A vehicle energy supplement forward road network active decision optimization method and system based on agent tool call arrangement and a storage medium

Abstract

The application discloses a vehicle energy supplement forward road network active decision optimization method and system based on agent tool call arrangement. The agent obtains the vehicle position, SoC, energy consumption model and navigation main path in real time, and when the trigger condition is met, a forward road network candidate set defined by a dynamic fan-shaped preliminary screening area and a road network corridor is constructed. By calling the map and charging network interface, the site reachable path, yaw time, U-turn identifier, peak-valley electricity price and queuing state are obtained. The system calculates the cross-period expenditure by section according to the predicted start charging time, and switches among the economic, balanced and safe modes through the SoC state machine with hysteresis. The optimal energy supplement station and navigation strategy are output by using the deterministic comprehensive cost model. The application can effectively reduce the reverse search, U-turn detour and price misjudgment, and improve the on-route nature, economy and safety of the energy supplement decision.

Description

Technical Field

[0001] This invention relates to the fields of intelligent vehicle navigation, new energy vehicle refueling decision-making, map route planning, charging network data processing, and artificial intelligence tool invocation and orchestration. In particular, it relates to a method, system, and computer-readable storage medium for optimizing and recommending refueling sites based on intelligent agents actively perceiving vehicle status, invoking external maps and charging data interfaces, and combining forward road network constraints, peak and off-peak electricity prices, queuing waiting time, and user preferences during vehicle operation. Background Technology

[0002] As the number of new energy vehicles increases, users typically consider factors such as whether a charging station is on their route, whether a U-turn is required, station availability, charging price, waiting time, and remaining battery power when choosing a charging station while driving. Existing map or charging service software often uses a circular search centered on the current location, which can easily recommend stations located behind the vehicle, on the opposite side of the road, or requiring significant deviation from the intended route, forcing users to make additional U-turns, detours, or replan their routes.

[0003] On the other hand, charging prices typically vary during peak, off-peak, and valley periods. The actual cost a user pays depends on the moment the vehicle arrives at the station and begins charging, not just the current electricity price at the time the query is initiated. If the navigation system only displays static prices, it may lead to inaccurate cost calculations in scenarios where prices are about to change, queue times are long, or charging occurs across different time periods.

[0004] Meanwhile, in low SoC scenarios, charging recommendations should not only pursue the lowest cost, but should prioritize ensuring that vehicles can safely reach available stations. In high SoC scenarios or scenarios with sufficient range, economic optimization can be appropriately carried out by utilizing price periods and route conditions. Therefore, a proactive optimization scheme is needed that can unify vehicle status, current navigation route, road network ahead, dynamic prices, available charging stations, queuing status, and user preferences into executable charging decisions. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a method, system, and storage medium for proactive decision-making optimization of the forward road network for vehicle refueling based on intelligent agent tool invocation orchestration, in order to solve the problems existing in the current refueling recommendation, such as backward station misrecommendation, U-turn detour, price timing misjudgment, insufficient security in low-battery scenarios, and difficulty in converting natural language intent into executable refueling constraints.

[0006] Technical solution:

[0007] a. The intelligent agent obtains the vehicle status and the current main navigation path in real time;

[0008] b. When the active energy replenishment triggering condition is met, construct a forward road network candidate set jointly defined by the dynamic sector initial screening area and the road network corridor extending along the main navigation path;

[0009] c. Verify the road network topology of candidate stations through the map route planning interface to exclude stations that make U-turns, enter in the wrong direction, reverse backtrack, or deviate excessively from the course.

[0010] d. Call the charging network interface to obtain the price time table, service fee, charging power, available charging stations, queue status and data update time;

[0011] e. Predict the expected arrival time, expected start time of charging, expected energy replenishment, and SoC of the candidate site;

[0012] f. Determine the economy-first, balance-first, or security-first mode through a SoC state machine with hysteresis;

[0013] g. Then, the deterministic integrated cost model is used for filtering, scoring, and ranking, and the optimal charging station, alternative stations, navigation instructions, and charging strategies are output to the user.

[0014] Beneficial effects:

[0015] First, by using the forward road network candidate set, invalid charging stations behind vehicles, on the opposite side of the road, and where U-turns are required are reduced;

[0016] Second, improve the accuracy of cost estimation by predicting the start time of charging and calculating electricity prices in segments across different time periods;

[0017] Third, by automatically switching between economy and security through the SoC state machine, the goal of minimizing price is avoided when the battery is low.

[0018] Fourth, by using tools to call the orchestration module, the user's natural language intent is transformed into structured constraints, enabling the agent's language understanding ability to work in conjunction with deterministic optimization algorithms, thereby reducing the uncertainty of purely subjective recommendations;

[0019] Fifth, reduce the misleading effect of expired prices, expired stakes, or missing data on recommendation results by using data credibility penalty items. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating how the dynamic sector-shaped initial screening area and the navigation main path road network corridor together constitute the forward road network candidate set in an embodiment of the present invention.

[0021] Figure 2 This is a flowchart of the intelligent agent tool invocation orchestration and power replenishment decision-making process in an embodiment of the present invention.

[0022] Figure 3This is an interactive architecture diagram of the vehicle, cloud-based intelligent agent, third-party map platform, and charging network data source in an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of SoC state machine switching with hysteresis in an embodiment of the present invention. Detailed Implementation

[0024] The present invention will be further described below with reference to embodiments. It should be understood that the following embodiments are used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent substitutions, parameter adjustments or module combinations made based on the technical concept of the present invention should fall within the scope of protection of the present invention.

[0025] Example 1

[0026] In this embodiment, the state perception module obtains the vehicle's current location, driving direction, speed, SoC, battery capacity, vehicle energy consumption model, and current navigation main path from the vehicle diagnostic interface, the vehicle manufacturer's cloud interface, or the vehicle system. The current navigation main path can be represented as a polyline composed of multiple road nodes or broken line segments, and may include road level, speed limit, road segment travel direction, estimated travel time, and congestion status.

[0027] The agent triggers an active charging decision under any of the following conditions: the current SoC is below a preset charging reminder threshold; the predicted remaining SoC after the vehicle reaches its destination is below the destination's safety threshold; there is an available charging station ahead on the current path with a price significantly lower than the average price in the current area; or the user expresses intents such as "find a charging station along the way," "find a cheap charging station," or "find a supercharger that doesn't require a U-turn" using natural language. Upon triggering, the agent does not simply perform a circular search at its current location but instead generates a candidate set of forward road networks.

[0028] The forward road network candidate set first establishes a dynamic fan-shaped initial screening area based on the vehicle's travel direction. The fan angle can decrease as vehicle speed increases; for example, it narrows to a smaller angle in highway scenarios to avoid recommending side and rear stations, while in low-speed urban scenarios, the angle is appropriately widened to improve candidate coverage. Then, the agent generates a road network corridor along the current main navigation path. The width of the road network corridor is dynamically adjusted based on the future travel distance of several kilometers or minutes along the main path, combined with road grade, intersection density, SoC margin, and destination distance. Only stations that simultaneously meet the conditions of forward projection, road network accessibility, and acceptable yaw are included in the subsequent scoring.

[0029] For each candidate station, the tool's orchestration module calls a third-party map route planning interface to obtain the route from the current location to that station, as well as the route back to the main path or continuing to the destination from that station. The data returned by the map interface may include estimated travel time, estimated distance, turning instructions, road direction, toll information, and congestion level. If the route instructions include a U-turn, or if the route's projection on the main navigation path falls after the vehicle's current location, or if the yaw time exceeds the user-constrained threshold, the station is excluded.

[0030] The tool calls the orchestration module to obtain the candidate site's electricity price schedule, service fee, charging pile power, number of available charging guns, number of idle charging piles, number of vehicles in queue, operator, site facilities, data update time, and data source. For data that exceeds the preset refresh time, the agent can mark it as low-confidence data and add a data confidence penalty to the comprehensive cost model, instead of directly treating it as real-time accurate data.

[0031] The spatiotemporal prediction module calculates the estimated arrival time ETA_i based on current road conditions and candidate routes, and calculates the estimated waiting time W_i based on the number of vehicles in the queue, available charging stations, charging power, and historical turnaround speed. The estimated start time of charging T_i is equal to the sum of ETA_i and W_i. The estimated replenishment energy Q_i can be determined by the target replenishment SoC, the estimated arrival SoC, the battery capacity, and the charging efficiency. For example, when the target replenishment SoC is 80%, the estimated arrival SoC is 25%, the battery capacity is 70kWh, and the charging efficiency is 0.95, the estimated replenishment energy is approximately 70×(0.80-0.25) / 0.95.

[0032] When the expected charging process spans multiple price periods, the cost is no longer calculated using a single current electricity price, but rather accumulated segment by segment based on the expected amount of electricity charged within each price period. If a candidate site is in peak price for the first 20 minutes after the expected start of charging and in flat price for the next 30 minutes, the cost for the corresponding amount of electricity is calculated using both peak and flat prices, and a service fee is added to obtain an expected energy replenishment expenditure that more closely approximates the actual payment result.

[0033] The comprehensive cost assessment module can adopt a deterministic model in the following form: Ci = Qi×Pi(Ti) + λt×ΔTi + λw×Wi + λr×Ri + λs×Si + λp×Ui. Where Ci is the comprehensive cost of the i-th candidate station, Qi×Pi(Ti) is the estimated energy cost, ΔTi is the increased yaw time relative to the original main navigation path, Wi is the estimated waiting time, Ri is the data reliability penalty, Si is the safety margin penalty, Ui is the user preference violation penalty, and λt, λw, λr, λs, and λp are weights adjusted according to user settings and decision-making modes.

[0034] The SoC state machine is used to avoid unreasonable recommendations from a single strategy under different power conditions. In the economy-first state, the system prioritizes the lowest overall cost. In the safety-first state, the system first ensures that the site is within the remaining range and that there is an available charging station, then prioritizes sites based on the shortest estimated arrival time, the largest SoC margin at the destination, or the highest probability of successful charging. In the balance-first state, the system increases the weight of safety margin and yaw time, while retaining a safe backup site. The state machine sets different entry and exit thresholds; for example, the SoC threshold for entering the safety-first state is lower than the SoC threshold for exiting the safety-first state, to avoid repeated switching within critical intervals.

[0035] Example 2

[0036] In the natural language interaction implementation, the user inputs "find a fast-charging station that's conveniently located ahead, doesn't require a U-turn, and preferably has a coffee shop nearby." The agent's language understanding module parses this statement into structured constraints: the direction constraint is forward, the maximum yaw time is a preset value or the user's historical preference value, the U-turn constraint is prohibited, the minimum charging power is a fast-charging threshold, and the facility preference includes coffee shops. Subsequently, the deterministic optimization module performs candidate station filtering and scoring based on the above constraints. The language model can be used to explain the reasons for the recommendation, but it does not replace the deterministic scoring results.

[0037] Example 3

[0038] In the high-speed driving embodiment, the vehicle travels along the main highway at a higher speed, with a System-on-Chip (SoC) of 30%. The system dynamically narrows the fan-shaped angle and generates road network corridors near service areas or exits along the main highway. If the nearest station A requires exiting the highway and turning back, while the slightly farther station B is located in a service area ahead with available supercharging stations, the system will exclude station A or increase its yaw penalty, and actively recommend station B, explaining its route proximity, estimated cost, and estimated arrival time.

[0039] Example 4

[0040] In the urban peak-shaving implementation, the vehicle is currently in peak pricing time, but is expected to enter off-peak pricing time in 15 minutes. The system predicts that the current SoC is sufficient to support the vehicle to continue driving to the nearest accessible station, and determines that the station is expected to start charging when it falls within the off-peak pricing time. If the time cost of waiting or continuing to drive is lower than the energy cost saved, the system can recommend continuing to drive to that station; if the SoC enters a safety-priority state, the system will no longer implement the peak-shaving waiting strategy, but will instead prioritize recommending the nearest accessible and available station.

[0041] The system of this invention can be deployed on an in-vehicle computing unit, a cloud server, a navigation platform plugin, or a vehicle manufacturer's service platform. Each module can be implemented in software, hardware, or a combination of both. When the computer program is executed by a processor, it can perform the steps described in the claims of this invention.

Claims

1. A method for proactive decision-making optimization of a vehicle refueling forward road network based on intelligent agent tool invocation orchestration, characterized in that, Executed by intelligent agents deployed on the vehicle, cloud, or navigation platform, including: The vehicle status, current navigation main path, and user power replenishment constraints of the target vehicle are obtained. The vehicle status includes at least the current location, driving direction, vehicle speed, remaining battery SoC, and vehicle energy consumption model. When the active power replenishment trigger condition is met, a forward road network candidate set is generated based on the driving direction and the current main navigation path. The forward road network candidate set is jointly defined by the dynamic fan-shaped initial screening area and the road network corridor extending along the main navigation path. Call the map route planning interface and charging network interface for the charging stations in the candidate set to obtain the reachable path, yaw time, yaw mileage, U-turn or back sign, estimated arrival time, price time table, service fee, charging power, available charging piles and queuing status of each candidate station. The estimated energy replenishment and estimated energy replenishment expenditure for each candidate site are determined based on the estimated arrival time, the estimated start time of charging, the vehicle's arrival SoC, and the target energy replenishment SoC. The decision-making mode of prioritizing economy, balance, or security is determined based on the SoC state machine with hysteresis. Under the decision-making mode, a deterministic comprehensive cost model is used to filter, score, and rank candidate stations to generate the optimal refueling station, alternative refueling stations, and corresponding navigation instructions. The optimal charging station, the reason for recommendation, the estimated cost, the estimated yaw time, and the charging strategy are proactively pushed to the user terminal or vehicle system.

2. The method according to claim 1, characterized in that, The generation of the forward road network candidate set includes: A dynamic sector-shaped initial screening area is established with the vehicle's current position as the vertex and the driving direction vector as the central axis. The fan angle, fan radius, and road network corridor width are dynamically adjusted based on vehicle speed, road grade, main navigation path curvature, and current SoC. The refueling stations located within the dynamic sector initial screening area and whose map projection points are located in front of the current vehicle are selected as initial candidate stations. The initial candidate stations are then further filtered based on the road topology of the current main navigation path.

3. The method according to claim 1 or 2, characterized in that, The secondary screening includes: Call the map route planning interface to calculate the path from the current location to the candidate stations and the path from the candidate stations back to the main navigation path; If the path contains a U-turn instruction, reverse entry, reverse reversal along the main path exceeding a preset distance, yaw time or yaw mileage relative to the original navigation main path exceeding a preset threshold, or the expected arrival SoC is lower than the safety margin threshold, then the corresponding candidate station will be excluded.

4. The method according to claim 1, characterized in that, The intelligent agent includes a tool invocation orchestration module, which parses the user's natural language charging intention into structured constraint parameters. The structured constraint parameters include at least operator preference, maximum acceptable yaw time, minimum charging power, minimum number of available charging piles, target charging SoC, time value coefficient, and supporting facility preference. Based on the structured constraint parameters, the system selectively calls the map path planning interface, charging price interface, charging pile status interface, and queuing status interface, so that the natural language understanding results are converted into executable site filtering conditions and cost model parameters.

5. The method according to claim 1, characterized in that, The determination of the estimated energy replenishment expenditure includes: The estimated arrival time is calculated based on current road conditions, vehicle speed, and candidate station routes; The estimated waiting time is calculated based on the queuing status of candidate sites and available charging positions, and the estimated start time of charging is determined by the estimated arrival time and the estimated waiting time. When the expected charging process spans more than two electricity price periods, the expected charging amount in each electricity price period is multiplied by the corresponding electricity price and service fee, and then summed up in segments to obtain the expected energy replenishment expenditure for the candidate site.

6. The method according to claim 1, characterized in that, The deterministic comprehensive cost model calculates the comprehensive cost Ci for the i-th candidate station. The comprehensive cost Ci includes at least the energy cost item, yaw time item, queuing time item, data reliability penalty item, safety margin penalty item, and user preference penalty item. The energy cost item is determined by the electricity price and service fee corresponding to the expected energy replenishment and expected start charging time. The yaw time item is determined by the increased travel time relative to the original navigation main path and the user time value coefficient. The data reliability penalty item is determined by the update time, source reliability, and missing degree of charging price data or station status data. The agent outputs the station ranking under the economic priority or balance priority mode according to the comprehensive cost from small to large.

7. The method according to claim 1, characterized in that, The SoC state machine with hysteresis includes an economy priority state, a balance priority state, and a safety priority state. When the SoC is above a first entry threshold, it enters the economy priority state. When the SoC is below a second entry threshold or the remaining range is predicted to be insufficient to cover the reachable charging station ahead and a safety margin is reserved, it enters the safety priority state. When the SoC is between the first entry threshold and the second entry threshold, it enters the balance priority state. Each state has a different exit threshold to avoid frequent switching of decision modes caused by the vehicle SoC within the critical range.

8. The method according to claim 7, characterized in that, Under the safety-first state, the agent first filters out candidate stations within the remaining range coverage area, with a path topology reachable, an estimated SoC at the destination with sufficient safety margin, and available charging piles. Then, it prioritizes stations based on the shortest estimated arrival time, the largest SoC margin at the destination, or the highest probability of successful charging. Under the economy-first state, the agent prioritizes stations based on the lowest overall cost. Under the balance-first state, the agent increases the weight of the safety margin penalty and yaw time terms, and retains at least one safe alternative station.

9. A proactive decision-making and optimization system for vehicle refueling forward road network based on intelligent agent tool invocation orchestration, characterized in that, include: The state awareness module is used to obtain the target vehicle's position, driving direction, speed, SoC, vehicle energy consumption model, and current main navigation path; The forward road network candidate generation module is used to generate a forward road network candidate set jointly defined by the dynamic sector initial screening area and the main navigation path road network corridor; The tool calls the orchestration module to parse the user's natural language intent into structured constraint parameters and calls the map route planning interface, charging network interface, pricing interface, and charging pile status interface. The spatiotemporal prediction module is used to calculate the expected arrival time, expected waiting time, expected start time of charging, expected energy replenishment, and SoC of the candidate station; The state machine module is used to switch between economy priority, balance priority and safety priority based on SoC and predicted battery life; The comprehensive cost assessment module is used to filter and sort candidate sites based on energy costs, yaw time, queuing time, data reliability, safety margin, and user preferences. The interactive execution module is used to output the optimal charging station, alternative stations, navigation instructions, and charging strategies to the user terminal or vehicle system.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the vehicle refueling forward road network active decision optimization method based on intelligent agent tool invocation orchestration as described in any one of claims 1 to 8.

Images