The rapid growth of electric vehicle adoption is placing new pressure on power systems, not only because electric vehicle charging loads are increasing, but also because those loads are becoming more uncertain. An electric vehicle is no longer merely a mobile electrical load. It is also a travel-choice agent whose charging behavior is shaped by route conditions, congestion, travel time, comfort, weather, charging availability, risk attitude, and the psychological reference points that users apply when judging gains and losses. The analysis under review examines this problem from the perspective of bounded rationality. It argues that electric vehicle users do not behave as perfectly rational decision makers. Their travel uncertainty and their differences in risk appetite produce bounded rational behavior, and that behavior feeds directly into electric vehicle charging demand.
The study combines prospect theory with a risk-based multi-attribute decision-making method. It considers three forms of uncertainty in electric vehicle travel decisions: interval numbers, crisp numbers, and triangular fuzzy numbers. It analyzes how reference points for these three attribute types relate to risk preference coefficients, builds a variable-coefficient risk multi-attribute electric vehicle travel decision model, and forms travel decisions based on the comprehensive prospect value of each route. It then uses the method of successive averages to update and allocate regional electric vehicle traffic flow dynamically. On that basis, it builds an electric vehicle charging demand model driven by comprehensive prospect values and examines how bounded rational behavior affects daily electric vehicle charging demand. The model is validated in a Nguyen-Dupuis network.
The central message is that electric vehicle charging demand cannot be understood solely through travel demand, battery state, and charging power. It is also influenced by how electric vehicle users perceive uncertain travel conditions and how they weigh gains and losses against reference points. When electric vehicle users are more risk-seeking, charging demand curves can become more volatile. When electric vehicle users become more risk-neutral, the curves can become smoother. When attribute weights change, the overall height of the electric vehicle charging demand curve can shift, even if the degree of volatility remains broadly similar under an unchanged risk appetite. These findings have practical meaning for charging infrastructure planning, grid load optimization, power system security, and electric vehicle aggregators that participate in electricity markets.
-
Why bounded rationality matters for electric vehicle charging demand
Electric vehicle travel is a dynamic, multi-stage process. A complete trip involves choices about travel time, purpose, route, mode, cost, weather exposure, road congestion, convenience, and comfort. These factors interact and constrain one another. Different electric vehicle users react differently when these factors cannot all be optimal at the same time. At each travel node, an electric vehicle user must weigh the real-time situation against personal needs and select the route that appears to provide the greatest perceived utility. That perceived utility is not the same as a purely objective minimum-cost calculation. It reflects psychological reference points, loss aversion, risk appetite, and the user’s reading of uncertain conditions.
The analysis groups multiple travel uncertainty factors into three attribute classes. The first class uses interval numbers and includes factors such as electric vehicle travel congestion and travel time. The second class uses crisp real numbers and includes factors such as congestion rate and charging pile availability. The third class uses language variables that are converted into triangular fuzzy numbers and includes factors such as weather conditions and convenience and comfort. Each class can be further divided into benefit-type and cost-type attributes. For benefit-type attributes, a better state corresponds to a larger attribute value. For cost-type attributes, a worse state corresponds to a larger attribute value. The model considers three states for each attribute: good, medium, and poor.
In the specific decision model, the first attribute is electric vehicle travel congestion rate, defined as the number of electric vehicles traveling on a road segment divided by the maximum capacity of that segment. The second attribute is electric vehicle travel time on the road segment. The third attribute is convenience and comfort, defined as the psychological feeling of electric vehicle users under the influence of weather, vehicle speed, road conditions, and related factors. Electric vehicle users generally prefer routes with lower congestion, shorter travel time, and higher convenience and comfort. Therefore, the first and second attributes are cost-type attributes, while the third attribute is a benefit-type attribute.
- Travel uncertainty: Electric vehicle users face uncertain congestion, travel time, weather, comfort, and charging conditions.
- Risk appetite differences: Electric vehicle users do not share one fixed attitude toward gains and losses.
- Reference-point effects: Electric vehicle users evaluate outcomes relative to expectations, not only in absolute terms.
- Charging demand consequences: Route choices influenced by bounded rationality change the spatial and temporal distribution of electric vehicle charging demand.
This matters for electric vehicle aggregators because aggregators collect charging demand across time periods and participate in electricity market transactions. A deviation between expected and actual electric vehicle charging demand can lead to penalties and lower expected revenue. If electric vehicle charging demand analysis ignores bounded rationality, it may miss the psychological and behavioral drivers that make demand fluctuate. The analysis therefore treats bounded rationality not as a secondary detail but as a core source of electric vehicle charging demand uncertainty.
-
A prospect-theory risk multi-attribute decision framework for electric vehicle travel
The framework begins with a set of electric vehicle travel route alternatives. If an electric vehicle travels from an origin to a destination, multiple routes may be available. The decision model treats each route as an alternative and evaluates it across the three uncertain attributes. The expected value of each attribute across states is used as the reference point. For each state, the model compares the attribute result of a route with the reference point and computes gains and losses. This is the point at which prospect theory enters the electric vehicle route-choice problem: electric vehicle users are assumed to evaluate outcomes as gains or losses relative to a reference point rather than as absolute outcomes.
For interval-number attributes, the actual attribute value is assumed to lie within an interval and to follow a normal distribution. The expected value is the midpoint of the interval, and the standard deviation is one-sixth of the interval width. The model considers three possible relationships between the interval attribute value and the reference point: the lower bound is at or above the reference point, the upper bound is at or below the reference point, or the reference point lies inside the interval. Gains and losses are calculated by integrating the relevant part of the probability density function. This allows the electric vehicle travel decision model to represent uncertain travel time or congestion conditions without reducing them to a single deterministic number.
For crisp-number attributes, the attribute value is a real number. The model compares that number directly with the reference point. If the attribute is cost-type and the value is below the reference point, the route obtains a gain; if the value is above the reference point, the route incurs a loss. For triangular fuzzy-number attributes, the attribute value is represented by a triangular fuzzy number with lower, middle, and upper values. The model compares the fuzzy attribute value with the reference point through its membership function. The attribute value may be entirely above the reference point, entirely below it, or straddle it. Gains and losses are obtained by integrating the corresponding parts of the membership function. This structure allows convenience, comfort, weather, and similar language-based factors to enter the electric vehicle decision model in a consistent way.
Language variables and corresponding triangular fuzzy numbers used for electric vehicle travel attributes Language variable Corresponding triangular fuzzy number Very low (1, 2, 3) Low (2, 3, 4) Relatively low (3, 4, 5) Medium (4, 5, 6) Relatively high (5, 6, 7) High (6, 7, 8) Very high (7, 8, 9) After the gain and loss matrices are obtained, the model applies prospect theory value functions and probability weighting functions. The gain value function uses the gain raised to a risk-preference exponent. The loss value function uses a loss-aversion coefficient multiplied by the absolute loss raised to the same exponent. The probability weighting functions transform objective state probabilities into decision weights for gains and losses. In the baseline parameterization, the risk-preference exponent is 0.88, the loss-aversion coefficient is 2.25, and the curvature parameters are 0.61 for gains and 0.69 for losses. These parameters reflect loss aversion and the nonlinear weighting of probabilities in electric vehicle travel decisions.
The model then computes a prospect value for each route and attribute by combining gain and loss value functions with gain and loss probability weights across the three states. The resulting prospect decision matrix is normalized. For each attribute, the maximum absolute prospect value across routes is used as the normalization denominator. The normalized prospect values are then weighted by attribute importance, producing a comprehensive prospect value for each route. A larger comprehensive prospect value means that the route is more attractive to electric vehicle users under bounded rationality. This comprehensive value is not simply a travel-time score. It combines congestion, travel time, convenience and comfort, uncertainty, loss aversion, probability weighting, and risk preference in one decision index.
-
Variable risk preference and electric vehicle route ranking
The analysis emphasizes that the risk-preference coefficient in prospect theory should not be treated as an unchanging constant for all electric vehicle users and all contexts. In much of the existing literature, the risk-preference coefficient is fixed at 0.88. That approach can show that electric vehicle users are not fully rational, but it cannot represent differences in risk attitude across users or across reference-point conditions. In reality, electric vehicle users differ in how willing they are to accept uncertain outcomes. Their risk preference is also related to the reference point they use. If the reference point is larger, the same absolute gain or loss may attract less attention, and the user’s risk preference may be lower. The reference point is the boundary between gain and loss, so a shift in that boundary changes how electric vehicle users perceive and respond to uncertainty.
The study therefore introduces a variable risk-preference coefficient for each attribute and state. The coefficient depends on the attribute reference point relative to the sum of reference points across states and on a travel-scale parameter. The travel-scale parameter lies between 0 and 1 and represents the size of the survey sample or travel scale. As the travel scale becomes larger, electric vehicle users gradually move from risk-seeking toward risk-neutral behavior. The variable coefficient is substituted into the prospect value function. This makes the electric vehicle travel decision model more responsive to differences in risk appetite and to the relationship between reference points and risk sensitivity.
For the initial case, the attribute weight vector is set to 0.2 for congestion rate, 0.5 for travel time, and 0.3 for convenience and comfort. With the fixed risk-preference coefficient of 0.88, the comprehensive prospect values for six routes from origin N1 to destination N3 are as follows: route 1 has a value of -0.7486, route 2 has 0.0729, route 3 has -1.0000, route 4 has -0.2004, route 5 has 0.0733, and route 6 has 0.0874. The ranking is route 6, then route 5, then route 2, then route 4, then route 1, then route 3. Routes 2, 5, and 6 have positive comprehensive prospect values, which means that under the three uncertainty factors and the assumed risk preference, electric vehicle users are likely to select those routes with higher probability.
Comprehensive prospect values for electric vehicle routes from origin N1 to destination N3 under the initial attribute weights Route alternative Comprehensive prospect value Ranking position Route 1 -0.7486 5 Route 2 0.0729 3 Route 3 -1.0000 6 Route 4 -0.2004 4 Route 5 0.0733 2 Route 6 0.0874 1 The variable risk-preference analysis shows that route rankings can change when the travel-scale parameter is varied from 0 to 1. As the parameter increases, the risk-preference coefficient decreases, and electric vehicle users become less sensitive to risk. The comprehensive prospect values of the routes respond to different degrees. Near a travel-scale parameter of 0.5, the curves for route 5 and route 6 cross, and the preferred electric vehicle travel route changes. When risk preference is relatively high, users lean toward risk-seeking behavior and are more likely to choose route 5, which is relatively strong across the three uncertainty factors. As risk preference decreases, users move toward risk neutrality and are more likely to choose route 6, which is relatively stable and less challenging. This result demonstrates that risk preference differences can materially affect electric vehicle route choice and, consequently, the spatial distribution of electric vehicle charging demand.
The practical implication is that a single fixed risk coefficient cannot fully represent the bounded rational behavior of electric vehicle users. Combining risk preference with reference points produces a more realistic picture of how electric vehicle users respond to uncertain travel conditions. This is important because route choice determines where electric vehicle charging demand appears and when it appears. If an electric vehicle user chooses a route with different congestion, travel time, and comfort characteristics, the resulting battery state and charging decision may also differ. The variable-coefficient model therefore provides a more behaviorally grounded input for electric vehicle charging demand analysis.
-
Traffic flow updating and the electric vehicle charging demand model
After the route-choice model produces comprehensive prospect values, the analysis updates traffic flow dynamically. A region may contain multiple origin-destination pairs, and each pair may contain multiple routes. Each route consists of several road segments. If all electric vehicle users were to choose only the route with the highest comprehensive prospect value in a single step, traffic flow on that route would rise sharply, congestion would change, and the original prospect values would no longer hold. The model therefore uses the method of successive averages to adjust and reallocate traffic flow across routes. The updated segment flows are fed back into the electric vehicle route-choice model, and the comprehensive prospect values are recalculated. The process continues until the traffic flows in adjacent iterations are sufficiently close, so that the final result gradually approaches equilibrium allocation.
The traffic-flow update uses the classic BPR function to revise electric vehicle segment travel time. Segment travel time increases as the ratio of segment flow to segment capacity rises. The model also derives speed under basic capacity conditions from capacity, segment length, and a unit conversion factor. These updated travel times feed back into the electric vehicle travel decision model. Because congestion and travel time are among the uncertain attributes, the feedback loop connects traffic conditions with bounded rational route choice. The comprehensive prospect value of a route is not static. It changes as electric vehicle flows redistribute across the network.
The charging demand model then uses the updated comprehensive prospect values. For an electric vehicle traveling on a selected route in a given time period, charging time depends on the difference between the desired state of charge after charging and the remaining state of charge before charging, multiplied by battery capacity and divided by charging power. The model also considers anxiety range. The state of charge after charging must be sufficient to cover the remaining route distance plus the user’s anxiety range. This constraint ensures that the charging decision is not only a matter of convenience but also a response to range anxiety and route length.
The total charging demand in a region during a time period is obtained by summing over origin-destination pairs. For each pair, the model multiplies the share of electric vehicle travel demand for that pair, the total electric vehicle travel demand in the period, and the sum over routes of the route-choice probability multiplied by the charging demand of an electric vehicle on that route. The route-choice probability is determined by the comprehensive prospect value. A positive comprehensive prospect value means that, after considering the three travel uncertainty factors and risk preference differences, the electric vehicle user obtains a perceived gain from choosing that route. The charging demand calculation therefore focuses on routes with positive comprehensive prospect values. Under the Wardrop traffic equilibrium principle, the flows on routes finally selected by users reach a balanced state, and the choice probability is treated as an average probability at equilibrium.
This modeling chain is important for electric vehicle charging demand analysis. It starts with bounded rational route choice, feeds route choice into traffic flow redistribution, updates travel times and congestion, recalculates prospect values, and then converts the resulting route probabilities into electric vehicle charging demand. The electric vehicle is thus represented as a traveler and as a charging load at the same time. The model does not assume that electric vehicle users simply choose the shortest path or the fastest path in a deterministic sense. It allows uncertainty, reference points, loss aversion, and risk appetite to shape the route and, through the route, the charging demand.
-
Validation in the Nguyen-Dupuis network
The analysis is validated in a Nguyen-Dupuis traffic network. The study scenario is the morning commute to work. Electric vehicle users are assumed to start from residential locations at nodes N1, N4, and N12 and to travel to the workplace at node N3. The electric vehicle battery capacity is 24 kWh, energy consumption is 30 kWh per 100 km, and the user’s anxiety range is 20 km. Each daily commute is assumed to begin with a full battery. The simulation is performed in MATLAB R2018b. These assumptions provide a controlled setting for examining how bounded rational electric vehicle route choice affects charging demand.
The network contains multiple route alternatives for each origin-destination pair. From N1 to N3, six routes are considered. From N4 to N3, five routes are considered. From N12 to N3, two routes are considered. The route sequences are defined by road segment numbers. The route set allows the electric vehicle decision model to compare alternatives with different congestion, travel time, and comfort profiles.
Electric vehicle route alternatives in the Nguyen-Dupuis network Origin-destination pair Route number Segment sequence N1 to N3 1 1-4-7-12-18 N1 to N3 2 1-4-11-15-18 N1 to N3 3 3-6-7-12-18 N1 to N3 4 3-6-11-15-18 N1 to N3 5 3-10-14-15-18 N1 to N3 6 3-10-17-19 N4 to N3 1 5-6-7-12-18 N4 to N3 2 5-6-11-15-18 N4 to N3 3 5-10-14-15-18 N4 to N3 4 5-10-17-19 N4 to N3 5 9-17-19 N12 to N3 1 4-7-12-18 N12 to N3 2 4-11-15-18 The network parameters include free-flow travel time, capacity, length, and speed for each road segment. These parameters determine how electric vehicle traffic flow affects congestion and travel time. The table below reports the road-segment data used in the validation.
Nguyen-Dupuis network road-segment parameters used in the electric vehicle travel analysis Segment Node order Free-flow time (min) Capacity (veh) Length (km) Speed (km/h) 1 N1 to N12 12 4000 14.2 56.8 2 N12 to N8 36 3500 22.4 78.4 3 N1 to N5 12 3000 11.2 33.6 4 N12 to N6 12 3000 11.2 33.6 5 N4 to N5 12 4000 14.4 57.6 6 N5 to N6 12 3000 4.8 14.4 7 N6 to N7 12 3000 8.0 24.0 8 N7 to N8 12 3000 8.0 24.0 9 N4 to N9 24 4000 19.2 76.8 10 N5 to N9 12 3000 14.4 43.2 11 N6 to N10 12 3000 20.8 62.4 12 N7 to N11 12 3000 14.4 43.2 13 N8 to N2 12 3000 14.4 43.2 14 N9 to N10 12 4000 16.0 64.0 15 N10 to N11 12 3000 9.6 28.8 16 N11 to N2 12 3000 14.4 43.2 17 N9 to N13 24 3500 14.4 50.4 18 N11 to N3 12 3000 12.8 38.4 19 N13 to N3 12 4000 17.6 70.4 The risk gain and loss matrices show how each electric vehicle route performs relative to the attribute reference points. For the first attribute, route 2 has the largest gain in all three states. Because the first attribute is cost-type, route 2 has the smallest attribute value among the six routes and the largest deviation from the expectation, which produces a high gain. For the second attribute, the attribute values of the routes are not far from the expectation, so the gains are generally not large. Since the second attribute is cost-type, route 5 has a relatively small attribute value and therefore a relatively higher gain than other routes. For the third attribute, routes 2 and 6 have relatively large gains because the third attribute is benefit-type and those routes have larger attribute values and larger deviations from the expectation.
The loss side tells a complementary story. For the first attribute, route 3 has the largest loss in all three states. Because the first attribute is cost-type, route 3 has the largest attribute value among the six routes and the largest deviation from the expectation, which produces a high loss. For the second attribute, the losses are generally not large because the route values are close to the expectation. Since the second attribute is cost-type, routes 1 and 3 have relatively larger attribute values and therefore relatively higher losses. For the third attribute, route 3 has the largest loss in all three states because the third attribute is benefit-type and route 3 has the smallest attribute value and the largest deviation from the expectation.
The normalized prospect values show that routes 2, 5, and 6 have relatively large values, while routes 1 and 3 have relatively small values. The comprehensive prospect values then produce the route ranking described earlier. This result is consistent with the idea that electric vehicle users do not respond to a single attribute in isolation. They integrate congestion, travel time, and comfort through a prospect-theory value function, and their bounded rationality changes the route probabilities and thus the spatial pattern of electric vehicle charging demand.
After dynamic traffic flow updating across the region, the final comprehensive prospect values show a clear pattern. For the N1 to N3 pair, routes 2, 5, and 6 have positive comprehensive prospect values. For the N4 to N3 pair, routes 3, 4, and 5 have positive values. For the N12 to N3 pair, route 2 has a positive value. These positive-prospect routes are the ones that enter the electric vehicle charging demand calculation. The result is not a single deterministic route assignment but a prospect-based allocation shaped by uncertainty, reference points, and risk appetite.
-
How risk appetite changes electric vehicle charging demand
The analysis compares several travel-scale parameter values, corresponding to different risk-preference coefficients, and also compares them with the traditional fixed coefficient of 0.88. The travel-scale parameter values considered are 0, 0.5, 0.7, and 1.0. As the parameter increases, the risk-preference coefficient decreases. Electric vehicle users become less sensitive to risk, and their risk preference gradually shifts from risk-seeking toward risk-neutral. They place more weight on relatively stable travel outcomes and are less willing to accept unnecessary uncertainty. The electric vehicle charging demand curve responds to this shift.
The result is that the fluctuation range of the electric vehicle charging demand curve gradually decreases as the travel-scale parameter increases. In other words, when electric vehicle users are more risk-seeking, the charging demand curve is more volatile. When they are more risk-neutral, the curve is smoother. This does not mean that total travel demand disappears or that charging demand becomes flat in absolute terms. It means that the behavioral route-choice component of electric vehicle charging demand becomes less erratic as users become more stable in their risk attitudes. The variable-coefficient risk multi-attribute electric vehicle travel decision model is therefore able to describe dynamic changes in electric vehicle charging demand that a fixed-coefficient model would miss.
Risk-appetite scenarios used to examine electric vehicle charging demand behavior Scenario representation Travel-scale parameter Risk-preference interpretation Scenario A 0 Stronger risk-seeking tendency in electric vehicle route choice Scenario B 0.5 Transition region where route 5 and route 6 cross in preference Scenario C 0.7 Lower risk sensitivity and movement toward risk neutrality Scenario D 1.0 Most risk-neutral among the variable-coefficient scenarios Reference scenario Fixed coefficient 0.88 Traditional fixed risk-preference treatment The route-level behavior behind this result is important. Around the middle of the travel-scale range, route 5 and route 6 exchange positions in the route ranking. Route 5 is relatively strong when the electric vehicle user is more risk-seeking and is willing to accept a route whose performance across the three uncertainty factors is relatively favorable. Route 6 is more attractive when the electric vehicle user becomes more risk-neutral and prefers a relatively stable route with less exposure to uncertainty. The switch between these routes changes the traffic flow pattern and, through it, the distribution of electric vehicle charging demand. This is a direct demonstration that risk appetite is not a cosmetic parameter. It changes electric vehicle travel behavior and electric vehicle charging demand.
For electric vehicle aggregators, this finding suggests that demand forecasting should not rely only on average travel patterns. If the aggregator ignores the risk attitudes of electric vehicle users, it may misjudge the volatility of charging demand. A risk-seeking population can produce sharper fluctuations, while a more risk-neutral population can produce smoother demand. Because aggregators face deviation penalties and revenue uncertainty, a behaviorally informed view of electric vehicle charging demand can improve market positioning and risk management.
-
Attribute weights and the level of electric vehicle charging demand
The analysis also examines how attribute weights affect electric vehicle charging demand. Four attribute-weight scenarios are used. The first scenario uses weights of 0.2 for congestion rate, 0.5 for travel time, and 0.3 for convenience and comfort. The second scenario uses 0.2, 0.6, and 0.2. The third scenario uses 0.1, 0.5, and 0.4. The fourth scenario uses 0.3, 0.4, and 0.3. These scenarios keep some weights fixed while changing others, allowing the analysis to isolate how the relative importance of electric vehicle travel uncertainty factors influences route choice and charging demand.
Attribute-weight scenarios and positive-prospect electric vehicle routes Scenario Attribute weights for congestion rate, travel time, and comfort Positive-prospect routes from N1 to N3 Positive-prospect routes from N4 to N3 Positive-prospect routes from N12 to N3 Scenario 1 (0.2, 0.5, 0.3) 2, 5, 6 3, 4, 5 2 Scenario 2 (0.2, 0.6, 0.2) 2, 5, 6 3, 5 2 Scenario 3 (0.1, 0.5, 0.4) 2, 5, 6 5 2 Scenario 4 (0.3, 0.4, 0.3) 2 3, 5 2 For the N1 to N3 pair, scenarios 1, 2, and 3 produce positive comprehensive prospect values for routes 2, 5, and 6. In scenario 4, only route 2 has a positive value. For the N4 to N3 pair, scenario 1 produces positive values for routes 3, 4, and 5. Scenarios 2 and 4 produce positive values for routes 3 and 5. Scenario 3 produces a positive value only for route 5. For the N12 to N3 pair, only route 2 has a positive value under all attribute-weight scenarios. These differences show that electric vehicle users with different attribute weights may choose different routes, and the set of routes that enter the charging demand calculation can change.
The comparison between scenario 1 and scenario 2 illustrates the mechanism. When the weight on congestion rate is unchanged, and the weight on travel time increases while the weight on comfort decreases, the comprehensive prospect values of routes 1, 2, and 6 decrease, while the comprehensive prospect values of routes 4 and 5 increase. The reason is that routes 1, 2, and 6 have smaller normalized prospect values under the second attribute than under the third attribute. When the second attribute becomes more important and the third attribute becomes less important, their comprehensive prospect values fall. Routes 4 and 5 have relatively larger normalized prospect values under the second attribute, so their comprehensive prospect values rise.
Across the four scenarios, the general pattern is that when the weight of a given attribute increases, routes with larger normalized prospect values under that attribute tend to gain, while routes with smaller normalized prospect values under that attribute tend to lose. The resulting set of positive-prospect routes changes. Because the electric vehicle charging demand model uses only positive-prospect routes, a change in attribute weights can shift the overall level of electric vehicle charging demand. The analysis reports that different attribute weights can make the regional electric vehicle charging demand curve higher or lower as a whole. However, because the risk-preference degree is held constant in this comparison, the fluctuation range of the curve remains broadly similar.
This result reinforces the distinction between the level and the volatility of electric vehicle charging demand. Attribute weights influence which electric vehicle travel uncertainty factors matter most and therefore influence the height of the demand curve. Risk preference influences how users respond to uncertainty and therefore influences the volatility of the demand curve. Both are part of bounded rationality, and both are relevant to electric vehicle charging demand analysis.
-
Implications for grid operations and electric vehicle aggregators
The analysis offers several practical implications. For charging infrastructure planning, the results suggest that electric vehicle charging demand should be evaluated with attention to route choice under uncertainty. A planning process that assumes deterministic shortest-path behavior may miss the concentration or dispersion of electric vehicle charging demand caused by bounded rationality. For power grid load optimization, the results suggest that the volatility of electric vehicle charging demand is linked to user risk appetite. A grid operator that understands this link can better anticipate periods of higher uncertainty and can design more robust operating strategies.
For electric vehicle aggregators, the results are particularly relevant. An aggregator collects electric vehicle charging demand across time periods and participates in electricity market transactions. If actual electric vehicle charging demand deviates from the expected profile, the aggregator may face penalties and reduced expected revenue. The analysis shows that bounded rational electric vehicle route choice can change the spatial and temporal distribution of charging demand. A behaviorally informed model can therefore help aggregators reduce deviation risk and improve market performance. It can also support charging facility planning, grid load optimization, and the safe and stable operation of the power system.
The variable-coefficient risk multi-attribute electric vehicle travel decision model provides a way to connect psychological factors with engineering demand analysis. It does not treat electric vehicle users as identical rational agents. It allows different electric vehicle users to have different risk attitudes and different reference points. It also allows the same user to respond differently as the travel context changes. This is a more realistic basis for analyzing electric vehicle charging demand than a fixed-coefficient approach alone. The comprehensive prospect value becomes a bridge between electric vehicle travel behavior and electric vehicle charging load.
The study also shows why electric vehicle charging demand analysis should be dynamic. Traffic flow changes after electric vehicle users choose routes. Congestion and travel time then change. Prospect values are recalculated, and route probabilities are updated. The charging demand model uses the updated route probabilities. This feedback process is closer to the real interaction between electric vehicle travel and traffic conditions. It also provides a structured way to update electric vehicle charging demand as network conditions evolve.
-
Limitations and future electric vehicle research directions
The analysis recognizes several directions for further work. One direction is to combine the risk-preference value with additional factors such as travel scale, user type, and travel purpose. The present model links risk preference to reference points and travel scale, but electric vehicle users may differ by commuter type, commercial vehicle operation, leisure travel, and other characteristics. Including those factors could make the electric vehicle travel decision model more complete and could make route predictions more consistent with actual electric vehicle travel behavior.
A second direction is to build more comprehensive bounded rationality models from additional perspectives. The current analysis considers three travel uncertainty attributes and risk preference differences. Future work could extend the behavioral dimensions and examine how electric vehicle users form expectations, update reference points, and respond to charging availability. Such work could help electric vehicle aggregators analyze electric vehicle charging demand more accurately and could support more refined market strategies.
A third direction is to improve the connection between route-choice prediction and actual electric vehicle travel data. The analysis validates the model in a Nguyen-Dupuis network, which provides a controlled test environment. Further validation with real electric vehicle travel data could strengthen the behavioral realism of the model. It could also reveal how different user groups and different travel purposes affect the relationship between bounded rationality and electric vehicle charging demand.
-
Conclusion
The analysis leads to several conclusions about electric vehicle charging demand under bounded rationality. First, the bounded rationality of electric vehicle users introduces uncertainty into route choice. The comprehensive prospect value obtained from the electric vehicle travel decision model includes the joint influence of three types of travel uncertainty factors and risk-preference differences. Using that value to select the preferred electric vehicle route makes the result more accurate and more consistent with real user behavior than a purely deterministic route choice approach.
Second, the risk-preference coefficient in prospect theory is not immutable. Changes in that coefficient have a substantial effect on electric vehicle route choice. Electric vehicle users with higher risk preference are more likely to choose routes whose outcomes are strongly affected by travel uncertainty factors. Electric vehicle users with lower risk preference are more likely to choose routes whose outcomes are less affected by those factors. Combining electric vehicle user risk preference with the reference points of travel uncertainty factors provides a more realistic representation of bounded rational electric vehicle travel behavior.
Third, electric vehicle charging demand is affected by both risk-preference differences and travel uncertainty factors. Risk-preference differences affect the fluctuation range of the regional electric vehicle charging demand curve. A higher risk-preference degree produces a larger fluctuation range, while a lower risk-preference degree produces a smaller fluctuation range. The attribute weights of travel uncertainty factors affect the height of the electric vehicle charging demand curve. When the factor with a larger weight corresponds to larger charging demand, the electric vehicle charging demand curve is higher. These results show that behaviorally grounded modeling can describe the dynamic variation of electric vehicle charging demand more effectively than models that ignore bounded rationality.
The broader significance is clear for electric vehicle integration. As electric vehicle adoption grows, charging demand will become more than a technical load profile. It will reflect how electric vehicle users perceive uncertainty, how they weigh gains and losses, and how they respond to risk. Incorporating those behavioral dimensions into electric vehicle charging demand analysis can support charging infrastructure planning, grid load optimization, power system security, and electric vehicle aggregator participation in electricity markets. The analysis therefore provides a structured behavioral foundation for understanding electric vehicle charging demand in a future with large-scale electric vehicle penetration.