PROPIEDADES GENERALES DE LOS ACEROS INOXIDABLES
4. SISTEMA ELECTRÓNICO PARA CONTROL DE DISPOSITIVO PORTA SUERO
4.4 DISEÑO ELÉCTRICO – ELECTRÓNICO
4.4.5 Funcionamiento de dispositivo porta-suero La idea general del sistema completo del porta-suero se muestra en la figura 36 y se explica a continuación: el
The coverage of the network geographically and in time depends upon the way the dataset has been collected. In the UK the major roads have a lot more data available as there is a lot more demand and the travel time is likely to change over time. A small country lane may only have 10 vehicles traverse it each day and most have local knowledge from past use. Collecting detailed information on the very large number of smaller roads is impractical. However if a logistics company needs to deliver along the road some estimate is required. The practitioner may therefore need more than one dataset to cover the entire network. If datasets overlap, the one with the highest frequency of collection and longest history should be chosen.
CHAPTER 2. EVALUATING SINGLE LINK MODELS 66
specific. Some datasets are publicly available, whereas others must be purchased. Since our focus is on scheduling the transportation petrol for the next day we require data that gives an accurate picture of how the travel time varies throughout the day. The majority of datasets with the level of detail required to predict this level of variation come from government sources. They maintain and monitor the roads and it would be very difficult for anyone else to install and maintain a network of automatic numberplate recognition (ANPR) cameras across an entire network.
We exclude less detailed datasets in this analysis which restricts the possible links in the vehicle routing problem network. The removed links are the minor roads and when transporting hazardous materials staying on major roads for as long as possible is preferable as they are better suited to large vehicles. Due to the lack of data for these links, the majority of prediction methods detailed in Section 2.2 are either inappropriate or give predictions on a very large scale.
A report on a variety of considerations for collecting travel time data is provided by Turner et al. (1998). An overview of the main collection techniques, with their characteristics are provided.
A summary of the characteristics for each method presented in Table 2.3.1. This links back to Section 2.3.1 as it summarises the frequency and the even intervals. Of the other three properties, missing data occurs in all of the collection methods and the time period and dataset length depend upon the dataset not the collection method. The other two characteristics in the table depend upon the method of collection, rather than the dataset. Pre-processing is the level of processing on the data that is required before it is in the format to use in the models and collection is who collects and stores the data.
ANPR cameras. One common method is using two ANPR at the start and end of a section of road. The travel time of a vehicle is calculated by taking the time between the two photos. Unfortunately not all vehicles will be captured by both
CHAPTER 2. EVALUATING SINGLE LINK MODELS 67
Pre processing Even intervals Frequency Collection
ANPR camera L Y H Road managment
Government
Fleet collection L N L Companies
Floating car H N but Av L/H Various
Mobile phones H N but Av H Phone network
Table 2.3.1: Summary of data collection characteristics. L and H are low and high, Y and N are yes and no and Av is average. The characteristics are the level of pre- processing required, the frequency of collection, who collects the data and if the data is collected at even intervals.
cameras and hence if the information is taken too regularly there may be times with no data. ANPR cameras are usually on all the time, and due to the large volume of cars the travel time calculations are recorded as averages at set intervals. The travel times are therefore for exiting the link not entering the link as the travel time function requires. Cameras failing, or poor weather leads to missing data.
Vehicle collection. Another method of travel time collection is to use the travel times of the vehicles of the VRP company. This can lead to very sparse datasets as there aren’t many vehicles and some of the potential links may not be traversed at all. If we consider the problem across multiple time steps this becomes even more problematic. Thus the collection intervals aren’t even.
Floating car data. Floating car data is collected by a fleet of vehicles travelling within the network (Rahmani et al., 2015). The vehicles are fitted with GPS and other sensors. The sensor information is recorded at fixed time intervals, then sent to the recording station for real time analysis. This data usually requires a lot of pre- processing to extract travel time information, the locations are unlikely to conform to the VRP links, and cars may even travel over multiple links. TomTom collect data from their sat nav users and this is available to purchase (TomTom, 2015). As there are many users, and this has been collected since 2006 then the dataset should have sufficient depth. TomTom clean the data before use. As more people are using
CHAPTER 2. EVALUATING SINGLE LINK MODELS 68
internet connected phone apps with GPS enabled rather than a sat nav then there is potentially much more frequent floating car data that is being collected.
Mobile phone network. Using the mobile phone network is similar to floating car data except that the data collection occurs only under certain conditions. A phone must be either making a call or have a data connection. Locations are determined by triangulation. This means that there is less data available, as few drivers use their phones while driving, and it is unlikely that phones would be in use both at the start and end of the link. Janecek et al. (2015) suggest an alternative method of using idle phone signals. These signals are generated when phones check for new messages and hence are much more frequent. Both types of mobile phone data have data processing issues, and must eliminate phones that are being used by people walking or cycling if the traffic is stationary. The data also relies upon the phone companies so would have to be purchased.