Then to monitor the adjustment, there are additional outputs in the outputs tab of all the adjustments within Aimsun Next.
Matrix cell values
To see the changes in matrix cell values, select Cell-by-Cell and the user class that you would like to analyse, then select the scatter graph option. This creates a scatter graph comparing the original and adjusted demands and gives the line of best fit, R2 as well as additional information such as the RMSE.
It can be seen here that the slope of the line of best fit is 1.306 which is above the significance value (0.99-1.01). The user should check that the correct prior matrix was used and why the adjustment is increasing the traffic so much. A change to the slope that is this high, suggests there is a mismatch in what is being compared e.g., time of day, user class, year and that the real data set, centroid configuration and prior matrix methodology should be evaluated.
Trip ends
To see the changes in matrix trip ends, select Trip Ends and the user class that you would like to analyse, then select the scatter graph option. This creates a scatter graph comparing the original and adjusted demands, aggregated by the origin and destination total. It gives the line of best fit, R2 as well as additional information such as the RMSE of these totals.
It can be seen here that the slope of the line of best fit is 1.306 which is above the significance value (0.99-1.01). The user should check that the correct prior matrix was used and why the adjustment is increasing the traffic so much. A change to the slope that is this high, suggests there is a mismatch in what is being compared e.g., time of day, user class, year and that the real data set, centroid configuration and prior matrix methodology should be evaluated.
Trip length distribution
Note: This is only currently available in the Static OD adjustment as this is the adjustment most likely to change the trip length distribution.
To see the changes made to the trip length distribution, go to the trip length distribution tab and select the user class you would like to analyse. A graph is histogram is shown showing the original and adjusted matrix distributions. It should be noted that the bars for the original demand in the histogram are displaced along the x axis by 0.5km so that both histograms can be easily seen. The data can be exported to show externally using the copy distribution data button.
The mean and standard deviation of both the adjusted and original trip length distributions are displayed below the chart to allow the modeler to easily check the significance of the changes to the trip length distribution after the adjustment.
The changes to the trip length distribution mean and standard deviation (1.12% and -4.45%) are under the significance levels given by TAG.
Sector to sector
To see the changes at a sector-to-sector level, select cell by cell, the user class that you would like to analyse and the grouping category which defines your sectors; then select the list option. This shows the percentage difference between the sector-to-sector totals.
Many of the sector-to-sector pairs have changes above the significance level given by TAG (5%). As this measure looks at the structural change to the matrix, the modeler must look at the prior matrix and real data set to understand why there is such a difference between that and what is needed to meet the calibration and validation counts.
All these outputs can be retrieved from an adjustment when the results are saved into an .adj file by right clicking the replication and clicking retrieve adjustment results.
Trip distribution without running the model
It is also now possible to compare the trip distribution of two matrices without running the model by using the shortest path calculator. This can be used to evaluate the trip length distribution for instance after a departure adjustment or dynamic OD adjustment. It does not take into account multiple routing options that may be used in assignment, but it can be a good method to check the trip distribution of a matrix for a large-scale model without having to wait to run the model.
