|Sažetak (engleski)|| |
Mass public transport is the backbone of sustainable urban mobility, and it is the only transport mode with high capacity to meet transport demand in cities. Traffic system of today is characterized by an increasing private car usage because of economic prosperity for citizens and unattractive transport alternatives. In the current modal split of city trips, the increasing private car usage is putting pressure on the existing urban traffic system, and by space consumption, traffic congestion, noise pollution, air pollution and poor road safety, unnecessary external costs are generated. Public transport systems in cities worldwide experience traffic congestion and overcrowded public transport vehicles. In timetable design, there are three traditional principles: meeting transport demand, reducing stop waiting time, and avoiding vehicle crowding. Quality of service is the successfulness of public transport operations considering passengers, operator, and the local community. In the literature, service reliability is one of categories belonging to quality of service, and service reliability is defined by comparing real service to the timetables. Service reliability is traditionally demonstrated by punctuality (for each measuring point in the network, difference between the observed arrival time and timetable arrival time) and regularity (for each measuring point in the network, difference between the observed interval and timetable interval). Both punctuality and regularity are crucial for service reliability, quality of service, attractiveness of mass transit and modal split of city trips. Service reliability in mass transit can be observed in a single or multiple parts of passenger trips, and each uses time to complete – these times are walking time (from source to network, for transfers, from network to destination), stop waiting time, and time spent in vehicle (consisting of running time between stops and stop dwell time on passenger routes). Due to technical capabilities in the past, service reliability research was primarily focused on stop waiting time. But with the development of automatic vehicle location and automated passenger counting technologies, a more detailed insight was provided into time spent in vehicles. Most in-vehicle time research of today is focused on predicting stop arrival time, to provide reliable vehicle arrival times for passengers on timetables, stop information displays or smart devices. Mass public transport is characterized by high transport supply and demand, and passengers arrive to stops randomly, without considering timetables. However, in public transport conducted by trams, trolleybuses and buses, vehicle performance is limited by the infrastructure. Completely segregated corridors are less common, and transport is often limited by the external disturbance factors (not manageable by the operator) and internal disturbance factors (manageable by the operator). The external ones are transport demand, signalized intersections, other vehicles, pedestrians, passenger behaviour and weather conditions, and the internal ones are vehicle bunching, driver behaviour, timetable quality, vehicle availability and driver availability. Because of these factors, travel time (between any two points on the line or between the terminals at maximum) is characterized by variability. Travel time variability reduces service reliability for passengers, such that their waiting time at stops and time spent in vehicles becomes more unpredictable, and they must include additional time in their trips to prevent late arrivals to their destinations. The research on travel time perception by passengers revealed that the reduction of travel time variability decreases uncertainties when estimating destination arrival time and decreases uncertainties in trip planning. Passengers also prefer reduction of variability over the reduction of average travel time itself because they can estimate arrival to the destination more precisely. Reducing travel time variability also makes transport supply more evenly distributed, reducing the number of uncomfortable rides. For the operator, less travel time variability means fewer operating costs. Travel time variability can be described by a distribution or descriptive statistics. Research has shown that distributions such as gamma, normal, log-normal, and log-logistic are most common. Normal distribution is better in commuter periods, and log-normal in non-commuter periods. The most common descriptive measures are two absolute ones (in minutes) – difference between 90th and 10th percentiles and standard deviation, and two relative ones (in percent) – ratio of 90th minus 10th percentile and median, and the coefficient of variation. Absolute measures are better for estimating time losses for passengers, and relative ones are better for estimating network performance. Research has shown that median is a better measure than the average, and that both measures increase linearly with the average travel time. Travel time is predicted by three types of models: primitive, based on data and based on traffic flow theory. The existence of many prognostic models indicates that each has advantages and disadvantages in terms of complexity, amount of input data, traffic theory, and applicability. Primitive models, consisting of instant, historical and hybrid, are easy to implement, but their disadvantage is the assumption that variables affecting travel time are constant in time. In models based on data, relationships between the dependent and the independent variables is established, so these models do not require extensive traffic theory knowledge, but they require a large amount of input data. They consist of parametric and non-parametric models. In parametric modes, two techniques are used – in time series techniques, which cannot show variations in real time, autoregressive methods are used, and in regression techniques, which may have problems with interdependence of independent variables, ridge regression, regression tree, bagging regression, random forest regression and support vector regression is often used. In non-parametric models, the relationships between variables are obtained directly from the data using machine learning – artificial neural network or support vector machine are most common. In models based on traffic flow theory, travel time is predicted using theoretical models. Unlike data-based models, these models do not require a large amount of data, but they may be inaccurate due to network specifications. They consist of macroscopic (by particle filters and Kalman filters) and microscopic (by source-destination matrices). The goal of this research is to develop a travel time prediction model for mass transit by establishing relationships between travel time and disturbance factors. The purpose of research is to improve travel time prediction in mass transit, ensuring a sustainable transport system with less external costs by shifting passengers from private cars to public transport, and reducing impact on the environment, energy consumption and space consumption. There are two research hypotheses: • hypothesis 1 – “It is possible to develop a travel time prediction model in urban mass transit based on disturbance factors” • hypothesis 2 – “It is possible to establish relationships between travel time and transport supply irregularity”. The research was conducted in five phases: • in phase 1, travel time data were collected by observing tram traffic in the city of Zagreb, by observing vehicles in specific locations on tram line 4; besides travel time data, all other data regarding distances, stops, intersections, pedestrian crossings, lane type and traffic volume were also collected • in phase 2, the collected data were processed, and the relevant measures of travel time and disturbance factors were calculated; based on the geometry, legislation and vehicle characteristics, ideal travel time was calculated • in phase 3, relationships between travel time and disturbance factors were established by correlation matrices; and insignificant variables were eliminated • in phase 4, based on correlation results, travel time prediction models using multiple linear regression were developed • in phase 5, the models were validated on a different dataset using scientific methods • in phase 6, the models were tested between the terminals, to check applicability for long segments. There were three models established by calculating three dependent variables, because all of them are important for service reliability: • deviation of median from ideal travel time – to predict travel time for average vehicle • deviation of median from the 10th percentile – to predict travel time for early vehicles • deviation of 90th percentile from ideal travel time – to predict travel time for late vehicles. After the correlation and model development, only three independent variables (predictors) remained: • traffic lights – total time lost at intersections divided by ideal travel time; total time lost at intersections is the sum of proportions of red-light times squared divided by double intersection cycle time for every intersection on the observed segment • ideal travel time – the reciprocal of ideal travel time on the observed segment was used • traffic volume – the ratio of total ideal time on sections with intense other traffic and segment ideal travel time. The model passed linear regression tests. The model was also validated by using different sample of travel time data, and the validation showed an average 9% error for 10th and 90th percentile of travel time, which was then used to estimate minimum and maximum to be used in timetables. The results showed that, by applying minimum and maximum from the validation, 10% of total vehicles will have travel time less than minimum and 14% of total vehicles will have greater travel time than maximum, which was acceptable. Therefore, the hypothesis 1 of the research was approved. Relationships between travel time and transport supply irregularity was subjected to correlation analysis as well, and several measures of supply irregularity based on vehicle interval and frequency were chosen as predictors. There were some minor correlations; however, in the multiple linear regression, all variables describing transport supply irregularity did not improve previously constructed models, and therefore the hypothesis 2 of the research was rejected. Scientific contribution of the research is achieved through: • determining disturbance factors affecting travel time in mass transit, such that the factors are richer in traffic context than in previous research • determining impact of each disturbance factor on travel time in mass transit, by introducing ideal travel time used for comparison • developing a new travel time prediction model in mass transit based on tram traffic rarely conducted in the past. The limitations of this research were manual data collection, and the assumption that timetable frequency has significant influence on travel time, resulting in analysing travel time for short segments. Therefore, future research should predict travel time on longer segments. Additional limitation of the research is the assumption of ideal travel time based on observations. Therefore, driver behaviour, and vehicle characteristics may be considered to improve the accuracy of the model. Since the data collection could not provide stop time data, future research should include stop time data, for modelling riding time and stop time separately.