5. Análisis de la evidentia en las Coplas de los pecados mortales
5.4. Sobre las figuras del narrador y del lector
1 48.08 24.48 0.22 3.29 186.7 318.8 0.30 4.90 27.48 0.32 2 49.90 22.61 0.21 3.32 189.9 313.4 0.31 4.93 28.33 0.35 3 49.82 23.52 0.22 3.41 192.9 321.2 0.31 5.04 27.67 0.35 4 47.83 22.44 0.24 3.21 181.7 304.8 0.29 4.70 28.53 0.30 5 48.00 22.68 0.22 3.05 175.2 291.9 0.29 4.80 28.01 0.32
Average 48.73 23.14 0.22 3.26 185.3 310.0 0.30 4.87 28.00 0.32
Std. Dev. 1.04 0.86 0.01 0.14 7.01 11.95 0.01 0.13 0.44 0.02
COV (%) 2.14 3.69 5.16 4.21 3.78 3.86 3.58 2.72 1.56 6.95
4.3 Sensitivity Analysis
An essential step when developing a modeling methodology is to analyze the model sensitivity to variations in input parameters. A sensitivity analysis provides insight into the internal function of a
model and helps to develop understanding and intuition on how the variation of the input parameters propagates through the model.
The input parameters to the model may be divided into three groups: (1) emissions (or fuel consumption) measurements over the baseline cycles, (2) cycle’s properties for the unseen cycle, and (3) cycle’s properties for baseline cycles. There are many ways to do sensitivity analyses. An analytical sensitivity analysis based on partial derivative was done for the inputs of groups (1) and (2) because those terms appear explicitly in equations 21 and 22. For group (3), the input‐output is somewhat more complicated so an empirical sensitivity analysis was conducted for a “typical” model.
4.3.1 Sensitivity to Emissions Measurements and Unseen Cycle Properties
The model output can be written mathematically in two equivalent ways. Equations 21 and 22 represent the model for the case of prediction of fuel consumption (FC) using three baseline cycles. Equation 21 shows the relationship between model output (predicted FC) and the inputs of measured FC. On the other hand, Equation 22 shows the relationship between model output (predicted FC) and the inputs of baseline cycles’ (or routes’) properties. The equations are equivalent and are just two different ways of representing the model (the equation of the plane defined by the three baseline cycles).
· · · (21)
· · (22)
Sensitivity functions for the model can be defined in terms of partial derivatives. The absolute‐sensitivity of the function F to variations in the parameter x is given by Equation 23 (Smith et al., 2008).
(23)
The relative‐sensitivity of the function F to variations in the parameter x is defined by Equation 24 (Smith et al., 2008).
∆
∆ (24)
The sensitivity functions are used to calculate changes in the output due to changes in the inputs or model parameters. They are useful to compare the effects that different parameters have on the output of the model. The absolute sensitivity functions show the most important parameters for a fixed size change in the parameters, while the relative‐sensitivity functions show the most important parameters for a certain percent change in the parameters (if the relative‐sensitivity function has a value of 10, that means that 1% change in the input parameter produces a 10% change in the output of the model).
For the model under study the absolute sensitivity functions for the measured FC (Equations 25, 26, and 27), and for the unseen cycle properties (Equations 28 and 29) are calculated from Equations 21 and 22, respectively. The analytic sensitivity functions for the measured fuel consumptions and the unseen cycle properties are constants so the sensitivities to changes on these parameters do not depend on the variability of other input parameters. Note that the linear interpolation methodology produces a different model for each attempted (unseen cycle) prediction, and therefore these constants will be different for each prediction made.
(25)
(26)
(27)
(28)
(29)
Because of the linear nature of the model (a plane in this case), the partial derivatives at each point will remain constant regardless of where the point is in the surface. Variations in emission measurements and input properties will affect the outcome of the model in a linear fashion. For example, if fuel consumption for all the baseline cycles varies by 10%, the unseen cycle predicted fuel consumption will
also vary by 10%. If the level of variation is different for each baseline cycle, the expected variation will depend on the relative weights of the baseline activity but its value will never surpass the level of variation of the most variable measurement.
4.3.2 Sensitivity to Baseline Cycle Properties
Variations in properties of the baseline cycle values will affect the calculation of the weight coefficients of the baseline cycles. At the same time, the changes to the weight coefficients are going to affect the output of the model (predicted FC for the unseen cycle). Due to the fact that the equations describing the input‐output relationship in this case are complicated (for this specific example there are three equations and three unknowns) a static sensitivity test was applied where a ±10% variation was applied to the property values and the change in predicted values was calculated and analyzed. A “typical” case was selected for illustration purposes. One should have in mind that the methodology produces a different model per each attempted prediction. The weight coefficients will vary for each prediction and performing an individual sensitivity analysis for each model will prove cumbersome.
Vehicle J (see Section 5.2.1, Table 16) was selected for this part of the study. A model was generated using three baseline routes (Idle, WashPA2, and WashPA3) and two route properties (standard deviation of speed, and average positive road load power). CO2 mass rate emissions for five different routes were predicted. Baseline properties were varied +/‐10% one metric at a time. The sensitivity analysis results for this “typical” model are summarized in Table 13. It can be seen that the sensitivity factors (SF) varied from 0.15 to 0.85. The model seems to be more sensitive to road load power (SF from 0.64 to 0.85) than to standard deviation of speed (SF between 0.15 and 0.24). Individual prediction errors varied from 1.69% to 8.82% for the different cases. Weight coefficient values showed more sensitivity varying from 7% to 148%. Note that if all the metrics (baseline and unseen) were varied the model would have produced the same results (being a linear model, the new system of equations would be the same as the original system multiplied by some factor). The use of relative‐sensitivity functions in this case allows a comparison of parameters’ changes on model outputs because they are dimensionless normalized functions. One drawback of using the partial derivative to quantify the influence of an input parameter is that the partial derivative is influenced by the units of measurement of the parameter.
Table 13 Static sensitivity analysis for a “typical” case
Variation (%) Weight coefficients Unseen Route
CO2 mass rate (g/s)
Error (%) SF StdSpeed AvPosRLP w1 w2 w3 Measured Predicted
Base Case
0 0 0.15 0.58 0.27 WashPA1
23.3 21.4 ‐8.3 0
Varying standard deviation of speed 10 0 0.22 0.44 0.34 WashPA1
23.3 21.0 ‐10.0 0.19
10 0 0.01 0.64 0.34 BM2Sab
27.0 24.8 ‐8.1 0.20
10 0 ‐0.06 0.14 0.92 Sab2BM
39.0 32.8 ‐16.0 0.15
10 0 ‐0.14 0.20 0.94 Sab2SW
34.6 34.6 0.2 0.16
10 0 ‐0.08 0.19 0.89 SW2Sab
32.1 32.7 2.1 0.16
‐10 0 0.07 0.76 0.17 WashPA1
23.3 21.9 ‐6.2 0.23
‐10 0 ‐0.17 1.05 0.13 BM2Sab
27.0 26.0 ‐3.9 0.24
‐10 0 ‐0.25 0.55 0.70 Sab2BM
39.0 33.9 ‐13.1 0.19
‐10 0 ‐0.35 0.65 0.70 Sab2SW
34.6 35.9 3.7 0.19
‐10 0 ‐0.27 0.61 0.66 SW2Sab
32.1 33.9 5.7 0.19
Varying average positive road load power
0 10 0.16 0.67 0.17 WashPA1
23.3 20.0 ‐14.1 0.64
0 10 ‐0.06 0.93 0.13 BM2Sab
27.0 23.7 ‐12.2 0.64
0 10 ‐0.13 0.48 0.65 Sab2BM
39.0 31.0 ‐20.7 0.70
0 10 ‐0.22 0.57 0.65 Sab2SW
34.6 32.7 ‐5.3 0.70
0 10 ‐0.14 0.53 0.61 SW2Sab
32.1 30.9 ‐3.5 0.70
0 ‐10 0.14 0.47 0.39 WashPA1
23.3 23.0 ‐1.1 0.78
0 ‐10 ‐0.09 0.69 0.39 BM2Sab
27.0 27.3 1.1 0.78
0 ‐10 ‐0.17 0.14 1.03 Sab2BM
39.0 36.1 ‐7.4 0.85
0 ‐10 ‐0.26 0.20 1.06 Sab2SW
34.6 38.2 10.5 0.85
0 ‐10 ‐0.19 0.19 1.00 SW2Sab
32.1 36.1 12.5 0.85