating predictions and schedule tracking control strategy allow allocation of the available transportation means, supported by an intelligent system that is evolved over the time.

INTELIGENTNY ANTYCYPACYJNY SYSTEM DYSPOZYTORSKI TRANSPORTU PUBLICZNEGO Niezawodność i poziom usług w komunikacji zbiorowej odgrywają kluczową rolę w aglomeracjach miejskich. W pracy bazując na obecnie dostępnych nowoczesnych technologiach APTS systemów zaproponowano rozwiązanie powyższego problemu w postaci nadzoru rozpoznającego incydenty ruchowe w czasie rzeczywistym zintegrowanego z inteligentnym antycypacyjnym sterowaniem dyspozytorskim reagującym na rozpoznane sytuacje ruchowe. W podejściu tym predykcyjne sterowanie umożliwia inteligentną dynamiczną alokację zasobów systemu komunikacji zbiorowej w zależności od realizowanego zadania

1.

Introduction The public transport reliable and high quality on-time service is a key operational

problem in urban areas. The operational system complexity, fast dynamics, functional non-homogeneity, high level of uncertainty and human behavioural anisotropy, occurrence of unpredictable operational incidents, unreliable operation of the driver-vehicle complex, complex interactions from other vehicles, passengers demand and system environment cause that reliable and high quality service offer is a challenging problem. In papers [1-3,5-7,10] the APTS hierarchical multi-layer management, surveillance and control system was proposed with properly integrated broad spectrum of decision-making and control functions (adaptive control, scheduling, management) (see Fig.1). The embedded into APTS control is concerned with the total integration of the control plant by control structure (horizontally from supply resources to passenger service, vertically from global system management to single vehicle control). The control

123

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

TiBT’06

problem in this generalized context is extremely difficult to handle in the traditional way. Therefore, the solution of this problem is split into the direct dispatching control layer realizing regulatory (schedule follow-up) control providing a stable steady-state operation of the lines, and optimization layer realizing optimizing (transit schedule set-point) control of the steady-state operated processes. Formally, the purpose of dispatching control is stabilization of bus trajectories (punctuality control) or headways (regularity control) around schedule trajectories/headways and consequent counteraction to off-schedule deviations. However, the provision of high level-of-service on this level, under high complexity of control plant 2-D processes circumstances, call for very efficient and dedicated control tools. This is visible on the real public transport lines where the normal operational related off-schedule deviations of buses if not properly and on time compensated by real-time dispatching control actions, are amplified by positive feedback mechanisms (e.g. bunching phenomenon), and propagate along the routes. In papers [2,4,6-10] flexible 1-D and 2-D dynamic network control models has been proposed and DISCON control method developed that offering wide spectrum of control capabilities, starting from control of interacting network elements with different levels of aggregation, in space (individual stops, route zones, common segments of different routes, overall routes), in vehicles population (one vehicle, groups of vehicles) or in time periods (rush hours, transient service periods during a day) and on multi-criteria and robustness features of the dispatching control actions ending. In this paper the new intelligent anticipative dispatching control idea is developed and dedicated mainly to cope with traffic incidents and system faults. The basic idea is to use predictive control after intelligent early recognition of such events i.e. a model to simulate the future effects of various anticipative control actions and the choose by on-line optimization the best one on the basis of the resulting service standards offered by various scenarios. The output measurements provide feedback being used to revise and improve the predictions. Since predictive control relies on an explicit internal model so we can dealing with traffic incidents by automatic updating this internal model. Artificial Intelligence (AI) based techniques are in natural way

1

Institute of Automatics, AGH University of Science and Technology, 30-059 Kraków, Al. Mickiewicza 30, tel. 0-12 6341568, [email protected],

124

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

TiBT’06

dedicated to the operation analysis of the transport system (i.e. system components, the infrastructure, the users and their complex dynamic interactions). The Intelligent Transport Systems (ITS) base on AI applications e.g. involving the integrated application of communications, management, surveillance, control and information processing technologies, use Ambient Intelligence (AmI) tools to support systems interactions intelligence in order to improve transport system operation. In this context it is essential to distinguish between information, which (if relevant) can advise the user, and intelligence, which implies knowledge of the user’s purpose, and an understanding of what information is relevant to his or her circumstances. The transport intelligence concern first of all to the self-recognizing spatial-temporal system state understanding intelligence (e.g. traffic incidents and faults) that influence the other system functional, connective, systems-cooperative, safety/security and logic intelligences. The intelligent control has more general meaning than the conventional control. The processes of interest are more general and may be described by hybrid control systems e.g. discrete event system like traffic incidents (DES) models or differential/ difference equation models or both. The intelligence at the upper public transport management IPTM layer

125

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

Fig.1.

TiBT’06

APTS: Advanced Public Transport management and control system

is required to support network-wide, pro-active traffic management, instead of the locally oriented and reactive traffic management which is observed today. Improved traffic management models are also necessary to deal with the huge amount of realtime traffic data to get from detectors and AVL+I/GPS systems (e.g. probe vehicles which are able to report their position and traffic conditions in real time) that need to be interpreted and analyzed to support the decision making process. The incorporation of AI into APTS systems for integrated management results from several reasons: limitations of current (locally-oriented) systems with facing operational and traffic incidents and congestion situations (requiring system-level precognitive expert behavioral support based integrated actions), due to presence of operators i.e. “man in the loop” in TMC centers with demand of on-line operator DSS tools to help cope with the complexity of integrated management schemes. 2.

Anticipative dispatching control The traffic incidents in dispatching control often cause undesirable passengers

reactions and shut-down of service properties of controlled plant. The incident-tolerant anticipatory dispatching control to be an aim of this paper is of paramount importance. Advanced traffic surveillance systems offer the capabilities of real-time on-line incident diagnosis, automatic incidents conditions and specifications assessment (e.g. location by AVL+I or OBU units and effective communication system) and selection of appropriate remedial actions. In this context the anticipatory dispatching control actions are able to avoid high costs and negative impacts consequences. In some cases the existence of “control by opportunity” situation enables on simple retuning of dispatching actions. In other cases, accommodation of the incident require the integrated actions on several lines. The main idea is to use LQ schedule tracking dispatching con126

TiBT’06

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

trol DISCON method in an interactive way by calculation of MPC constrained predictive control problems [11]. In DISCON, punctuality control model has the form [9 ]: x j +1 = A1j x j + B j u j + A nj x j − n + A1j z j

y j = C j x j + Djv j

(1)

s

xj=trajj-traj j state variables (i.e. deviations real trajectories from scheduled); uj control variables, zj /vj -disturbances (random travel and service times, driver behaviour) xj∈[xLj, xUj]; uj∈[Lj,Uj] ; aij = λk ∏l = k +1 (1 − λl ) ;Bj∈R

mxm

Aj1∈R x m

i

mxr

(2)

:bik=aik/λk;λkare eigenvalues of A1; Ajn∈R

m

is a zero matrix with last column: cim = (−1) i +1 ∏l =1 (1 − λl ) . Aggregated in space i

1

punctuality models are created from (1) with block matrices A =[Msl]s,l=1,.,q called intra/inter

zonal

i-k

Mll:aik=λl(1-λl)

/Msl: aik = λl (1 − λl ) Nl − k [ ∏ip = l + 1 (1 − λ p )](1 − λ s )i − N s−1 where +1,…,Nl.

1

i≥k; (s≥1);

i,k=Nl-1+1,.,Nl

i=Ns-1+1,.., Ns; k=Nl-

In addition to (1) the passenger load balance equation is used : lij=lij-1+λij(xij-

uij-xi-1j)-µijlij-1+∆λijHij where λij/∆λij- average/random passenger arrivals;µij-alighting proportions or in vector form: lj=Njlj-1+Ljhj+wj with l0=0 at terminals. The Fig. 2 illustrate the basic idea of anticipative predictive control. At the current time k the real bus trajectory is at yk whereas scheduled trajectory at sk. The reference trajectory r(t⎟k) determined by Intelligent Supervisor (Fig. 3) starts at yk and defines an ideal trajectory along which the plant should return to the schedule s(t). The rate of convergence r(t⎟k) to s(t) is an important “tuning knob”. Internal model is used to predict the behaviour of the plant y*(k+i⎟k), starting at the current time k, over future prediction horizon Hp. This predicted behaviour depends on input trajectory u*(k+i⎟k) i=0,1,..,Hp that is judged and selected on the basis of r(k+Hp⎟k)= y*(k+i⎟k) condition or a least squares solution with minimization of performance index of the differences [r(k+Hp⎟k)-y*(k+i⎟k)]2. The control law can be “tuned” by introducing weights into the cost function being minimized and adding term to penalise input energy [11].

127

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

Fig.2.

Fig.3.

TiBT’06

Basic idea of anticipative predictive dispatching control

Anticipative intelligent dispatching control structure

The Anticipative Dispatching Control ADC repeated cycle consist of: State estimation- Output prediction over prediction horizon- MPC predictive control optimization. For the state estimation and prediction purposes the punctuality model is extended on z and v related measured /unmeasured disturbances (modeled as outputs of corresponding LTI systems with Gaussian white noises). The estimates are computed from measured outputs (current and past) and past inputs by linear state observer and Kalman filtering. For a given reference trajectory and constraints over prediction horizon the MPC controller solves the optimization problem (weighted least square with control and output constraints) over selected control horizon M and implements the first control to the plant to be obligatory until new cycle starts at the next sampling instant. As an illustrative example the bus line no. 115 in Cracow with 28 bus stops was selected which are especially sensitive to unpredictable traffic incidents (see Table 1 for off-schedule deviations on this line during rush hours). The punctuality model (1) 128

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

TiBT’06

was aggregated in this case to three zones (q=3) [1-7],[7-21], [21-28]. These zones are controlled by DISCON-ADC control scheme initiated by Intelligent Supervisor. We assume the existence of AVL+I and APC systems at the Intelligent Supervisor level and for clarity we concentrate only on buses which are directly influenced by traffic incidents. We assume model=plant case with solutions related premise of exponential type of reference trajectories (with time constants Tref ∈[2-3]), the sampling interval Ts =0.3, prediction and control horizons Hp =8 and M=1 respectively. In Table 1 three typical zone control cases: Zone 1: the incident generated bunching phenomenon; Zone 2: demand peak; Zone 3: traffic jam are presented. In all cases the incident impacts are efficiently compensated by optimal DISCON-ADC constrained control actions. 3.

CONCLUSIONS The advantage proposed DISCON-ADC control scheme is capability to tune the

controller to achieve multiple schedule tracking objectives e.g. supporting priorities for realizing transfers. The explicit constraints on control and output variables enables to specify service quality imposed off-schedule tolerances and bounds for control actions. The ADC Supervisor-Control Cycle: State Estimation-Prediction-OptimizationMPC Control offers high efficiency and flexibility in control plant dedicated requirements representation. Model Predictive Control (MPC) automates a “plant” by combining a prediction and a schedule tracking control strategy. The control strategy compares predicted plant states to a set of objectives, then adjusts available dispatching control actions to achieve the objectives while respecting the plant’s constraints. Such constraints can include the control limits, boundaries of safe operation, and lower limits for service quality. MPC’s constraint-tolerance differentiates it from other “optimal control” strategies and allows to allocate the available plant resources intelligently as the system evolves over time. The future works will concern the robust features of the proposed control approach.

129

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

TiBT’06

BIBLIOGRAPHY

Adamski A, Optimal dispatching control of bus lines. 9th IFIP Conf. on Optimization Tech., Lecture Notes in Control and Information Science 23 334-344, 1980. [2] Adamski A., Optimal dispatching control of bus lines. Proc. of the 4th IFAC/IFIP/ IFORS Inter. Conference on Control in Transportation Systems, 6772, 1983.. [3] Adamski A., Optimal Dispatching Control in Public Transport. Habil. Thesis. Scientific Bulletins of AGH, AUTOMATICS no 50, 1989. [4] Adamski A., Probabilistic models of passengers service processes at bus stops. Transportation Research B, 26, pp. 253-259, 1992. [5] Adamski A., Optimal adaptive dispatching control in an integrated public transport management system. 2nd Meeting EURO WG on Transportation 913-38, 1993. [6] Adamski A., Distributed dispatching control in public transport. Proc. TRISTAN II Conference, vol. II, pp. 917-933, Capri- ITALY 23-28 June 1994. [7] Adamski A, Real-time computer-aided control in public transport from the point of schedule reliability. Lecture Notes in Economics and Math. 430, 23-38, 1995. [8] Adamski A., Flexible Dispatching Control Tools in Public Transport. Advanced Methods in Transportation Analysis eds. L Bianco P Toth Springer 481-506, 1996 [9] Adamski A., Turnau A., Simulation support tool for real-time dispatching control in public transport. Transportation Research A, vol. 32/2, pp. 73-87, 1998. [10] Adamski A, Multi-criteria robust real-time dispatching control method. 8th Meeting of the EURO WG on Transport. M. Bielli P. Carotenuto eds. 313-318, 2000. [11] Adamski A, ITS: control surveillance and management Monograph Cracow 2003 [12] Maciejowski J. Predictive control with constraints. Prentice Hall, 2002. [1]

130

V I  Konf er enc ja   Telema ty ka  i  B ez p i ecz eń s two   Tra ns portu   Katowice, 12 – 13 październik 2006 r.

Table.1.

Illustrative examples of zonal anticipative dispatching control on bus line no 115

131

TiBT’06