Factors affecting fuel economy

Introduction

The fuel economy figures published by car manufacturers are obtained in ideal conditions: cars running in a laboratory on rollers, highly trained driver, etc (see [1]).  Such idealized figures are useful for comparing the fuel economy of various vehicle models, but they are not always easy to reproduce in our day to day driving.  The problem is that fuel economy seems to depend on many factors, and those conditions are not necessarily taken into account in the idealized tests.

My personal experience is that the fuel economy that can be obtained with a car varies a lot, as it may depend on several factors: temperature, traffic, road topography, speed, etc.  Although the fuel economy obtained with the Prius does not vary as much, it still shows significant variance.  Consequently, it is a bit hazardous to try comparing the fuel economy obtained by different Prius drivers across the continent, as they drive in very different conditions, and probably in a very different way. 

My main motivation when I purchased a Prius was to drive a car that can provide the best possible fuel economy.  But in addition to being equipped with high tech devices to save fuel, the Prius offers a nice display board that informs the driver, in real time, on how good (e.g. "ecological") his driving is, and how those results compare with those obtained yesterday or last month.  However, in spite of how advanced the technology might be, it is believed that the way the driver operates the car has a strong influence on fuel economy - and that is independent of the car.  So the idea behind those real time displays is to "put the human into the equation", by providing him with info that will help him achieve better fuel economy.  Thanks to those gadgets, one thing I often try to do is to drive in such a way to get the best possible fuel economy.  The problem is I don't know whether my efforts do have an effect, because the fuel economy figures vary a lot.  It is very hard to make any comparison, as the variance is too high.

In my last post, I studied the problem of the fuel economy gauge accuracy: I recorded the amount of fuel, distance, and fuel economy (FC) every time I filled the tank up.  It revealed a nice seasonal variation, but it was based on average FC between the fill ups.  Considering that I might drive in very different conditions between the fill ups, I needed something more accurate, and more real time.

Therefore, I designed another scientific experiment, which goal was to:

1. Find the various factors affecting fuel economy, and their relative importance.
2. Find out how much the way I drive does affect fuel economy.

Method

In the remainder of this document, I will use "Fuel consumption" (FC) instead of "Fuel economy" (FE).  The reason is that in Canada, vehicles' fuel efficiency is measured in L/100 km which, according to [2], is actually called "fuel consumption".

The experiment took place between April 2010 to May 2011, in Québec, Canada.  It consisted of recording FC for each trip, along with various other variables that might affect it.  To minimize the variance between observations, the fuel consumption was always measured on the same 2 trips each day: going from home to work, and coming back from work to home, about 21 km each way.  The same road was always used each way, and whenever a significantly different road had to be used (e.g. because of road works), then those records were discarded.  By measuring the FC always obtained on the same routes, the hope was that all the variation observed in the FC readings would come only from factors related with the driving conditions.  This way, the relative effect of those factors could more easily be estimated.

For each trip, the following data was recorded:

• Start of trip
• Date and Time of day.
• Battery level (bars).
• Outside temperature as displayed by the climate control display (deg C).
• End of trip
• Distance traveled (km) (as measured by odometer).
• Fuel consumption for trip (L/100km), as displayed by trip stats, and corrected to take account of temperature, using the method described in my previous post.
• Average speed for trip (km/h) (as displayed by trip stats).
• Battery level (bars).
• Outside temperature (deg C).
• Arrival time.
• Fan speed for trip when A/C in use (1-8, 0 if no A/C).
• Road surface state (dry / wet / snow / slush).
• Weather (sunny / cloudy / rain etc.).
• Notes (e.g. stopping to post office, heavy traffic, etc.).
  

To minimize the variance between observations, some elements that might affect FC were kept constant as much as possible:

• Tire pressure was maintained to 34 PSI all year round, but pressure was only adjusted about 4 times a year, to counter the effect of varying temperature.
• The vehicle was always kept in the garage all night.  Our garage is not heated, but since it is attached to the house, its temperature goes only about -5 deg C during the cold winter nights.  Whenever the car was used before going to work, data was not recorded when going to the office on that day.
• The Prius was always kept outdoor during the day, at the office. 
• Whenever the car was used at lunch time or at any other time during the day (except for going to the office in the morning), data was not recorded for the evening return trip.  The goal was to start always with a cold engine, that is after the car had been parked without running for at least 7 hours.
• The cabin heater was maintained at 20.5 deg C during the winter, and A/C at 23 deg C during the summer.  Climate control "auto mode" and "recycle" were not used.
• There were always 2 people for the trips, each way.  The driver was always the same, and so was the passenger.  
• Cargo varied slightly, from 2 lunch boxes to an additional 2 laptop computers + a few little items.
• Fuel level in tank was not recorded (actually, I forgot to do it...).
• No grille blocking procedure was done (the procedure being aimed at keeping the engine warmer during the cold months).

To evaluate the effect of the normal/ECO mode on FC, a driving mode was randomly chosen using OpenOffice Calc random number generator, and printed out on the data recording sheet.  Before each trip, I read the driving mode to use for that trip on the recording sheet, and set the mode right after starting the hybrid system.

To evaluate the effect of the driving technique on FC, a driving technique was randomly chosen using the same method as above, and printed out on the data recording sheet.  Before each trip, I read the name of the driving technique to use for that trip on the recording sheet, and used that technique for the entire trip.  Two driving techniques were used:

1. "Normal", in which the Prius was driven pretty much like I used to drive my previous vehicle.  Before starting to drive in this mode, I switched the display to something different than the Hybrid System Indicator, to make sure I did not get real time feedback on my driving.
2. "Care", in which I would try to optimize the use of gliding and stealth mode, and I would often watch the Hybrid System Indicator to make sure I left the ECO zone only when absolutely necessary, therefore making very progressive accelerations most of the time.

In all cases, however, I put safety ahead of fuel consumption: I did some hard braking operations or hard acceleration, even when driving in "care" mode, when required for safety.

It is important to mention that although at first my driving in "normal" mode was significantly different than the one in "efforts" mode, that difference seemed to decrease with time, as I got used to driving the Prius.  For instance, the gliding has now become a reflex for me, and I even "try" to glide when I drive non-hybrids, just because I am used to do it.

Finally, the topography between home and work is hilly: my home being about 250 m higher in altitude than the office (see Figure 1).  The road between the office and home is approximately the same, but in the opposite direction.  Consequently, the FC in the morning (downhill) is significantly lower than in the afternoon (uphill), as the road presents many opportunities for gliding.

Figure 1: Road altitude between home (left) and office (right).  Data interpolated from [3].

Therefore, the data was collected for 2 different paths: going to work in the morning (downhill), and going back home in the afternoon (uphill).  The 2 data sets were studied independently, and also compared.

Results

Results for downhill trip

A total of 105 data points were collected for the downhill trip. To roughly get an idea of which parameter has an influence on fuel consumption, I have run a linear regression analysis on the following variables:

Dependent variable:

• Fuel consumption (L/100 km) 

Independent variables:

• Winter tires (0=summer, 1=winter)
• A/C (0=no AC, >0=fan speed)
• Distance (km)
• Average speed (km/h)
• Temperature (C)
• Eco mode (0=normal, 1=Eco)
• Driving technique (0=normal driving, 1=care)
• Battery level difference (bars, end value - start value)
• Road (0=dry, 1=wet)

Results of the analysis are given below:

• R squared = 0.8658    That means 86.58% of the change in fuel consumption can be explained bye the change in the independent variables
• Standard error = 0.2512    That value is to +/- on result of the regression equation
• F - Statistic = 68.0742    That means the analysis is significant

Table 1: Results of a linear regression analysis on several variables 

thought of having an influence on FC.

	R Squared	Tires	A/C	Distance	Speed	Temp	Eco	Driving	Batt	Road
Tires	64.76%	100%	2%	11%	0%	65%	0%	1%	0%	3%
A/C	1.45%	2%	100%	17%	0%	11%	1%	0%	0%	1%
Distance	6.88%	11%	17%	100%	0%	12%	0%	0%	2%	3%
Speed	0.03%	0%	0%	0%	100%	0%	0%	3%	4%	10%
Temp	78.13%	65%	11%	12%	0%	100%	0%	1%	0%	4%
Eco	0.04%	0%	1%	0%	0%	0%	100%	0%	0%	1%
Driving	7.82%	1%	0%	0%	3%	1%	0%	100%	30%	0%
Batt	3.43%	0%	0%	2%	4%	0%	0%	30%	100%	1%
Road	3.98%	3%	1%	3%	10%	4%	1%	0%	1%	100%

The R Squared value roughly indicates which variables are important in the analysis.  A rule of thumb is that a R Squared value above 50% indicates a strong relationship.  Here, we see that Tires and Temperature have a strong influence on fuel consumption.  The rest seems minor.

The R-Square matrix (columns to the right of the R Squared value in Table 1) is also interesting.  It tells us how much collinearity exists between the independent variables.  Variables that behave the same way will have a high % in the matrix.  As a rule of thumb, values higher than 90% indicate a strong collinearity and should be omitted from the analysis.  There are no such high values in our matrix.  However, there are some indications of collinearity.  For instance, the table reveals relationships between the following variables:

• Tires and temperature (65%) - this is to be expected, as winter tires are installed used only during the winter.
• Tires and distance (11%) - that is interesting.  That probably means the diameter of the winter tires is different from the one of the all-season tires, which creates a different distance measurement.  The results show that the covered distance is, on average, 0.44% longer in the winter than in the summer.  On the Prius tire diameter, that corresponds roughly to a difference of 1.3 mm in wheel radius between the 2 types of tires, which is quite realistic.
• A/C and temperature (11%) - that was expected, as A/C is used during the warmer months only.
• A/C and distance (17%) - probably an indirect effect of the relationship between distance and temperature (see below), since A/C is used only when temperature is high.
• Distance and temperature (12%) - probably an indirect effect of the Distance - Tires and Tires - Temperature relationships.
• Speed and Road surface state (10%) - that is interesting.  That means when the road surface is dry, the mean speed is different than when the road surface is wet.  That is probably something we could expect, as traffic tends to be slower in the presence of rain or snow.
• Driving and Battery level difference (30%) - that was expected.  When driving in such a way to minimize the use of fuel, battery usage increases, which probably creates this relationship.

The analysis yielded the following equation, giving fuel consumption value as a function of the independent variables:
FC = 0.294*Tires + 0.411*A/C + 0.060*Distance + 0.001*Speed - 0.042*Temp - 0.046*Eco - 0.197*Driving + 0.038*Batt + 0.074*Road

The R-Squared column of Table 1 tells us the relative importance of each factor on fuel consumption.  Let's study those factors one by one.

Effect of temperature on fuel consumption

Figure 2 shows the seasonal variation of temperature during the downhill trips.  Since the vehicle was always parked in the garage, there is a difference between the temperature at the start of the trip, and the temperature at the end.  The difference increases in the winter, as even though my garage is not heated, it is attached to the house so its indoor temperature does not really go below about -5 C.

Figure 2: Seasonal temperature variation for the downhill trip (morning), 

at trip start (vehicle in garage) and trip end.

The seasonal variation of fuel consumption is shown in Figure 3.  The shape of the curve, compared to the one of Figure 2, clearly suggests a dependence of fuel consumption on temperature.  Such a dependence is well known, and was also highlighted in my previous post.

Figure 3: Seasonal FC variation for the downhill trip (morning).

 The linear regression analysis above indicated that the main factor affecting the variance in FC is temperature (64.76%).  My previous study highlighted a strong dependence of fuel consumption on temperature.  This dependence is confirmed here, and illustrated in Figure 4.

Figure 4: Dependence of fuel consumption on temperature for the downhill trip.

The dependence is clear: the vehicle needed less fuel at higher temperatures.  As the temperature decreases, the vehicle needs an extra 0.05 L/100 km for each degree C to do the downhill trip.  The difference between the best FC value (obtained at about +28 C) and the worst (obtained at about -25 C) is large, going from about 2.25 to about 5.5 L/100km.

When one looks carefully, the relationship between temperature and FC is not exactly linear.  A quadratic equation seems to fit better.  This is illustrated in Figure 5. As it can be seen, the curve fits better, and the coefficient of determination (R2) is slightly higher.

Figure 5:  Dependence of fuel consumption on temperature for the downhill trip, fitted with a quadratic equation. 

This quadratic relationship is interesting.  It means 2 things:

• Fuel consumption does not get infinitely better as temperature increases (it reaches a point where it no longer improves. It even gets worse with increasing temperature past a certain point.
• Fuel consumption gets worse with decreasing temperature, but the decrease is not linear: a difference of 5 degrees C has little influence on FC at warm temperature, but has major influence at cold temperature.

Let's remove the effect of temperature from the data (using the quadratic equation shown in Figure 5).  Doing so, the FC graph looses its seasonal variation, as shown in Figure 6.  On that new graph, no clear seasonal trend is visible.

Figure 6: Seasonal FC variation for the downhill trip (morning)
with the effect of temperature removed.

For studying the remaining factors, the effect of temperature was removed from the FC data.

Effect of tires on fuel consumption

As shown in the R-squared column of Table 1, tires is the second strongest factor that seems to have an effect on FC.  Indeed, winter tires are softer, designed for adherence, and probably require more power from the engine to propel the vehicle.  I equipped my Prius with Blizzaks, installed onto steel rims, further increasing wheel weight and inertia, and therefore the amount of energy required to spin them.  However, the tires factor has a strong dependence on temperature (65% in the correlation matrix).  That was expected since winter tires are used only during the cold season.  The temperature dependence could therefore explain the apparent strong dependence of FC on tires.

Figure 7 shows the same data as in figure 6, but where data points are presented into groups: winter tires (blue) and all-season (summer) tires (red). FC values for all season (summer) tires have an average of -0.026 L/100 km, whilst FC values for winter tires have an average of 0.071 L/100 km.  That represents a difference of 0.097 L/100 km that can be attributed to the use of the 2 different types of tires.

Figure 7: Effect of tire type on fuel consumption.

For studying the remaining factors, the effect of tire type was removed from the FC data.

Effect of driving technique on fuel consumption

As shown in Table 1, the third strongest factor that appears to have an influence on FC is the driving technique.  Again, that was expected.  The study of the influence of the driving technique on FC is particularly interesting for the Prius, because of its instruments panel which provides the driver with feedback on his driving.  The Prius' Hybrid System Indicator (Figure 8) was designed to provide instantaneous, real time feedback, that is useful to adjust one's driving to get better FC.  Such feedback indicator is also found in other hybrids, the idea behind it being to "put the human into the equation", helping the driver adjust his behavior to achieve better FC.

Figure 8: The Prius' Hybrid System Indicator

To estimate how strong the driver effect can be on FC, I have randomly chosen a driving technique each morning: "normal" (where the car would be driven normally), or "care", where the vehicle was driven with FC in mind.  Those are described in the introduction.   The results are shown in Figure 9, that shows the seasonal variation of FC, with the effect of temperature and tires removed, for the 2 driving techniques.

Figure 9: Effect of driving technique on fuel consumption

Figure 9 shows that that driving with care to lower FC does have an effect on FC: the average FC obtained when driving carefully was 0,21 L/100km lower than the average for normal driving.  That difference value is statistically significant.

The  effect of the driving technique is interesting and deserves a deeper investigation.  It is interesting to plot the original FC figures as a function of temperature, but this time highlighting the difference between the 2 driving techniques.  This is illustrated in Figure 10.  The graph shows that although driving with care generally produces lower FC , the difference between the 2 driving techniques decreases at lower temperatures. 

Figure 10: Fuel consumption as a function of final temperature, for each driving technique.

This result is very interesting.  It can probably be interpreted by the fact that the morning drive is downhill.  As shown in Figure 1, the road presents several opportunities for gliding, and gliding is one of the main characteristics of the careful driving technique.  At low temperatures, the engine is required for cabin heating, decreasing the time during which gliding can take place in stealth mode.  In such cases, the careful driving technique becomes much closer to the normal one.

For studying the remaining factors, the effect of the driving technique was removed from the FC data. 

Effect of the "ECO" mode on fuel consumption

The Prius comes with 3 driving "modes": Power, normal, EV and ECO.  The ECO mode is similar to the normal mode.  The 2 modes have some slight differences, the main one being the mapping of the accelerator function: in ECO mode, the driver has to press more on the pedal to get the same acceleration.  Consequently, running in ECO mode often results in softer accelerations and therefore lower FC.  I have always wondered what was the actual effect of that button on FC.  This is particularly important, as after driving in a given mode for a while, on gets "used" to that mode, and "learns" to press more on the accelerator to get the same acceleration as with normal mode.  In the end, I hypothesized the ECO mode might not have a very strong effect after all.

The seasonal variation of FC with the effect of temperature, tire type and driving technique removed is displayed in Figure 11, for both the normal and ECO modes.  It shows a very little difference between the 2 modes, the ECO mode being on average 0.047 L/100km lower than in normal mode.  However, that difference is not statistically significant.

Figure 11: Seasonal variation of FC (with the effect of temperature, tire type and driving technique removed) for Normal and ECO modes.

For studying the remaining factors, the effect of the ECO mode was removed from the FC data.

Effect of the vehicle's weight

The vehicle's weight was relatively constant.  There were always the same 2 people in the car, with roughly the same amount of stuff in the boot.  In the winter, we did carry an extra 4 liters of windshield washing liquid, as well as a set of metallic traction aid.  But since we only carried those during the winter, the effect of that constant additional weight was probably interpreted as the effect of tires in the analysis.  One thing that did vary was the amount of fuel in the regression analysis.  The only varying weight that I could have measured is the amount of fuel in the tank - it would have been easy, an extra column in the table, where I would have recorded the number of bars on the fuel level indicator.  Unfortunately, that information was not recorded in this study.  However, in my previous study, I recorded the date at which I filled the tank, and the amount of fuel purchased each time.  Using simple temporal linear interpolation (assuming I used the same amount of fuel each day), I "regenerated" a column in the file estimating the amount of fuel in the tank.  Figure 12 shows the effect of the estimated amount of fuel remaining in the tank on FC .  The graph shows a lot of variation, but does show a linear tendency, basically adding 0.0054 L/100km for each liter of fuel in the tank.  Knowing the Prius tank can hold 45 L of fuel, that means a difference of 0.25 L/100km between a full tank and an empty one.  That is non negligible. 

Figure 12: Effect of the amount of fuel remaining in tank on fuel consumption.  

For studying the remaining factors, the effect of the fuel weight was removed from the FC data.

Effect of average speed

Speed is a factor known to have an effect on fuel consumption.  In my experiment, speed mostly varied because of traffic conditions and traffic lights.   Figure 13 shows the effect of the vehicle's average speed on FC .  The graph shows a non linear relationship - FC seems to increase from 50 to 65 km/h, and from 50 to 30 km/h. 

Figure 13: Effect of average speed on fuel consumption.

One problem with the concept of average speed is that does not say anything about the actual variance of speed during the trip.  A low average speed value could be obtained by driving at that constant low speed, or by periods of high speed followed by periods of highway parking.  I hypothesized that this non linear relationship was actually caused by temperature.  The effect of temperature was highlighted in Figure 14, that shows the same information as Figure 13, but only for the data points where temperature was higher or equal to 5 deg C.  Figure 15 shows the data points where temperature was lower than 5 deg C.

Figure 14: Effect of average speed on fuel consumption

when temperature was higher or equal to 5 deg C.

 Figure 15: Effect of average speed on fuel consumption

when temperature was lower than 5 deg C.

It is interesting to note that when the temperature is above 5 deg C (Figure 14), speed seems to have a positive linear relationship with FC. On the other hand, when temperature is lower than 5 deg C, the relationship is negative: the lower the speed, the higher the FC.  That interesting effect is probably caused by the fact that at low temperature, engine is running more often, sometimes even in slow traffic, to keep the cabin warm.  Lower speed means more time spent in traffic, therefore higher fuel consumption.  Unfortunately, I do not have enough data points to determine whether a low average speed does have a similar effect in the summer.

Considering those 2 equations, the effect of vehicle average speed can be removed from the data.

Results for uphill trip

A total of 76 data points were collected for the uphill trip. 

Effect of temperature on fuel consumption 

The seasonal variation of fuel consumption is shown in Figure 16, for both the uphill and downhill trips.  Both curves have the same look, showing a similar dependence of FC on temperature.  The graph also shows that the uphill trip approximately requires 3 L/100km more fuel than the downhill trip.

Figure 16: Seasonal variation of the fuel consumption, for the uphill and downhill trips.

Figure 17 shows the dependence of FC on temperature.  Again, the same relationship is observed between temperature and FC: the colder, the higher the FC.  However, since the uphill trip always occurred at the end of the afternoon, we captured more data at high temperatures.  As a result, A/C was used more often.  In Figure 17, the 2 data series are plotted with different symbols, depending on whether A/C was used or not.  The graph clearly shows that when A/C is used, more fuel is required.  However, what proportion of this increase in FC is due to temperature vs A/C is unclear - I do not have enough data points to evaluate it.

Figure 17: Effect of temperature on fuel consumption, for both the uphill and downhill trips.  

The relationship between temperature and FC is therefore not linear.  The effect of temperature can be removed from the data, using the 2 linear equations displayed in Figure 17, based on the use of A/C.

It is interesting to study the effect of A/C fan speed on FC - this is shown in Figure 18.  The graph shows that the higher the fan speed, the higher the fuel consumption.  Again, I do not have enough data points to see how much of that effect is due to the fan compressor, requiring more energy from the engine to recharge the battery, or from higher temperature affecting the hybrid system.

Figure 18: Effect of A/C fan speed on fuel consumption, for the uphill trip.

For studying the remaining factors, the effect of temperature was removed from the FC data. 

Effect of tires on fuel consumption

The uphill trip data shows that on average, the FC (with temperature effect remove) obtained with all season tires is -0.02 L/100km whilst FC values for winter tires have an average of 0.06 L/100 km.  That represents a difference of 0.08 L/100 km.  Those values are very close to those obtained for the downhilltrip, where a difference of 0.09 L/100 km was observed.

For studying the remaining factors, the effect of tire type was removed from the FC data.

Effect of driving technique on fuel consumption

The effect of the driving technique on fuel consumption is shown in Figure 19.  Again, the average value of the FC is lower for the "Care" driving technique than the normal one - a difference of 0.149 L/100 km being observed. 

Figure 19: Effect of driving technique on FC for uphill trip, once the effects of temperature and tires has been removed.

That difference is lower than the one observed for the downhill trip (0.21 L/100 km).  It could be explained by the fact that during the downhill trip, there are several opportunities for gliding, which is one of the main advantages of the careful driving technique.  The uphill trip requires more effort from the engine, and therefore in this case the careful technique is not as advantageous.

Figure 20 shows the effect of the driving technique on FC for varying temperature.  As in Figure 10, it tends to show that the difference between the FC obtained with careful technique and the normal technique is larger at higher temperatures, but the amount of data I have is probably not sufficient to conclude.

 Figure 20: Effect of driving technique on fuel consumption vs temperature, for uphill trip.

This result is also interesting.  It can probably be interpreted by the fact that the afternoon drive is uphill, therefore it does not present many opportunities for gliding.  Now since gliding is one of the main characteristics of the careful driving technique, and that gliding is very much influenced by temperature, it is natural that temperature does not have an effect that is as strong for the uphill than the downhill trip.

For studying the remaining factors, the effect of the driving technique was removed from the FC data.

Effect of the "ECO" mode on fuel consumption

The seasonal variation of FC with the effect of temperature, tire type and driving technique removed is displayed in Figure 21.  It shows a very little difference between the 2 modes, the ECO mode being on average 0.022 L/100km lower than in normal mode.  That is twice as small as the one observed for the uphill trip.

Figure 21: Effect of the use of the EC model on FC (with the effect of temperature, tire type and driving technique removed).

For studying the remaining factors, the effect of the ECO mode was removed from the FC data.

Effect of the vehicle's weight

Using the same linear interpolation technique as above, I "regenerated" a column in the file estimating the amount of fuel in the tank.  Figure 22 shows the effect of the estimated amount of fuel remaining in the tank on FC.  The graph shows a lot of variation, but does show a clear tendency, basically adding 0.0020 L/100km for each liter of fuel in the tank.  Knowing the Prius tank can hold 45 L of fuel, that means a difference of 0.09 L/100km between a full tank and an empty one.   

Figure 22: Effect of the amount of fuel remaining in tank on fuel consumption, for uphill trip.  

Those figures are twice as small as for the downhill trip.  That is rather strange, as one would think that carrying a weight has more effect uphill than downhill.  However, considering the variance of the data, the difference is probably within the error on the slope estimate.

For studying the remaining factors, the effect of fuel weight was removed from the FC data.

Effect of average speed

Figure 23 shows the effect of the vehicle's average speed on FC.  The graph shows a roughly linear relationship.  I tested the same hypothesis as for the downhill trip: that is the relationship is also linked with temperature.

 Figure 23: Effect of average speed on Fuel consumption.

Figure 24 shows the same information as Figure 23, but only for the data points where temperature was higher or equal to 5 deg C.  Figure 25 shows the data points where temperature was lower than 5 deg C.

Figure 24: Effect of average speed on fuel consumption

when temperature was higher or equal to 5 deg C.

 Figure 25: Effect of average speed on fuel consumption

when temperature was lower than 5 deg C.

As in the downhill case, when the temperature is above 5 deg C (Figure 24), speed seems to have a positive linear relationship with FC. Also, when temperature is lower than 5 deg C (Figure 25), the relationship is negative: the lower the speed, the higher the FC.  The slopes are only about 50% of the slopes observed for the downhill case, which indicates here that speed has less of an effect on FC on the uphill trip.

Considering those 2 equations, the effect of vehicle average speed can be removed from the data.

Analysis & Interpretation

The results are summarized in Table 1, that shows the % of the variance that is explained by each variable, both for the uphill and downhill trips.

Table 1: Percentage of variance explained by each variable, for the uphill and downhill trips.

Removed effect of	.	% expl Uphill	.% expl Downhill
Temperature		81.33%	76.33%
Tires		0.33%	0.63%
Driving technique		1.61%	3.16%
ECO		0.04%	0.16%
Weight		0.07%	0.41%
Speed		1.02%	4.14%
Total		84.40%	84.83%

The table shows that temperature accounts for about 80% of the variance, and has a stronger influence for the uphill trip.  The next 2 factors are the driving technique and speed.  Both have more influence on the downhill trip.  Next are the type of tire, followed by fuel weight, then the ECO button, that are all negligible.

I was surprised by those results.  I knew temperature effect was strong, but I would have hoped that the driving technique and the ECO button had more effect.

The strong effect of temperature can be explained by several factors.  Temperature affects grease and oil viscosity, and low temperature requires cabin heating.  Also, air drag is directly proportional to air density [4].  According to [5], the air density at 20 C is 1.2041 kg/m3, whilst at -20 C, it increases to 1.3943 kg/m3.  That causes an increase of 15% in air drag, that the engine must compensate for.  The presence of snow or slush on the road may also affect FC.

It is interesting to note that temperature has a stronger effect for the uphill trip.  This could be caused by the fact that the hybrid system temperature goes to higher extremes during the day as the vehicle is outside, whilst it is parked in a garage during the night.  That possibly creates the stronger influence of temperature on the uphill trip.

The effect of the fuel weight, relatively stronger for the downhill trip, is opposite to what seems logical, and I can find no explanation for this, apart from the fact that the values are very small, and could be just due to random error.

The effect of speed, 4 times as strong for the downhill trip, is also puzzling.  One problem though is that the speed parameter, a one figure value, says nothing about the speed during the trip, which varies a lot.  I suspect we could find a lot of answers if we studied that variation.

The various factors that I studied can only explain about 85% of the variance.  I suspect other factors like speed variation during trips could explain the rest.

Conclusion

The question I wanted an answer for was: what are the factors that influence fuel consumption, and how much does the way I drive affect fuel consumption.

Factors were identified, and quantified.  The driving technique was seen to influence FC by only a small factor w/r to temperature.

Of all the factors that I studied, only a few of them are under my control.  Those are:

• Driving technique
• Driving with care can lower FC by 0.18 L/100km on average
• ECO button
• Using the ECO button can lower FC by 0.034 L/100km on average
• Speed
• Driving slower by 10 km/h can lower FC by 0.15L/100km on average during the summer
• Vehicle weight
• Driving with a half tank instead of a full one can lower FC by 0.08 L/100 km on average

The combined effect of those measures amounts to 0.444 L/100km.  Considering that my yearly average FE on both trips approximately 5 L/100km, that represents 8% of the FC that is in my hands.

Since I drive about 10920 km each year for commuting, that means I use about 546 L of fuel for going to work.  Knowing that 8% of that FC is in my hands, that means 43 L of fuel depends on the way I drive, the weight of the car, my speed and my use of the ECO button.

That is not much, but in a way, knowing that each L of fuel creates 2.3 kg of CO2 [6], then by doing these techniques, I can save about 100 kg of CO2 per year, just by going to work in a wiser way. At the current price of fuel, that saving amounts to about $60 annually. 

Of course, I could move to a warmer country :-)  or move closer to work, and save much more. 

Future work

We could further increase our understanding of the factors that affect FC by doing new experiments.  For instance:

Effect of grille blocking

The results of this study show that temperature is by far the stronger factor affect FC.  Many Prius drivers living in the colder climates use a grille blocking technique - the purpose being to keep the hybrid system warmer, and get lower FC.  It it really worth doing that?  How much fuel can be saved using such a technique?  A new study could easily be designed to test that.

FC measurements with higher temporal frequency

The use of the average speed is limiting, as it does not consider the variations of speed, accelerations and decelerations during the trip.  Another study, with more advanced equipment, could be used to study those variations, and perhaps propose best speed variation through the trip to achieve the lowest possible FC at any time of the year.

Test Power mode

It might be interesting to test Power mode, to estimate how much using that mode does affect FC.

Try various makes of snow tires

The ones I use (blizzaks) are known to by soft, possibly affecting drag.  Other makes may be different in that respect.

Evaluate the effect of using A/C

We know that driving at higher temperatures does affect FC, but we do know know how much of it is specific to the use of A/C.  One possibility would be to drive without A/C, and estimate the difference.  (I would have to convince my wife first, though... )

Improve my driving technique

My "driving with care" technique is basic, and could very likely be improved.  That could result in higher savings when using the care driving technique.

Compare results with plug-in Prius

Of course, it would be great if I could compare those results with FC values obtained with the plug-in Prius.  This way, I could not only evaluate how good the plug-in system is for that specific situation, but also estimate whether that type of hybrid would be appropriate for my type of commute.

References

[1] http://auto.howstuffworks.com/fuel-efficiency/fuel-economy/28004-epa-fuel-economy-explained.htm
[2] http://en.wikipedia.org/wiki/Fuel_efficiency#Fuel_efficiency_of_vehicles
[3] http://atlas.nrcan.gc.ca/site/english/maps/topo/map
[4] http://en.wikipedia.org/wiki/Drag_%28physics%29
[5] http://en.wikipedia.org/wiki/Density_of_air 
[6] http://wiki.answers.com/Q/What_is_the_CO2_emission_per_liter_petrol

 

Notes

I hope you found this useful.  Do you think I forgot something or made a mistake?  Feel free to comment below!  I am very open to suggestions.

Thanks

My sincere thanks to my wife, for her help, patience and understanding during this long study.