Electric vehicle user charging scheduling optimization method based on large language model

Abstract

The invention relates to an electric vehicle user charging scheduling optimization method based on a large language model. The method comprises the following steps: S1, obtaining historical charging data and vehicle operation data; s2, performing charging demand simulation prediction by using a large language model to obtain a charging demand prediction result; s3, adopting a double-layer optimization algorithm based on the historical charging data and the vehicle operation data; s4, pushing the optimal charging scheduling strategy to a user side, obtaining a real-time power consumption behavior of the user side, and if the real-time power consumption behavior deviates from expectation, performing adjustment through price excitation or prompt information; s5, calculating a key index, judging whether the optimization target is completed or not based on the key index, if so, executing S6, and otherwise, returning to S2 after adjusting the large language model or the double-layer optimization algorithm; and S6, historical charging data and vehicle operation data. Compared with the prior art, the method has the advantages of realizing bidirectional interaction, improving the utilization efficiency of electric power resources and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle charging, and more particularly to an optimization method for electric vehicle user charging scheduling based on a large language model. Background Art

[0002] With the popularization of new energy vehicles globally, the number of electric vehicles has increased exponentially. The popularization of electric vehicles helps reduce carbon emissions and dependence on fossil fuels. However, due to the randomness of their charging demands and the uneven regional distribution, the large-scale connection to the power grid poses new load challenges to the power system. This uncertainty in demand has a significant impact on the stability of the power system, especially during peak load periods. Therefore, how to balance the charging demands of electric vehicles and the power supply capacity of the grid through effective demand response technologies has become an important research direction in smart grids.

[0003] Currently, traditional demand response methods have limitations in dealing with the distributed and high-frequency electricity demands of electric vehicles and are difficult to accurately grasp the charging behaviors of each user. For example, patents (CN116562013A, A charging load calculation method based on electric vehicle charging behavior simulation, August 8, 2023) and literature (Fu Kun, Research on Electric Vehicle Behavior Simulation and Charging and Discharging Scheduling Strategies [D]. Power Engineering and Engineering Thermophysics, December 2021). These methods do not consider individual characteristics such as the psychological factors of users, which affects the simulation and prediction of user charging behaviors. With the development of large language model (LLM) technology, the natural language processing and user behavior prediction capabilities based on LLM provide new possibilities for demand response. LLM can simulate the charging behaviors of users based on various data sources (such as charging history records, weather conditions, user preferences, traffic congestion conditions, etc.), thereby better understanding the demand patterns of electric vehicle users and accurately predicting charging demands. This simulation method can help the demand response system more intelligently match personalized needs at the user level, predict key factors such as user charging time, frequency, and required charging amount, and further support aggregators in formulating flexible and efficient response strategies.

[0004] However, most of the existing demand response methods still remain at the single-layer response strategy. Usually, aggregators simply aggregate user demands for unified regulation, without fully considering the uncertainty of user behavior and the complexity of personalized demand response. For example, Patent (CN117374954A, an optimized scheduling method and system for electric vehicle aggregators to participate in demand response, January 9, 2024) and literature (Pang Songling, Research on Pricing Strategy and Revenue Allocation for Electric Vehicle Aggregators to Participate in Demand Response [J]. Advanced Technology of Electrical Engineering and Energy, 2024, 43(07): 41-50). CN117374954A discloses an optimized scheduling model that includes a power purchase cost model for electric vehicle aggregators to participate in demand response, an electric vehicle response rate model corresponding to electric vehicles, an incentive cost model for electric vehicle aggregators to give incentives for electric vehicles to participate in demand response, and a revenue model for electric vehicle aggregators to participate in demand response. However, it does not consider the uncertainty of user charging behavior, which may lead to optimization deviation and inefficiency. User charging behavior is random, such as different charging times, demands, and responses to incentives. If the model assumes deterministic behavior, the power purchase plan may not match the actual demand, affecting the stability of the power grid and increasing the operating risk of aggregators. In addition, the two-layer demand response architecture is not considered, ignoring the game relationship between the grid operator and the aggregator, which may lead to scheduling incoordination, reducing the demand response efficiency. Although the optimized scheduling model of CN117374954A patent covers factors such as power purchase cost, incentive cost, response rate, and revenue, it does not consider the uncertainty of user charging behavior and the two-layer demand response architecture, which may lead to optimization deviation and inefficiency. And user charging behavior is random, such as different charging times, demands, and responses to incentives. If the model assumes deterministic behavior, the power purchase plan may not match the actual demand, affecting the stability of the power grid and increasing the operating risk of aggregators. In addition, the two-layer demand response architecture is not considered, ignoring the game relationship between the grid operator and the aggregator, which may lead to scheduling incoordination, reducing the demand response efficiency, and affecting the stability of the power market. Summary of the Invention

[0005] The purpose of the present invention is to provide an optimized method for scheduling electric vehicle user charging based on a large language model to overcome the problems of scheduling incoordination and reduced demand response efficiency.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An optimized method for scheduling electric vehicle user charging based on a large language model, the method includes the following steps:

[0008] S1. Obtain preprocessed historical charging data and vehicle operation data;

[0009] S2. Based on historical charging data and vehicle operation data, use a large language model to simulate and predict charging demand, and obtain the charging demand prediction result;

[0010] S3. Adopt a two - layer optimization algorithm based on historical charging data and vehicle operation data. The two - layer optimization algorithm includes global optimization and individual optimization. First, at the global optimization level, construct a global optimization model with the goal of minimizing the peak grid load to obtain the optimal electricity price strategy. Then, at the individual optimization level, optimize the charging time of users based on the optimal electricity price strategy to obtain the optimal charging scheduling strategy;

[0011] S4. Push the optimal charging scheduling strategy to the user side, obtain the real - time electricity consumption behavior of the user side. If the real - time electricity consumption behavior deviates from the expectation, adjust it through price incentives or prompt messages;

[0012] S5. Calculate key indicators, and judge whether the optimization goal is completed based on the key indicators. If it is completed, execute S6; otherwise, adjust the large language model or the two - layer optimization algorithm and then return to S2;

[0013] S6. Historical charging data and vehicle operation data.

[0014] Furthermore, the specific steps of using the large language model to simulate and predict charging demand and obtain the charging demand prediction result are as follows:

[0015] Use the Transformer structure to extract the time - series features of historical charging data and vehicle operation data, calculate the charging probability of users in different time periods, and combine with the user portrait to calculate the charging demand prediction result of users in a future period.

[0016] Furthermore, the optimization goal of the global optimization model is to minimize the peak grid load, specifically:

[0017]

[0018] Where: represents the total grid load at time t; T represents the set of all time periods within the optimization cycle.

[0019] Furthermore, the constraint conditions of the global optimization model include charging demand constraints and grid load upper - limit constraints.

[0020] Furthermore, the method for solving the global optimization model is: adopt the gradient - descent method to optimize the electricity price strategy.

[0021] Furthermore, the specific method of using the gradient - descent method to optimize the electricity price strategy is: the electricity price at the (k + 1) - th iteration is:

[0022]

[0023] Among them, represents the electricity price for the next iteration, η is the learning rate, and L is the loss function.

[0024] Furthermore, the loss function is specifically:

[0025]

[0026] Among them, P target is the target load level, and P t represents the current load level.

[0027] Furthermore, the optimal charging scheduling strategy is:

[0028]

[0029] Among them, D t represents the charging demand, and C t represents the electricity price of the optimal electricity price strategy.

[0030] Furthermore, the specific steps of S4 are:

[0031] Monitor the actual charging behavior of users, and the actual charging behavior is specifically the real-time charging data at the user end;

[0032] Based on the deviation degree, evaluate the deviation between the actual behavior of users and the optimal charging scheduling strategy. Among them, the calculation formula of the deviation degree D is:

[0033]

[0034] Among them: D represents the charging behavior deviation degree; T represents the total number of time periods within the optimization period; represents the actual charging power of the user at time t; represents the optimal charging power at time t in the global optimization strategy; w t represents the time weight factor;

[0035] If the deviation exceeds the threshold, execute the adjustment mechanism. The adjustment mechanism includes price incentives and prompt information. Among them, the price incentive is: adjust the electricity price, increase the electricity price during peak hours and decrease the electricity price during off-peak hours to guide users to optimize the charging time;

[0036] The prompt information is: send a notice to the user to remind the best charging period.

[0037] Furthermore, the preprocessing includes denoising and standardization, and the denoising adopts the sliding mean filtering method.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) The present invention designs a two - layer architecture. The first layer is the electric vehicle user layer, which is responsible for collecting and analyzing in real - time information such as users' charging demands, behavior patterns, and their responses to electricity prices. The second layer is the aggregator layer. The aggregator uses the collected user data and market information to formulate power supply strategies. This architecture can effectively coordinate user demands and aggregator supplies, achieve two - way interaction, and improve the utilization efficiency of power resources. In the present invention, the optimization objective of the global optimization model is to minimize the peak value of the power grid load. The aggregator formulates electricity price strategies through global optimization, and users adjust their behaviors based on electricity prices, forming two - way interaction between supply and demand, avoiding power grid overload and improving resource utilization efficiency. At the same time, the two - layer optimization algorithm of the present invention performs two - layer linkage of the optimal charging strategy. Through the optimization order of "global → individual", the aggregator and users form collaborative decisions, avoiding possible conflicts in single - layer optimization and significantly improving the scheduling efficiency.

[0040] The present invention also uses the gradient descent method to optimize the electricity price strategy. The real - time optimization of the electricity price strategy enables the aggregator to flexibly respond to changes in user behavior. Users then adjust their charging plans through electricity price signals, forming a dynamic feedback closed - loop and improving the supply - demand matching efficiency.

[0041] (2) The present invention can also deeply analyze users' historical charging data, market electricity price changes, and user feedback by integrating large language models. The powerful natural language processing ability of the LLM enables it to extract valuable information from complex data and generate accurate predictions of user demands. This process not only improves the intelligence level of demand response but also provides a more scientific decision - making basis for aggregators. Brief Description of the Drawings

[0042] Figure 1 is the two - layer model of electric vehicle users and aggregators of the present invention;

[0043] Figure 2 is the process diagram of electric vehicle user demand simulation;

[0044] Figure 3 is the two - layer decision - making optimization framework diagram. Detailed Embodiment

[0045] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives detailed implementation methods and specific operation processes. However, the protection scope of the present invention is not limited to the following embodiments.

[0046] The present invention proposes an optimization method for electric vehicle user charging scheduling based on a large language model. The method includes the following steps:

[0047] S1. Obtain pre - processed historical charging data and vehicle operation data;

[0048] S2. Based on historical charging data and vehicle operation data, use a large language model to simulate and predict charging demand, and obtain the charging demand prediction result;

[0049] S3. Adopt a two-layer optimization algorithm based on historical charging data and vehicle operation data. The two-layer optimization algorithm includes global optimization and individual optimization. At the global optimization level, construct a global optimization model with the goal of minimizing the peak grid load to obtain the optimal electricity price strategy. At the individual optimization level, optimize the user's charging time to obtain the optimal charging scheduling strategy;

[0050] S4. Push the optimal charging scheduling strategy to the user side, obtain the real-time electricity consumption behavior of the user side. If the real-time electricity consumption behavior deviates from the expectation, adjust it through price incentives or prompt messages;

[0051] S5. Calculate key indicators, and judge whether the optimization goal is completed based on the key indicators. If it is completed, execute S6. Otherwise, adjust the large language model or the two-layer optimization algorithm and then return to S2;

[0052] S6. Historical charging data and vehicle operation data.

[0053] The specific framework of the present invention includes a user layer and an aggregator layer, as well as the interaction between the aggregator layer and the power market. As Figure 1 shown, the two are closely connected through an information flow and feedback mechanism, and jointly promote the efficient implementation of the power demand response.

[0054] The specific steps of S4 are as follows:

[0055] Step 1: Monitor the actual charging behavior of the user. The actual charging behavior is specifically the real-time charging data of the user side, and compare it with the expected behavior in the optimization strategy.

[0056] Step 2: Calculate the deviation degree: Evaluate the deviation between the actual behavior of the user and the optimization strategy, and judge whether adjustment is needed.

[0057] The deviation degree calculation can use a mathematical formula to measure the difference between the actual charging behavior of the user and the optimal charging strategy. The formula for calculating the deviation degree D is as follows:

[0058]

[0059] Where:

[0060] D: Charging behavior deviation degree (total deviation).

[0061] T: Total number of time periods within the optimization period (such as the number of hours in a day).

[0062] Actual charging power (kW) of the user at time t.

[0063] The optimal charging power (kW) of time t in the global optimization strategy.

[0064] w t : The time weight factor, which can be used to assign higher weights to certain critical periods (such as peak periods).

[0065] Step 3: Execute the adjustment mechanism:

[0066] Price incentive: If the user does not charge according to the recommended time period, the electricity price can be adjusted (for example, increasing the electricity price during peak hours and decreasing it during off-peak hours) to guide the user to optimize the charging time.

[0067] Prompt message: Send notifications to the user, such as reminding the best charging time period and encouraging participation in the optimization plan.

[0068] Step 4: User feedback and adaptive adjustment: According to the user response, further optimize the price strategy or adjust the notification method to improve user acceptance and implementation effect.

[0069] The user layer focuses on the needs and behavior patterns of electric vehicle users, mainly including the following components:

[0070] 1) User data collection: Aggregate multi-dimensional data by investigating and statistically analyzing the user's charging behavior (such as charging time, frequency, and location), market electricity price information, and user feedback, providing a basis for demand analysis.

[0071] 2) Demand analysis: Analyze the collected data, construct a user profile, deeply understand the user's charging preferences and behavior patterns, and use LLM to simulate user behavior to generate personalized demand forecasts.

[0072] 3) Demand response participation: Design personalized demand response programs, including incentives for users, to encourage users to actively participate in demand response activities.

[0073] 4) Information flow: The user layer transmits demand information, feedback, and participation status to the aggregator layer in real time to ensure that the aggregator can timely grasp the changes in user needs.

[0074] The aggregator layer is mainly responsible for the scheduling of power resources and market interaction, including the following key contents:

[0075] 1) Resource scheduling: The aggregator real-time schedules power resources according to the demand information provided by the user layer and the market conditions, ensuring the effective allocation of power during high-demand periods and enhancing the stability and reliability of the system.

[0076] 2) Market Interaction: Aggregators interact with other market participants such as power generators and distribution companies to formulate power procurement and sales strategies, ensuring the best economic benefits in the power market.

[0077] 3) Demand Forecasting: Based on user behavior analysis and market information, aggregators use LLM for accurate demand forecasting, providing a scientific basis for power dispatch.

[0078] 4) Decision Support: Through an intelligent decision support system, aggregators can analyze data to generate decision suggestions and use dynamic optimization algorithms to adjust demand response strategies in real time to adapt to changes in the power market.

[0079] 5) Feedback Mechanism: The aggregator layer collects feedback information from users to further optimize demand response strategies, ensuring that aggregators can better meet user needs and improve user satisfaction.

[0080] The specific steps of the present invention include:

[0081] Step 1) Data Collection and Preprocessing

[0082] First, the system needs to collect the charging behavior data of electric vehicle users, including charging records, driving data, and external factors. For example, the charging record data includes charging time, location, and charging amount; vehicle usage data records the driving mileage and average energy consumption; external factors cover weather, traffic conditions, etc. After data collection, preprocessing is required, including operations such as denoising and standardization. The denoising process uses a moving average filtering method, and the formula is as follows:

[0083]

[0084] Standardization normalizes all variables so that the data can be processed on a unified scale.

[0085] Step 2) Simulation and Prediction of User Charging Demand Based on LLM. The process diagram of electric vehicle user demand simulation is as Figure 2 shown.

[0086] Based on historical charging data and vehicle operation data, a large language model is used for simulation and prediction of charging demand. First, the Transformer structure is used to extract time series features and calculate the charging probability P charge (t) of users at different time periods. Then, combined with the user profile, the charging demand of users in the future for a period of time is calculated

[0087] To improve the prediction accuracy, the prediction results will be weighted and corrected according to the user behavior pattern. For example, if a user usually charges at night, the model will assign a higher weight to this period to optimize the prediction accuracy.

[0088] Step 3) Two-layer optimization algorithm: Aggregate charging demand response optimization

[0089] This step adopts a two-layer optimization algorithm, which includes global optimization (macro level) and individual optimization (user level).

[0090] At the global optimization level, the goal is to minimize the peak grid load P peak , while ensuring that the charging demand is met. The following constraints are considered during the optimization process: charging demand constraint D t ≥D min and the upper limit of the grid load P t ≤P max . In the specific implementation, the gradient descent method is used to optimize the electricity price strategy C t : where η is the learning rate and L is the loss function, which measures the grid load balance.

[0091] At the individual optimization level, the goal is to minimize the user's charging cost U cost . The dynamic programming method is used to optimize the user's charging time T charge , and the optimal charging scheduling strategy is calculated as follows:

[0092] In this way, the system can optimize the charging schedule on the premise of ensuring the user's needs, so as to achieve the overall load balance.

[0093] Step 4) Demand response strategy execution

[0094] After the optimization strategy is formulated, the system will push the charging scheduling plan to the user side to guide the user to adjust the charging time to match the global optimization result. At the same time, the system will monitor the charging behavior in real time. If it is found that the user's behavior deviates from the expectation (such as not charging at the recommended time), adjustments will be made through price incentives or prompt messages to ensure the achievement of the optimization goal.

[0095] Step 5) Effect evaluation and model optimization

[0096] After the strategy is executed, the system needs to evaluate the optimization effect and calculate key indicators, including the grid load peak shaving rate R peak and the user cost reduction rate R cost . If the optimization goal is not achieved as expected, the system will return to Step 2 to adjust the prediction model or optimization algorithm to improve the applicability of the scheduling plan.

[0097] Step 6) System update and iteration

[0098] To ensure long-term optimization effects, the system will regularly update the LLM prediction model and retrain it with new data. In addition, the optimization strategy will also be dynamically adjusted according to changes in user behavior patterns to meet the charging needs of electric vehicle users, ensuring the long-term stable operation and continuous optimization of the system.

[0099] Through the above optimization process, the system can reduce the peak load of the power grid while meeting the charging needs of users, improving the overall charging efficiency. A two-layer decision-making optimization framework for electric vehicle users and aggregators based on LLM provided by the present invention is as Figure 3 shown.

[0100] The two-layer demand response method for electric vehicle users and aggregators based on LLM proposed by the present invention has the following specific optimization points and benefits:

[0101] 1. Realize personalized charging behavior prediction

[0102] The present invention uses LLM to construct a personalized user charging behavior model, and through deep learning of data such as users' historical driving behaviors, charging preferences, and travel habits, accurately predicts users' charging needs. Compared with the uniform scheduling method in traditional demand response strategies, in the part of user charging behavior prediction based on LLM in step 2 of the present invention, the Transformer structure is used to extract time series features, calculate the charging probability P charge (t), and correct it in combination with the user portrait to improve the prediction accuracy. This method greatly improves the accuracy of response and avoids the problem of uneven power grid load caused by inaccurate demand prediction.

[0103] 2. Improve demand response efficiency

[0104] The present invention adopts a two-layer optimization algorithm (step 3), which divides demand response into a user layer and an aggregator layer. The optimization of the user layer dynamically adjusts according to individual charging needs to ensure that the optimization results meet the specific needs of users; the aggregator layer, from the perspective of power grid load balance, globally analyzes the charging needs of all users and optimizes the electricity price strategy C t by using the gradient descent method to adjust users' charging behaviors, thereby reducing the peak load of the power grid. This method is more flexible than the single-layer optimization strategy and improves the overall scheduling efficiency.

[0105] 3. Optimize the balance of power grid load

[0106] In the global optimization part of step 3, the loss function L for optimizing the electricity price strategy is designed as: where P target is the target load level, and the electricity price strategy is updated by gradient descent optimization: This method effectively balances the power grid load, makes the power supply more stable, and reduces the operating costs at the same time.

[0107] 4. Improve user responsiveness and accuracy

[0108] In the implementation of the demand response strategy (step 4), the system pushes the optimized strategy to the user side and combines price incentives and hierarchical reward mechanisms to guide users to adjust their charging times, enabling them to participate in demand response more proactively and improving the response effect. This part combines incentive strategies for user behavior, transforming users from passive scheduling to active optimization, thereby improving the system execution efficiency.

[0109] 5. Support continuous improvement and evolution

[0110] In the system update and iteration part of step 6, the system regularly updates the LLM prediction model, retrains it based on new data, and dynamically adjusts the optimization strategy to make the demand response strategy more adaptable to future load changes. The introduction of the feedback mechanism endows the system with self - adaptability, not only improving the prediction accuracy but also enhancing the overall optimization level of the system, providing more reliable decision - making support for future demand response strategies.

[0111] The present invention proposes a two - layer demand response method for electric vehicle users and aggregators based on large language models, aiming to improve the utilization efficiency of power resources through an innovative two - layer architecture and intelligent technologies. This method effectively integrates the personalized needs of electric vehicle users and the supply strategies of aggregators to achieve two - way interaction and real - time regulation. The key technical points cover real - time user behavior analysis, personalized demand response strategies, aggregator scheduling optimization, data - driven decision - making support systems, adaptive feedback mechanisms, and security and privacy protection measures, etc. Through the synergistic effect of these technologies, the present invention can achieve more intelligent and efficient power demand response management, providing a new solution for the interaction between electric vehicle users and aggregators.

[0112] 1. Two - layer demand response architecture: The present invention designs a two - layer architecture. The first layer is the electric vehicle user layer, which is responsible for real - time collection and analysis of users' charging demands, behavior patterns, and their response to electricity prices, etc. The second layer is the aggregator layer. Aggregators use the collected user data and market information to formulate power supply strategies. This architecture can effectively coordinate user demands and aggregator supplies, achieve two - way interaction, and improve the utilization efficiency of power resources.

[0113] 2. Application of Large Language Model: By integrating large language models, the present invention can deeply analyze the user's historical charging data, market electricity price changes, and user feedback. The powerful natural language processing ability of the LLM enables it to extract valuable information from complex data and generate accurate predictions of user needs. This process not only improves the intelligence level of demand response but also provides a more scientific decision-making basis for aggregators.

[0114] 3. Real-time User Behavior Analysis: This technical point includes developing a real-time monitoring system that can track users' charging habits and electricity consumption patterns and analyze factors affecting users' charging decisions, such as time periods, weather, market electricity prices, etc. Through real-time analysis of this data, the system can timely adjust demand response strategies to better meet user needs and improve the timeliness and accuracy of responses.

[0115] 4. Personalized Demand Response Strategy: The present invention implements a personalized demand response strategy based on user behavior and preferences. The system uses the LLM to model users' historical behavior, identify each user's charging pattern and preferences, and thus generate customized response plans. The personalized strategy can not only improve user satisfaction but also encourage more users to participate in demand response programs, increasing the flexibility of overall load regulation.

[0116] 5. Aggregator Scheduling Optimization: To improve the resource allocation efficiency of aggregators, the present invention has developed an aggregator scheduling optimization algorithm. This algorithm takes into account the changing demands of users and the real-time situation of the electricity market, and intelligently adjusts the power supply strategy of aggregators to optimize the power scheduling arrangement. Through this optimization, aggregators can better balance the supply and demand relationship, reduce operating costs, and improve service quality.

[0117] 6. Data-driven Decision Support System: The present invention constructs a comprehensive data-driven decision support system that can integrate various information from electric vehicle users and the electricity market. Through data analysis and visualization tools, this system provides real-time decision support for aggregators, helping them quickly respond to market changes and user needs and achieve efficient demand response management.

[0118] 7. Adaptive Feedback Mechanism: An adaptive feedback mechanism is designed. Through real-time evaluation of demand response results, the system can continuously optimize demand response strategies based on user feedback and market changes. This mechanism not only improves the flexibility and adaptability of the system but also enhances users' sense of participation, making them more actively participate in power scheduling.

[0119] 8. Security and Privacy Protection Measures: This invention pays particular attention to the security and privacy protection of user data and designs a series of security mechanisms, including data encryption, access control, and anonymization processing. Through these measures, the security and confidentiality of user data during the demand response process are ensured, enhancing users' trust in the system and encouraging more users to participate in demand response activities.

[0120] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. An optimization method for electric vehicle user charging scheduling based on large language models, characterized in that, The method includes the following steps: S1. Obtain the preprocessed historical charging data and vehicle operation data; S2. Based on the historical charging data and vehicle operation data, use a large language model to simulate and predict the charging demand, and obtain the charging demand prediction result; S3. Adopt a two-layer optimization algorithm based on the historical charging data and vehicle operation data. The two-layer optimization algorithm includes global optimization and individual optimization. First, at the global optimization level, construct a global optimization model with the goal of minimizing the peak grid load to obtain the optimal electricity price strategy. Then, at the individual optimization level, optimize the user's charging time based on the optimal electricity price strategy to obtain the optimal charging scheduling strategy; S4. Push the optimal charging scheduling strategy to the user side, obtain the real-time electricity consumption behavior of the user side. If the real-time electricity consumption behavior deviates from the expectation, adjust it through price incentives or prompt messages; S5. Calculate the key indicators, and judge whether the optimization goal is completed based on the key indicators. If it is completed, execute S6. Otherwise, adjust the large language model or the two-layer optimization algorithm and return to S2; S6. Historical charging data and vehicle operation data.

2. The optimization method for the charging scheduling of electric vehicle users based on a large language model according to claim 1, wherein The specific steps of using the large language model to simulate and predict the charging demand to obtain the charging demand prediction result are as follows: Use the Transformer structure to extract the time series features of the historical charging data and vehicle operation data, calculate the charging probability of the user in different time periods, and combine the user portrait to calculate the charging demand prediction result of the user in the future period of time.

3. The optimized method for charging scheduling of electric vehicle users based on large language models according to claim 1, characterized in that, The optimization goal of the global optimization model is to minimize the peak grid load, specifically: Wherein: represents the total grid load at time t; T represents the set of all time periods within the optimization cycle.

4. The optimization method for the charging scheduling of electric vehicle users based on a large language model according to claim 3, characterized in that The constraint conditions of the global optimization model include charging demand constraints and grid load upper limit constraints.

5. The optimization method for electric vehicle user charging scheduling based on a large language model according to claim 4, wherein, The method for solving the global optimization model is: adopt the gradient descent method to optimize the electricity price strategy.

6. The optimization method for electric vehicle user charging scheduling based on a large language model according to claim 5, wherein Adopting the gradient descent method to optimize the electricity price strategy is specifically: the electricity price at the (k + 1)-th iteration is: Among them, represents the electricity price for the next iteration, η is the learning rate, and L is the loss function.

7. The optimization method for the charging scheduling of electric vehicle users based on a large language model according to claim 6, wherein, The specific loss function is: Among them, P target is the target load level, and P t represents the current load level.

8. A method for optimizing the charging scheduling of electric vehicle users based on a large language model according to claim 1, characterized in that, The optimal charging scheduling strategy is: Among them, D t represents the charging demand, and C t represents the electricity price of the optimal electricity price strategy.

9. The optimization method for the charging scheduling of electric vehicle users based on a large language model according to claim 1, wherein, The specific steps of S4 are: Monitor the actual charging behavior of the user. The actual charging behavior is specifically the real-time charging data of the user side; Evaluate the deviation between the user's actual behavior and the optimal charging scheduling strategy based on the deviation degree. Among them, the calculation formula of the deviation degree D is: Where: D represents the charging behavior deviation; T represents the total number of time periods within the optimization cycle; represents the actual charging power of the user at time t; represents the optimal charging power at time t in the global optimization strategy; w t represents the time weight factor; If the deviation exceeds the threshold, execute the adjustment mechanism. The adjustment mechanism includes price incentives and prompt messages. Among them, the price incentive is: adjust the electricity price, increase the electricity price during peak hours and decrease the electricity price during off-peak hours to guide the user to optimize the charging time; The prompt message is: send a notice to the user to remind the best charging period.

10. The optimization method for the charging scheduling of electric vehicle users based on a large language model according to claim 1, wherein The preprocessing includes denoising and standardization. The denoising adopts the moving average filtering method.

Images