Service sequence optimization method and device for electric vehicle charging station and medium

Abstract

The invention relates to the technical field of electric vehicle charging stations, in particular to a service sequence optimization method and device for an electric vehicle charging station and a medium, and the method comprises the steps: constructing a multi-dimensional database according to time-of-use electricity price, user data and station data of the electric vehicle charging station, based on the multi-dimensional database, constructing an initial dynamic pricing model taking total revenue maximization as a target function; the station data comprises a geographic position, the total capacity of a charging station, the number of queuing vehicles, the current service duration, the real-time power load and the operation cost; constructing a simulation environment containing an electric vehicle intelligent agent and a charging station intelligent agent to carry out reinforcement learning on the initial dynamic pricing model so as to obtain a trained dynamic pricing model; and inputting the time-of-use electricity price, the user data and the site data to the dynamic pricing model, and outputting a dynamic service charge sequence by the dynamic pricing model. The technical problem that the income regulation and control capability of an existing charging station fixed time period pricing mode is weak is solved, and the maximization of the total operation income of the charging station is realized.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging station technology, and specifically to a service sequence optimization method, equipment, and medium for electric vehicle charging stations. Background Technology

[0002] With the rapid growth of electric vehicle ownership, charging station operations are facing increasingly fierce market competition and revenue pressure. The fees charged to users by charging stations consist of two parts: the grid electricity price and the charging station service fee. While the grid electricity price is not adjustable, the service fee becomes a key lever for charging station operators to control revenue. Currently, the mainstream service fee pricing model in the industry is still fixed-time pricing, which sets a uniform service fee standard according to coarse-grained time periods such as weekdays / holidays and peak / off-peak periods. This static pricing method cannot accurately adapt to the real-time changing market supply and demand relationship. It often results in problems such as charging piles operating beyond capacity during peak hours due to excessively low pricing but insufficient revenue, and charging piles being idle and wasted during off-peak hours due to excessively high pricing.

[0003] Factors such as the prices of competitors around the charging station, the real-time vehicle density in the area, the station's own load rate, and the constantly changing time-of-use electricity prices create a complex decision-making environment. Existing methods cannot integrate and process the data foundation and core algorithms of these multi-dimensional factors in real time, making it impossible to establish a mathematical model that accurately quantifies the relationship between price and demand. This makes it difficult to calculate the theoretically optimal service fee price, which in turn affects the final charging volume and revenue. Summary of the Invention

[0004] The purpose of this invention is to provide a service sequence optimization method, equipment, and medium for electric vehicle charging stations based on time-of-use pricing and dual-agent reinforcement learning simulation, so as to overcome the technical problem of weak revenue regulation capability of existing fixed-time pricing methods for charging stations and maximize the total operating revenue of charging stations.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: The present invention provides a service sequence optimization method for electric vehicle charging stations, comprising: constructing a multi-dimensional database based on time-of-use electricity prices, user data, and site data of electric vehicle charging stations, and constructing an initial dynamic pricing model based on the multi-dimensional database with the objective function of maximizing total revenue; the site data includes geographical location, total capacity of charging stations, number of vehicles in queue, current service duration, real-time power load, and operating costs; constructing a simulation environment containing electric vehicle intelligent agents and charging station intelligent agents to perform reinforcement learning on the initial dynamic pricing model to obtain a trained dynamic pricing model; inputting time-of-use electricity prices, user data, and site data into the dynamic pricing model, and the dynamic pricing model outputs a dynamic service fee sequence.

[0006] Optionally, a first reward function is set for the electric vehicle intelligent agent. The first reward function is obtained by weighted summation of charging completion reward, economic reward, time efficiency reward and charging abandonment penalty; the charging abandonment penalty includes penalties for abandoning charging due to the target charging station being full, charging price being too high, or user waiting time exceeding the limit.

[0007] Optionally, a second reward function is set for the charging station's intelligent agent. The second reward function is obtained by weighted summation of economic benefit reward, utilization rate reward, and comprehensive penalty. The comprehensive penalty includes penalties for exceeding load limits, excessively high pricing, and excessive price gap with competitors.

[0008] Optionally, the state attributes of the electric vehicle intelligent agent include the remaining battery power of the electric vehicle, the charging power type, and the user's personalized preferences; the user's personalized preferences include price sensitivity, range anxiety, distance preference, and congestion tolerance; the state attributes of the charging station intelligent agent include the total capacity of the charging station, operating costs, and the charging station's dynamic attributes; the charging station's dynamic attributes include real-time power load, number of vehicles in queue, current service duration, instantaneous load rate within the station, historical demand sequence, and future reservation sequence.

[0009] Optionally, it also includes: a dynamic pricing model that receives pricing query commands from users for different time periods and outputs explanatory analyses for the corresponding time periods based on the pricing query commands, including natural language reports and visual attribution graphs.

[0010] Optionally, the objective function of the dynamic pricing model is as follows: In the formula, This represents the total revenue of electric vehicle charging stations; This indicates the service fee for a specific period of time. This indicates the current electricity price on the grid. This represents the overall marginal cost of providing a single charging service; A function representing the predicted number of charging cycles; These represent the total user cost of this website and the total user cost of its competitors, respectively. Indicates the number of vehicles in the area; This indicates the current status of the charging station.

[0011] Optionally, the explanatory analysis for different time periods can be output using the following formula: In the formula, Indicates the time period The state vector includes the number of vehicles in the area, competitor prices, station load, and weather factors. This indicates that the dynamic pricing model is for a specific time period. Pricing of service fees for output; This represents a pre-trained dynamic pricing model; It is an interpretable analytical function used for dynamic pricing models. In state Make a decision vector The model logic is explained and analyzed.

[0012] The present invention also provides a terminal device, including a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described above.

[0013] The present invention also provides a storage medium storing a computer program, which is invoked and executed by a computer to implement the method described above.

[0014] The beneficial effects of this invention are as follows: 1. By constructing a dynamic pricing model with the objective function of maximizing total revenue, multi-agent simulation verification, and an interpretable decision support process, this invention relies on a high-efficiency multi-agent simulation environment for reinforcement learning. In a market environment with fluctuating time-of-use electricity prices and changing user demands, it can quickly locate the service fee pricing point, achieve stable improvement in the operating revenue of electric vehicle charging stations, and optimize the service sequence of electric vehicle charging stations.

[0015] 2. This invention provides a multi-agent reinforcement learning simulation environment that can realistically simulate the complex charging decision-making process of electric vehicle users. It comprehensively considers multiple user decision-making factors such as price sensitivity, range anxiety, distance preference, and congestion tolerance, as well as the state attributes of the charging station agent, including the total capacity, operating cost, and dynamic attributes of the charging station. This provides a safe and low-cost simulation verification platform for optimizing charging station pricing strategies, avoiding the risk of strategy trial and error in real operation scenarios and overcoming the technical problem of blind strategy verification in existing dynamic pricing methods for charging stations.

[0016] 3. This invention receives pricing query commands from users for different time periods and outputs explanations and analyses corresponding to different time periods based on the pricing query commands. It can transform the dynamic service fee sequence output by the dynamic pricing model into easily understandable natural language reports and visual attribution diagrams, analyze and explain the decision-making logic of pricing suggestions for each time period, and overcome the technical problem of opaque decision-making logic in existing dynamic pricing methods for charging stations. Attached Figure Description

[0017] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0018] Figure 1 This is a flowchart of a service sequence optimization method for electric vehicle charging stations; Figure 2 This is a schematic diagram of the game simulation process between the intelligent agent of electric vehicles and the intelligent agent of charging stations. Figure 3 This is a schematic diagram illustrating the generation of pricing suggestions for electric vehicle charging stations based on time-of-use pricing. Figure 4 This is a schematic diagram of the interface of a smart pricing management visualization platform for electric vehicle charging stations; Figure 5 This is a schematic diagram of the interface for a real-time operation status monitoring dashboard for electric vehicle charging stations. Figure 6 This is a schematic diagram of an interface for analyzing dynamic adjustments to electricity prices at electric vehicle charging stations. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] As one implementation method, such as Figure 1 As shown, this embodiment provides a service sequence optimization method for electric vehicle charging stations. Based on collected time-of-use electricity prices, user data, and electric vehicle charging station site data, a knowledge-complete database is constructed. Then, based on the relationship between each data point in the database and the factors affecting service prices or overall prices, a specific dynamic pricing model is built. After the model is established, a simulation environment is constructed to verify the model's accuracy. A multi-agent reinforcement learning method is introduced, and a simulation environment including electric vehicle agents and charging station agents is designed to reproduce specific scenarios for verification. Finally, an interpretable analysis function is used to analyze the surrounding environment of the charging station and provide a specific price adjustment reasoning process. Specifically, this includes: S1. Construct a multidimensional database based on time-of-use electricity pricing, user data, and electric vehicle charging station site data, and build an initial dynamic pricing model based on the multidimensional database with the objective function of maximizing total revenue; site data includes geographical location, total charging station capacity, number of vehicles in queue, current service duration, real-time power load, and operating costs.

[0021] Specifically, a multi-dimensional database and dynamic pricing model are constructed. By integrating and building a dynamically updated multi-dimensional database covering factors such as surrounding competitor prices, real-time regional vehicle density, grid time-of-use pricing, site capacity utilization, vehicle charging characteristics, and geographical location, a dynamic pricing model with the goal of maximizing revenue is established.

[0022] Through the data acquisition and processing subsystem, raw information is obtained in real time from multiple heterogeneous data sources, including but not limited to: time-of-use electricity prices published by the national or regional power grid on the power grid side, with an update frequency of no less than every 30 minutes; real-time service fees of other charging stations within a 3-kilometer radius obtained from competitors via third-party platform APIs; macro-level traffic conditions such as regional vehicle density and road congestion index obtained from the traffic big data platform on the transportation side; operational data such as capacity utilization, queue length, current load rate, and historical reservation orders of target charging stations collected by IoT devices on the site side; and vehicle attribute data such as battery capacity, SOC, and fast / slow charging preferences obtained through vehicle terminal or APP authorization on the user side after anonymization.

[0023] The method for constructing the multidimensional database is as follows: raw data is acquired in real time from public services, IoT sensors, and internal databases, and then cleaned, normalized, and spatiotemporally aligned before being stored in the multidimensional database. The database time granularity is set to 30 minutes to ensure the real-time nature and accuracy of the data. Based on this database, a dynamic pricing model is constructed, with the objective function being the maximization of total revenue for charging stations.

[0024] S2. Construct a simulation environment containing both electric vehicle agents and charging station agents to perform reinforcement learning on the initial dynamic pricing model. Set the reward function of the agents to maximize the total revenue of the electric vehicle charging station to obtain the trained dynamic pricing model. Construct a multi-agent reinforcement learning simulation environment to provide a realistic simulation environment for the pricing strategy by simulating the real-world choices of a large number of heterogeneous users, enabling pre-validation and post-event review. The multi-agent training environment is a Markov game process simulating the real world, and its core elements are defined through the following framework: state space In a given context, the state space is the collection of all environmental information that an agent can observe at any given moment. For example, the state of an electric vehicle agent includes remaining battery power (SoC), distance to the nearest charging station, and the number of people in the queue. Its state This includes its own attributes and local environmental information. For charging station intelligent agents... Its state This includes its own dynamic attributes and pricing information of its competitors.

[0025] Action space The action space is the set of all possible operations that an agent can perform. For example, the action of an agent at a charging station is to set the service fee for the next 30 minutes (discretized in steps of 0.1 yuan / kWh). Electric vehicle agent action The decision to choose a charging station or continue driving. Charging station intelligent agent. action The service fee price set for it , is a continuous or discrete decision variable.

[0026] State transition function : Describes the environment in joint actions Next, from the current state Transition to the next state The probability, i.e. This function encapsulates the complex dynamics of traffic flow simulation, user arrival and departure, and the charging process.

[0027] A multi-agent reinforcement learning algorithm with centralized training and distributed execution is employed. During training, each agent... Learn a strategy This strategy is based on local observations. Make an action The probability distribution. The policy parameters are determined by maximizing its expected cumulative discount reward. To perform the update, among which This serves as a discount factor. By introducing course learning and a distributed parallel architecture, the environmental complexity and the number of agents are gradually increased to ensure the efficiency of the training process and the stability of the policy.

[0028] The reward function of the electric vehicle intelligent agent is obtained by weighted summing of charging completion reward, economic reward, time efficiency reward, and charging abandonment penalty. The state attributes of the electric vehicle intelligent agent include the remaining battery power, charging power type, and user personalized preferences; user personalized preferences include price sensitivity, range anxiety, distance preference, and congestion tolerance.

[0029] The first reward function for the electric vehicle agent is as follows: In the formula, This indicates a reward upon completion of charging; This indicates an economic incentive, which is negatively correlated with the total charging cost; This indicates a time efficiency bonus, which is negatively correlated with the total time spent traveling and queuing. This indicates a penalty for abandoning charging; penalties for abandoning charging include those caused by the target charging station being full, charging prices being too high, or the user waiting time exceeding the limit. These represent the weighting coefficients for the charging completion reward, economic reward, time efficiency reward, and penalty for abandoning charging, respectively. This indicates the charging station service fee; This indicates the electricity price on the power grid.

[0030] The reward function of the charging station agent is obtained by weighted summation of economic benefit rewards, utilization rate rewards, and comprehensive penalties. The state attributes of the charging station agent include the total capacity of the charging station, operating costs, and dynamic attributes of the charging station; the dynamic attributes of the charging station include real-time power load, number of vehicles in queue, current service duration, instantaneous load rate within the station, historical demand sequence, and future reservation sequence.

[0031] The second reward function for the charging station agent is as follows: In the formula, This indicates an economic benefit incentive, which is positively correlated with service fee revenue. This indicates a load utilization bonus, encouraging the load to be maintained at a healthy level and avoiding idling or overloading; This indicates a comprehensive penalty, which includes penalties for exceeding load limits, excessively high pricing, and excessive price differences with competitors. These represent the weighting coefficients for economic benefit rewards, utilization rate rewards, and comprehensive penalties, respectively, used to balance short-term gains with long-term operational efficiency.

[0032] As one implementation method, refer to Figure 2 In a virtual environment, multiple electric vehicles, such as electric vehicles A and B, and multiple charging stations, such as charging stations A and B, are simulated to determine their interactive selection relationships. Through a two-way game between the intelligent agents of electric vehicles and the intelligent agents of charging stations, an appropriate charging station service fee is ultimately output.

[0033] S3. Input time-of-use electricity pricing, user data, and site data into the dynamic pricing model. The dynamic pricing model outputs a dynamic service fee sequence. Set the objective function of the dynamic pricing model as follows: In the formula, This represents the total revenue of electric vehicle charging stations; This indicates the service fee for a certain period of time, in yuan / degree. This indicates the current electricity price, in yuan / kWh. This represents the overall marginal cost of providing a single charging service, expressed in yuan per kilowatt-hour. A function representing the predicted number of charging cycles; These represent the total user cost of this website and the total user cost of its competitors, respectively. Indicates the number of vehicles in the area; This indicates the current status of the charging station.

[0034] The rolling time-domain optimization method is adopted, with the next 24 hours as the planning window. The optimal service fee sequence is solved every 30 minutes. The solution algorithm can be mixed integer linear programming (MILP). After the calculation is completed, the result is returned. Sequence and sequence.

[0035] As one implementation method, refer to Figure 3 The system collects site data and user data and stores them in a database. Site data includes charging pile load and operational status, while user data includes charging preferences and behavioral habits. Based on the database, a simulation system with multiple agents is built. Then, the agents simulate actual charging scenarios and finally generate corresponding dynamic pricing suggestions.

[0036] This embodiment introduces a multi-agent reinforcement learning simulation environment capable of simulating user interaction with charging stations. This allows for strategy validation and optimization in a virtual world, reducing trial-and-error costs in actual operation. Price adjustments directly alter user behavior, impacting final charging volume and revenue; such complex market feedback is difficult to predict within traditional decision-making frameworks. The lack of an efficient simulation environment to model user charging choices prevents operators from assessing the potential effects of strategies before implementation or accurately reviewing their effectiveness afterward, leading to significant uncertainty and risk in the decision-making process. Using this simulation model can improve strategy accuracy.

[0037] Even data-driven and simulation-validated pricing recommendations will be difficult to adopt if their underlying decision-making logic is a black box to human operators due to a lack of interpretability. Therefore, transforming complex model outputs into intuitive visualizations and clear natural language explanations, revealing the degree of influence of different factors on pricing decisions, is crucial for building a bridge for human-machine collaboration and enhancing the system's practical value. Providing a real-time decision support interface with interpretable analytical capabilities is a key step in ensuring the successful implementation of intelligent pricing strategies.

[0038] S4 also includes: The dynamic pricing model receives pricing query commands from users for different time periods, and outputs explanatory analyses for each time period based on the pricing query commands. The explanatory analyses include natural language reports and visual attribution plots. The explanatory analyses for different time periods are output using the following formula: In the formula, Indicates the time period The state vector includes the number of vehicles in the area, competitor prices, station load, and weather factors. This indicates that the dynamic pricing model is for a specific time period. Pricing of service fees for output; This represents a pre-trained dynamic pricing model; It is an interpretable analytical function used for dynamic pricing models. In state Make a decision vector The model logic is explained and analyzed. The implementation, based on attention-based natural language generation models or post-attribution algorithms, involves calculating state vectors. The influence of various dimensions of characteristics on decision-making The system assesses the contribution of key influencing factors, outputting an interpretable analysis report that includes key influencing factors, the direction and extent of influence, presented through a visual interface. The results are presented in bar charts and text format, and can be exported as a PDF report. Real-time decision support is provided through the visual interface, displaying pricing recommendations, simulation results, and the basis for price adjustments, enabling operators to intuitively understand the pricing logic and adopt the recommendations.

[0039] S41. Provides a geographic information visualization interface to dynamically display the location of charging stations, heat maps of capacity utilization, traffic flow, and intelligent agent simulation process.

[0040] S42 provides a pricing strategy dashboard, displaying historical and future electricity prices, service fees, total costs, and projected profits in line graph format. An interactive cursor allows viewing detailed data for any time period. It dynamically displays the location of each charging station, current utilization rate, traffic flow, and the movement trajectory of the simulated agent. A line graph shows the recommended service fee, grid electricity price, total user costs, and projected revenue curves for the next 24 hours, with mouse hover support for viewing details for any time period. When the operator clicks on a pricing suggestion for a target time period, an interpretability function is invoked.

[0041] As one implementation method, refer to Figure 4 The system visualizes the distribution and status of charging stations within a region through a map, while integrating data such as station parameters, competitor prices, and user preferences to generate dynamic pricing suggestions. It also provides price trend charts to intuitively present pricing changes over different time periods, assisting operators in real-time monitoring and decision-making. (Reference) Figure 5 Focusing on core operational data of the building's charging stations, it displays real-time indicators such as the number of remaining charging stations, revenue, and current load. It also includes future time-to-time status trend predictions and charging reservation statistics, helping operators intuitively understand station resource utilization, revenue progress, and load status. (Reference) Figure 6 The system uses electricity price trend charts to compare the current electricity price with the market average price. Combined with information such as charging usage during peak hours and pricing of competing products in the surrounding area, it outputs specific suggestions for raising the electricity price (from 0.85 yuan / kWh to 1.05 yuan / kWh), while explaining the basis for the price adjustment and the expected benefits, to assist operators in making decisions.

[0042] This invention establishes a dynamic pricing model with the goal of maximizing revenue by integrating and constructing a dynamically updated multidimensional database covering factors such as competitor prices, regional vehicle density, time-of-use pricing, station capacity utilization, vehicle attributes, and geographical location. Secondly, to achieve high-fidelity simulation of user charging behavior, a multi-agent reinforcement learning framework is employed. Deep reinforcement learning algorithms are used to simulate the charging station selection decisions of heterogeneous users. A macro-level strategy optimization model is used to generate and validate time-of-use service fee pricing strategies, completing both pre-simulation and post-analysis of the strategies. Finally, a visual interface provides real-time decision support, enabling operators to intuitively understand the basis for price adjustments and adopt recommendations.

[0043] This invention effectively overcomes the market response lag problem of static pricing strategies by constructing a dynamic pricing model for service fees and using intelligent agent simulation. This significantly improves the operating revenue, resource utilization efficiency, and the scientific rigor and credibility of the decision-making process for charging stations, ultimately optimizing charging station profitability. This invention not only achieves data-driven dynamic pricing but also reveals the inherent logic of the strategy through behavioral simulation and interpretable analysis, effectively overcoming the market response lag problem of static pricing strategies and thus significantly improving the operating revenue, resource utilization efficiency, and the scientific rigor and credibility of the decision-making process for charging stations.

[0044] As one implementation method, the electric vehicle's intelligent agent observes the real-time prices, queue lengths, distances, and its own battery level of the five nearest charging stations. The charging station's intelligent agent observes the global vehicle distribution, competitor prices, and its own load. The electric vehicle selects a target station or continues driving; the charging station sets a service fee for the next 30 minutes in discrete steps of 0.1 yuan / kWh. The MAPPO algorithm is used for centralized training and distributed execution. The training process incorporates course learning, gradually transitioning from simple scenarios with few vehicles to complex urban road networks. The charging station's intelligent agent runs for 24 hours in a simulation environment, and key indicators such as total revenue, user satisfaction, load balancing, and churn rate are statistically analyzed and compared with the baseline strategy to complete the pre-simulation evaluation. Historical operational data can also be replayed for post-event review and analysis. This invention realizes a complete chain of data-driven, behavioral simulation, and reliable decision-making, significantly improving the scientific rigor, adaptability, and operability of dynamic pricing for charging stations.

[0045] As one implementation method, this scheme can use traditional machine learning models such as gradient boosting trees to directly predict user demand or revenue under different service fees, thus replacing dynamic pricing models. At the same time, simulation systems based on simple probability rules such as Logit models can replace complex multi-agent reinforcement learning environments, and tools such as SHAP can be used to provide ex-post interpretation of tree models, which can replace specially designed interpretability analysis functions.

[0046] As one implementation method, the entire market is abstracted into a black box response function, and the charging station is allowed to act as a single intelligent agent. The optimal service fee is directly output through the DQN deep reinforcement learning algorithm, thereby eliminating the need for independent mathematical models and multi-agent simulation. The advantage of this embodiment is that the architecture is simple and the training efficiency is relatively high.

[0047] As one implementation method, dynamic pricing models can be abandoned, and a rule base developed by domain experts can directly replace dynamic pricing models and simulation environments. The rules themselves are highly transparent and naturally interpretable. Their advantage is that the decision-making logic is extremely clear and easy to understand, maintain and audit.

[0048] As one implementation, this embodiment also provides a terminal device, including a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the method described above.

[0049] As one implementation method, this embodiment also provides a storage medium storing a computer program, which is called and executed by a computer to implement the method described above.

[0050] Compared with the prior art, the present invention has the following beneficial effects based on the above embodiments: 1. This invention constructs a dynamic pricing model with the objective function of maximizing total revenue, verifies it through multi-agent simulation, and implements an interpretable decision support process. Relying on an efficient multi-agent simulation environment for reinforcement learning, it can quickly locate the service fee pricing point in a market environment with fluctuating time-of-use electricity prices and changing user demands, thereby achieving a stable increase in the operating revenue of electric vehicle charging stations and optimizing the service sequence of electric vehicle charging stations.

[0051] 2. This invention provides a multi-agent reinforcement learning simulation environment that can realistically simulate the complex charging decision-making process of electric vehicle users. It comprehensively considers multiple user decision-making factors such as price sensitivity, range anxiety, distance preference, and congestion tolerance, as well as the state attributes of the charging station agent, including the total capacity, operating cost, and dynamic attributes of the charging station. This provides a safe and low-cost simulation verification platform for optimizing charging station pricing strategies, avoiding the risk of strategy trial and error in real operation scenarios and overcoming the technical problem of blind strategy verification in existing dynamic pricing methods for charging stations.

[0052] 3. This invention receives pricing query commands from users for different time periods and outputs explanations and analyses corresponding to different time periods based on the pricing query commands. It can transform the dynamic service fee sequence output by the dynamic pricing model into easily understandable natural language reports and visual attribution diagrams, clearly explaining the decision-making logic of pricing recommendations for each time period. At the same time, it quantifies the contribution of different influencing factors such as grid electricity price, site load, and competitor prices to the pricing results, overcoming the technical problem of opaque decision-making logic in existing dynamic pricing methods for charging stations.

[0053] The specific embodiments described above are preferred embodiments of the service sequence optimization method, equipment and medium for electric vehicle charging stations of this application, and are not intended to limit the specific implementation scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of this application are within the protection scope of this application.

Claims

1. A service sequence optimization method for electric vehicle charging stations, characterized in that, include: A multidimensional database is constructed based on time-of-use electricity pricing, user data, and electric vehicle charging station site data. An initial dynamic pricing model with the objective function of maximizing total revenue is then built based on the multidimensional database. The site data includes geographical location, total charging station capacity, number of vehicles in queue, current service duration, real-time power load, and operating costs. A simulation environment containing electric vehicle intelligent agents and charging station intelligent agents is constructed to perform reinforcement learning on the initial dynamic pricing model in order to obtain the trained dynamic pricing model. Input time-of-use electricity prices, user data, and site data into the dynamic pricing model, and the dynamic pricing model outputs a dynamic service fee sequence.

2. The service sequence optimization method for electric vehicle charging stations according to claim 1, characterized in that, The first reward function of the electric vehicle intelligent agent is set. The first reward function is obtained by weighted summation of charging completion reward, economic reward, time efficiency reward and charging abandonment penalty. Charging abandonment penalty includes penalties for abandoning charging due to the target charging station being full, charging price being too high or user waiting time exceeding the limit.

3. The service sequence optimization method for electric vehicle charging stations according to claim 1, characterized in that, The second reward function for the charging station's intelligent agent is set up. This second reward function is obtained by weighted summation of economic benefit reward, utilization rate reward, and comprehensive penalty. The comprehensive penalty includes penalties for exceeding load limits, excessive pricing, and excessive price gap with competitors.

4. The service sequence optimization method for electric vehicle charging stations according to claim 1, characterized in that, The state attributes of an electric vehicle's intelligent agent include the remaining battery power, charging power type, and user personalized preferences; user personalized preferences include price sensitivity, range anxiety, distance preference, and congestion tolerance. The state attributes of the charging station intelligent agent include the total capacity of the charging station, operating costs, and dynamic attributes of the charging station; the dynamic attributes of the charging station include real-time power load, number of vehicles in queue, current service duration, instantaneous load rate in the station, historical demand sequence, and future reservation sequence.

5. The service sequence optimization method for electric vehicle charging stations according to claim 1, characterized in that, Also includes: The dynamic pricing model receives pricing query commands from users for different time periods and outputs explanatory analyses for the corresponding time periods based on the pricing query commands. The explanatory analyses include natural language reports and visual attribution graphs.

6. The service sequence optimization method for electric vehicle charging stations according to claim 1, characterized in that, The objective function of the dynamic pricing model is as follows: In the formula, This represents the total revenue of electric vehicle charging stations; This indicates the service fee for a specific period of time. This indicates the current electricity price on the grid. This represents the overall marginal cost of providing a single charging service; A function representing the predicted number of charging cycles; These represent the total user cost of this website and the total user cost of its competitors, respectively. Indicates the number of vehicles in the area; This indicates the current status of the charging station.

7. The service sequence optimization method for electric vehicle charging stations according to claim 5, characterized in that, The following formula outputs the explanation and analysis for different time periods: In the formula, Indicates the time period The state vector includes the number of vehicles in the area, competitor prices, station load, and weather factors. This indicates that the dynamic pricing model is for a specific time period. Pricing of service fees for output; This represents a pre-trained dynamic pricing model; It is an interpretable analytical function used for dynamic pricing models. In state Make a decision vector The model logic is explained and analyzed.

8. A terminal device, characterized in that, The method includes a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the method as described in any one of claims 1 to 7.

9. A storage medium, characterized in that, The storage medium stores a computer program, which is invoked and executed by a computer to implement the method as described in any one of claims 1 to 7.

Images