Truck Volume Percentages

[kengodleskidot]

Below is a description of how truck volumes are estimated in PeMS. The methodology has it's limitations and drawbacks as described below. @kengodleskidot is currently evaluating Weigh-In-Motion (WIM) data that the Caltrans Census group collects and that is reported to FHWA. The preferred option is to incorporate the WIM data in a way where truck volume percentages can be determined and applied to 5-minute aggregate data. We will have further discussions regarding this topic in the future and this issue will be expanded on in subsequent posts.

PeMS Existing Estimation of Truck Volumes: For detectors that don't report truck volumes (currently none of the VDS stations or associated detectors report truck volume), the PeMS system has the ability to estimate them at each detector based on the measured 5-minute, lane-by-lane values of flow and occupancy. If there are sensors that do report truck volumes then we can use those directly. This is a configuration option that is configured during the initial installation.

The initial work on this was in the paper, "Estimation of truck traffic volume from single loop detectors using lane-to-lane speed correlation," by J. Kwon, P. Varaiya, A. Skabardonis (2002), which is available for download in our Resources section.

The algorithm attempts to break down the total flow into passenger cars and large trucks. It does not attempt to identify the proportion of vehicles in each of the 13 classes defined by the FHWA in the United States - that requires detailed measurements that are typically not collected with standard detectors (we assume that since the detectors can't report truck volumes that they are relatively simple sensors like single-loop detectors). The algorithm starts with the following assumptions:

There are no heavy trucks on the inner lanes. For multi-lane freeways, the vehicle speeds over different lanes are synchronized. The traffic volume consists mostly of short passenger cars and long trucks. The average length of passenger cars is 16 feet and the average length of trucks is 60 feet. With these assumptions, the algorithm estimates the proportion of trucks in each lane, starting with the first lane (where we know by assumption that the proportion of trucks is zero), and working to the outer lanes. We store the proportion of trucks in each lane, but we usually plot it only as an aggregate across all lanes since individual per-lane estimates tend to be noisy.

Warning: It is very important to understand that the truck volume estimates from the PeMS algorithm are not measured but estimated from 5-minute aggregated volume and occupancy data. The algorithm works best when the assumptions listed above are satisfied by the traffic at the given location. There are a number of cases when these assumptions can be violated and thus the estimates are unreliable. The two main cases are:

When there is a configuration error and the lane numbers are switched, especially when lane 1 is switched with one of outer lanes, the first assumption is clearly violated. In this case, the algorithm will incorrectly think that an outer lane is actually lane 1 and assign zero trucks to that lane. Further, since the algorithm hinges on the ratios of occupancy in lane 1 and outer lanes, results for other lanes also will be off. When there is a huge speed difference among lanes, the second assumption is violated. Such speed difference occurs especially when there is a lot of merging/diverging, i.e. near on- and off-ramps, or when traffic speed is very low. In these cases, the algorithm will perform poorly since it requires that the speeds in the adjacent lanes be correlated. It is not recommended to look at the truck volume estimate from a single location over a single day. A better way to use the PeMS truck volume estimates is to look at the outputs over many locations, preferably along the corridor in question over multiple days - say over 5 weekdays. The estimates may be unreliable for some of those locations on some of those days, but the empirical studies suggested that the overall trend is captured by the average over locations and times. Specifically, we recommend that the user computes the ratio (daily truck volume to the daily total volume) over each day and location and take their median as the final estimate of the ratio of truck traffic in the corridor. The PeMS displays allow users to view the ratio of truck traffic down the freeway as an easy way to get the estimate.

[junlee-analytica]

Ken to look into WIM data, daily variance on traffic that is trucks. Mintu will keep the item backlogged until the issue is prioritized.

[junlee-analytica]

Ken and Mintu will schedule a meeting to look into WIM and classification data.

[junlee-analytica]

Meeting is set for June 5th.

[junlee-analytica]

Data gathering phase has begun - access to historical time period data to be provided to Ken and Mintu. Provided data needs to be reviewed and analyzed to create scope of effort. WIM team will be updated once this has been completed.

[mmmiah]

May 5, 2024 Meeting resolutions with Census group:

Action Item: Stanley will help to get FHWA account to download WIM class historical data Ken will contact with the corresponding person of TSN to check if they have already processed data Mintu will schedule follow up meeting with Stanley and Vicky to discuss imputation progress and outcomes.

Leanings: Consider Hour of day (HOD), Day of week (DOW) and Season of the Year (SOY) variation to develop the factors (24 (hours) X 7 (days) X 4 season)= total 672 factors) Define Truck before calculating the factors (may be >class4) Consider lane closure impacts in imputation process WIM data quality is pretty good as it is prepared good data only for FHWA submission Truck percentage does not really change over time, we may apply the factors couple of longitudinal direction easily.

[jkarpen]

Per Ken this will likely not happen during the project timeline, will likely come later.

[jkarpen]

Per Mintu there is no update on this as of 7/18 so keeping in the backlog.