Charging strategy intelligent pushing method and system

Abstract

The invention provides a charging strategy intelligent pushing method and system in the technical field of charging pile management, and the method comprises the steps: S1, a server collects multi-source heterogeneous data including user portrait data, energy dynamic data, vehicle state data and household load data, and carries out the preprocessing and fusion to obtain a user charging data set; s2, inputting the user charging data set into the double-agent decision model to obtain a personalized charging strategy; s3, pushing the personalized charging strategy to the mobile terminal; s4, the mobile terminal displays the personalized charging strategy, and pushes the personalized charging strategy to the charging pile for execution based on an input execution instruction; and S5, the server collects the adoption rate of the personalized charging strategy and charging behavior feedback, and updates user portrait data and model parameters. The method has the advantages that the personalized charging strategy is dynamically generated and pushed, and the user experience and the energy utilization efficiency are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of charging pile management technology, and in particular to a method and system for intelligently pushing charging strategies. Background Technology

[0002] With the accelerated global energy structure transformation and the booming development of the new energy vehicle industry, electric vehicles, as the core carrier of the green transportation system, are experiencing rapid and continuous growth in their social ownership. Home charging stations, as a key infrastructure for electric vehicle energy replenishment, are increasingly becoming the mainstream charging choice for private users due to their flexible deployment, ease of use, and controllable costs. Currently, home charging stations generally adopt AC slow charging technology, with their power supply directly connected to the household power grid, sharing the same electricity metering system with other electrical appliances in the home (such as lighting, air conditioning, and other household appliances). This model has the following advantages:

[0003] 1. Ease of use: It supports users to charge as soon as they park, especially during nighttime parking hours, which can automatically complete the charging process. This effectively avoids the problems of queuing and parking space shortages at public charging stations during peak hours, and significantly improves the convenience of charging and the efficiency of time utilization for users.

[0004] 2. Economic advantages: It can make full use of the residential time-of-use electricity pricing policy. Users can adjust their main charging activities to low electricity price periods by setting up charging plans, which has a significant cost advantage over public fast charging piles or commercial charging stations.

[0005] 3. Exclusivity and security: Charging stations installed independently in private parking spaces or garages provide a dedicated charging environment, protecting user privacy and avoiding the risks of equipment misuse and management that may exist in public charging locations.

[0006] However, with the increasing diversification of energy structures, the continuous promotion of intelligent technologies, and the rising demand for personalized services, the static and single charging strategies of traditional home charging stations are gradually revealing the following limitations, hindering further optimization of user experience and energy utilization efficiency:

[0007] 1. Lack of dynamic response capability: It is unable to perceive changes in grid load, fluctuations in time-of-use electricity prices, and the intermittent power supply characteristics of renewable energy (such as household photovoltaic power generation) in real time. Therefore, it is difficult to automatically adjust charging behavior during periods of lowest electricity prices or higher proportion of clean electricity, thus failing to fully realize economic and environmental benefits.

[0008] 2. Insufficient Adaptation to User Behavior: Existing charging strategies largely rely on users manually setting fixed time schedules, failing to generate personalized solutions by incorporating multi-dimensional data such as user travel patterns, charging urgency, and vehicle battery status. For example, they cannot proactively fully charge the battery in advance based on long-distance travel needs the following day, or intelligently reduce charging priority in scenarios with low usage demand.

[0009] 3. Lack of home energy coordination mechanism: No intelligent coordination strategy has been established between the charging process and the total household electricity load. During peak electricity consumption periods, charging piles may continue to operate at high power, leading to a surge in the total household electricity load, which may even trigger circuit breaker protection or cause an increase in demand-based electricity charges. Furthermore, it is impossible to achieve linkage optimization with the Home Energy Management System (HEMS).

[0010] 4. Data value not fully exploited: A large amount of data generated during the operation of charging piles, such as charging time, energy consumption records, and user plugging and unplugging behavior, has not been collected and deeply analyzed by the system. It has also failed to combine user profiles (such as charging preferences, price sensitivity, and vehicle usage frequency) for strategy optimization and lacks the ability to achieve self-evolution through machine learning and training with historical data.

[0011] Therefore, how to provide a method and system for intelligently pushing charging strategies to dynamically generate and push personalized charging strategies in order to improve user experience and energy efficiency has become an urgent technical problem to be solved. Summary of the Invention

[0012] The technical problem to be solved by the present invention is to provide a method and system for intelligently pushing charging strategies, so as to dynamically generate and push personalized charging strategies to improve user experience and energy utilization efficiency.

[0013] In a first aspect, the present invention provides a method for intelligently pushing charging strategies, comprising the following steps:

[0014] Step S1: The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data and household load data, and preprocesses and merges the multi-source heterogeneous data to obtain the user charging dataset.

[0015] Step S2: The server inputs the user charging dataset into a dual-agent decision model constructed based on the demand prediction agent module, the energy optimization agent module, and the decision output module to obtain a personalized charging strategy.

[0016] Step S3: The server encrypts the personalized charging strategy into an encrypted charging strategy and pushes the encrypted charging strategy to the corresponding mobile terminal in real time.

[0017] Step S4: The mobile terminal decrypts the received encrypted charging strategy in real time, obtains and displays the personalized charging strategy, and pushes the personalized charging strategy to the charging pile for execution based on the input execution command.

[0018] Step S5: The server collects the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the server updates the user profile data and the model parameters of the dual-agent decision model through an online learning mechanism.

[0019] Furthermore, step S1 specifically includes:

[0020] The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. The user profile data includes at least historical charging records, travel schedules, charging urgency tags, price sensitivity coefficients, and vehicle usage frequency. The energy dynamic data includes at least real-time grid load, time-of-use electricity price curves, and predicted household photovoltaic power generation. The vehicle status data includes at least remaining battery capacity, battery health status, and expected charging time. The household load data includes at least real-time total electricity load and the operating status of key appliances.

[0021] The multi-source heterogeneous data is preprocessed as follows: Historical charging records undergo preprocessing including missing value handling, time alignment, and feature extraction; travel schedules undergo preprocessing including natural language parsing and structured processing; charging urgency tags, price sensitivity coefficients, and vehicle usage frequency undergo preprocessing including digital encoding and normalization; real-time grid load and time-of-use electricity price curves undergo preprocessing including parsing and alignment, and resampling; predicted household photovoltaic power generation power undergoes preprocessing including uncertainty handling and unit unification; remaining battery capacity undergoes preprocessing including unit conversion and range calculation; battery health status undergoes preprocessing including normalization; expected charging duration undergoes preprocessing including logical verification; real-time total power load undergoes preprocessing including filtering and noise reduction; and the operating status of key electrical appliances undergoes preprocessing including event detection and status encoding.

[0022] The user charging dataset is obtained by fusing the preprocessed multi-source heterogeneous data by determining the reference time axis, as well as data association and alignment.

[0023] Furthermore, in step S2, the demand prediction agent module is used to infer the user charging dataset through the LSTM network, predict the user's travel demand and charging urgency in the next 24 hours, and output the demand priority weight.

[0024] The energy optimization agent module is used to infer the user charging dataset through the DQN network and calculate the optimal charging time window; the objective function of the DQN network integrates three optimization objectives: the lowest electricity cost, the highest clean energy utilization rate, and peak avoidance of household load.

[0025] The decision output module is used to output personalized charging strategies based on demand priority weights and the optimal charging time window.

[0026] Furthermore, step S3 specifically includes:

[0027] The server obtains the current timestamp and the local MAC address, XORs the personalized charging policy with the MAC address to obtain first-level encrypted data, encrypts the first-level encrypted data and the timestamp using the AES algorithm to obtain second-level encrypted data, encrypts the second-level encrypted data and the MAC address using the RSA algorithm to obtain third-level encrypted data, calculates the hash value of the personalized charging policy using the SHA-256 algorithm, encrypts the third-level encrypted data and the hash value using the RC6 algorithm to obtain an encrypted charging policy, and pushes the encrypted charging policy to the corresponding mobile terminal in real time via the TLS protocol.

[0028] Furthermore, step S4 specifically includes:

[0029] The mobile terminal receives the encrypted charging strategy in real time, decrypts the encrypted charging strategy into three-level encrypted data and a hash value using the RC6 algorithm, decrypts the three-level encrypted data into two-level encrypted data and a MAC address using the RSA algorithm, decrypts the two-level encrypted data into one-level encrypted data and a timestamp using the AES algorithm, performs timeliness verification using the timestamp, XORs the one-level encrypted data with the MAC address to obtain a personalized charging strategy, performs integrity verification of the personalized charging strategy using the hash value, and displays the personalized charging strategy on the display screen.

[0030] Based on the input execution command, the mobile terminal pushes the personalized charging strategy to the charging pile via Bluetooth for execution.

[0031] Secondly, the present invention provides a charging strategy intelligent push system, comprising the following modules:

[0032] The user charging dataset construction module is used by the server to collect multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data and household load data, and to preprocess and fuse the multi-source heterogeneous data to obtain the user charging dataset.

[0033] The personalized charging strategy generation module is used by the server to input the user charging dataset into a dual-agent decision model constructed based on the demand prediction agent module, the energy optimization agent module, and the decision output module to obtain a personalized charging strategy.

[0034] A personalized charging strategy encrypted push module is used by the server to encrypt the personalized charging strategy into an encrypted charging strategy and push the encrypted charging strategy to the corresponding mobile terminal in real time.

[0035] The personalized charging strategy execution module is used by the mobile terminal to decrypt the received encrypted charging strategy in real time, obtain and display the personalized charging strategy, and push the personalized charging strategy to the charging pile for execution based on the input execution command.

[0036] The feedback optimization module is used by the server to collect the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the module updates the user profile data and the model parameters of the dual-agent decision model through an online learning mechanism.

[0037] Furthermore, the user charging dataset construction module is specifically used for:

[0038] The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. The user profile data includes at least historical charging records, travel schedules, charging urgency tags, price sensitivity coefficients, and vehicle usage frequency. The energy dynamic data includes at least real-time grid load, time-of-use electricity price curves, and predicted household photovoltaic power generation. The vehicle status data includes at least remaining battery capacity, battery health status, and expected charging time. The household load data includes at least real-time total electricity load and the operating status of key appliances.

[0039] The multi-source heterogeneous data is preprocessed as follows: Historical charging records undergo preprocessing including missing value handling, time alignment, and feature extraction; travel schedules undergo preprocessing including natural language parsing and structured processing; charging urgency tags, price sensitivity coefficients, and vehicle usage frequency undergo preprocessing including digital encoding and normalization; real-time grid load and time-of-use electricity price curves undergo preprocessing including parsing and alignment, and resampling; predicted household photovoltaic power generation power undergoes preprocessing including uncertainty handling and unit unification; remaining battery capacity undergoes preprocessing including unit conversion and range calculation; battery health status undergoes preprocessing including normalization; expected charging duration undergoes preprocessing including logical verification; real-time total power load undergoes preprocessing including filtering and noise reduction; and the operating status of key electrical appliances undergoes preprocessing including event detection and status encoding.

[0040] The user charging dataset is obtained by fusing the preprocessed multi-source heterogeneous data by determining the reference time axis, as well as data association and alignment.

[0041] Furthermore, in the personalized charging strategy generation module, the demand prediction agent module is used to infer the user charging dataset through the LSTM network, predict the user's travel demand and charging urgency in the next 24 hours, and output the demand priority weight.

[0042] The energy optimization agent module is used to infer the user charging dataset through the DQN network and calculate the optimal charging time window; the objective function of the DQN network integrates three optimization objectives: the lowest electricity cost, the highest clean energy utilization rate, and peak avoidance of household load.

[0043] The decision output module is used to output personalized charging strategies based on demand priority weights and the optimal charging time window.

[0044] Furthermore, the personalized charging strategy encrypted push module is specifically used for:

[0045] The server obtains the current timestamp and the local MAC address, XORs the personalized charging policy with the MAC address to obtain first-level encrypted data, encrypts the first-level encrypted data and the timestamp using the AES algorithm to obtain second-level encrypted data, encrypts the second-level encrypted data and the MAC address using the RSA algorithm to obtain third-level encrypted data, calculates the hash value of the personalized charging policy using the SHA-256 algorithm, encrypts the third-level encrypted data and the hash value using the RC6 algorithm to obtain an encrypted charging policy, and pushes the encrypted charging policy to the corresponding mobile terminal in real time via the TLS protocol.

[0046] Furthermore, the personalized charging strategy execution module is specifically used for:

[0047] The mobile terminal receives the encrypted charging strategy in real time, decrypts the encrypted charging strategy into three-level encrypted data and a hash value using the RC6 algorithm, decrypts the three-level encrypted data into two-level encrypted data and a MAC address using the RSA algorithm, decrypts the two-level encrypted data into one-level encrypted data and a timestamp using the AES algorithm, performs timeliness verification using the timestamp, XORs the one-level encrypted data with the MAC address to obtain a personalized charging strategy, performs integrity verification of the personalized charging strategy using the hash value, and displays the personalized charging strategy on the display screen.

[0048] Based on the input execution command, the mobile terminal pushes the personalized charging strategy to the charging pile via Bluetooth for execution.

[0049] The advantages of this invention are:

[0050] 1. The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. This data is preprocessed and fused to obtain a user charging dataset. This dataset is then input into a dual-agent decision-making model constructed based on a demand prediction agent module, an energy optimization agent module, and a decision output module to obtain a personalized charging strategy. The server encrypts the personalized charging strategy and pushes it to the corresponding mobile terminal. The mobile terminal decrypts the encrypted charging strategy in real time, obtains and displays the personalized charging strategy, and pushes it to the charging pile for execution based on the input execution command. The server collects the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the user profile data is updated through an online learning mechanism. The system also includes model parameters for a dual-agent decision-making model. This involves real-time collection and integration of multi-source data, such as user profiles, energy dynamics, vehicle status, and household load, by the server. This data is then input into a dual-agent decision-making model that combines demand forecasting and energy optimization agents. The model dynamically generates personalized charging strategies that accurately match users' individual charging needs (e.g., travel plans, cost sensitivity) while responding in real-time to grid conditions (e.g., time-of-use pricing, clean energy supply) and household electricity load. These strategies are encrypted and pushed to the user's mobile terminal in real-time. After decryption and display, the user authorizes execution. Simultaneously, the server continuously optimizes user profiles and model parameters using an online learning mechanism based on strategy adoption rates and feedback data from actual charging behavior, forming a closed-loop system. This ultimately enables the dynamic generation and delivery of personalized charging strategies, significantly improving user experience and energy efficiency.

[0051] 2. By collecting real-time user profile data (such as travel plans and price sensitivity), energy dynamic data (such as real-time electricity prices and photovoltaic power generation), vehicle status, and household load information from multiple dimensions, the data is preprocessed and integrated before being input into a dual-agent decision-making model. The LSTM network predicts the user's charging needs and urgency, while the DQN network integrates electricity cost, clean energy utilization rate, and household load peak-shaving targets to dynamically generate the optimal charging time window. This ultimately forms a personalized charging strategy that balances economy, environmental protection, and electricity safety. The strategy is securely pushed to the user's mobile terminal through multi-layer encryption and timeliness verification mechanisms. After manual confirmation, the strategy is executed. At the same time, the system continuously tracks the strategy adoption rate and user feedback, and uses an online learning mechanism to dynamically optimize the model and user profile, achieving continuous and accurate iteration of the strategy, thereby improving user experience and energy efficiency.

[0052] 3. The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. It then performs preprocessing (such as missing value handling, time alignment, and feature extraction) and fusion (by determining and associating data through a baseline time axis). This comprehensive data processing ensures data quality, consistency, and integrity, providing a reliable foundation for subsequent decision-making, improving the robustness and accuracy of the system, reducing decision-making errors caused by data noise or inconsistency, and demonstrating innovation in data integration.

[0053] 4. A dual-agent decision-making model is adopted, which combines a demand forecasting agent module (using an LSTM network to predict user travel demand and charging urgency) and an energy optimization agent module (using a DQN network to calculate the optimal charging time window, integrating multiple objectives such as lowest electricity cost, highest clean energy utilization, and peak avoidance of household load). This method, which combines deep learning and reinforcement learning, enables the generation of highly personalized and optimized charging strategies, improves the intelligence and efficiency of decision-making, and can dynamically balance user demand and energy constraints.

[0054] 5. Employing multiple encryption algorithms (such as AES, RSA, SHA-256, RC6) and protocols (TLS), including the integration of timestamps and MAC addresses for verification, this multi-layered encryption mechanism ensures the security of data during transmission and storage, prevents unauthorized access and data tampering, and effectively enhances user privacy protection.

[0055] 6. The server collects policy adoption rates and charging behavior feedback in real time, and updates user profile data and dual-agent decision-making model parameters through an online learning mechanism. It can continuously learn and improve, adapt to changing user behavior and external conditions (such as energy market fluctuations), realize the system's self-optimization and long-term performance improvement, reduce manual intervention, lower maintenance costs, and demonstrate the advanced application of artificial intelligence in energy management.

[0056] 7. The objective function of the energy optimization agent module integrates multiple optimization objectives such as the lowest electricity cost, the highest clean energy utilization rate, and peak load avoidance for households. It not only reduces users' charging costs, but also encourages the use of clean energy (such as photovoltaic power generation) and reduces peak load on the power grid.

[0057] 8. Through real-time acquisition and efficient processing and fusion of multi-source heterogeneous data, data quality and integrity are ensured. An innovative dual-agent decision-making model (combining LSTM and DQN networks) enables intelligent generation of personalized charging strategies, significantly improving decision-making accuracy and optimization capabilities. Meanwhile, multi-layer encryption mechanisms and real-time push ensure data security and privacy, while online learning mechanisms enable the system to adaptively optimize and continuously improve performance. In addition, multi-objective optimization promotes energy efficiency and sustainability, and a user-friendly interface and reliable technical details enhance practicality and ease of operation. Attached Figure Description

[0058] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0059] Figure 1 This is a flowchart of a smart charging strategy push method according to the present invention.

[0060] Figure 2 This is a schematic diagram of the structure of a smart charging strategy push system according to the present invention. Detailed Implementation

[0061] The technical solution in this application embodiment follows the general idea as follows: A server collects and integrates multi-source data such as user profiles, energy dynamics, vehicle status, and household load in real time. This data is then input into a dual-agent decision-making model that combines demand prediction and energy optimization agents. This dynamically generates personalized charging strategies that accurately match users' individual charging needs while responding to grid conditions and household electricity load in real time. These strategies are encrypted and pushed to the user's mobile terminal in real time. After decryption and display, the user authorizes execution. Simultaneously, the server continuously optimizes user profiles and model parameters using an online learning mechanism based on strategy adoption rates and feedback data from actual charging behavior. This forms a closed-loop system, enabling the dynamic generation and delivery of personalized charging strategies, thereby improving user experience and energy efficiency.

[0062] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the intelligent charging strategy push method of the present invention includes the following steps:

[0063] Step S1: The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data and household load data, and preprocesses and merges the multi-source heterogeneous data to obtain the user charging dataset.

[0064] Step S2: The server inputs the user charging dataset into a dual-agent decision model constructed based on the demand prediction agent module, the energy optimization agent module, and the decision output module to obtain a personalized charging strategy.

[0065] Step S3: The server encrypts the personalized charging strategy into an encrypted charging strategy and pushes the encrypted charging strategy to the corresponding mobile terminal in real time.

[0066] Step S4: The mobile terminal decrypts the received encrypted charging strategy in real time, obtains and displays the personalized charging strategy, and pushes the personalized charging strategy to the charging pile for execution based on the input execution command.

[0067] Step S5: The server collects the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the server updates the user profile data and the model parameters of the dual-agent decision model through an online learning mechanism.

[0068] Step S1 specifically involves:

[0069] The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. The user profile data includes at least historical charging records, travel schedules, charging urgency tags, price sensitivity coefficients, and vehicle usage frequency. The energy dynamic data includes at least real-time grid load, time-of-use electricity price curves, and predicted household photovoltaic power generation. The vehicle status data includes at least remaining battery capacity, battery health status, and expected charging time. The household load data includes at least real-time total electricity load and the operating status of key appliances.

[0070] The multi-source heterogeneous data is preprocessed, specifically by: processing the historical charging records for at least missing values ​​(for occasionally missing records, data interpolation between previous and subsequent time points can be used, or they can be directly marked as "no record"), and time alignment (converting all charging start times, end times, charging amounts, etc., into standard timestamps (such as Unix timestamps or ISO timestamps). The preprocessing of the travel itinerary includes at least natural language parsing (if the itinerary is user-input text (e.g., "go to the airport at 9 am tomorrow"), and feature extraction (deriving more useful features from the original records, such as average charging time, common charging periods, weekly charging frequency, average charging capacity, etc.); and structuring (converting the parsed information into structured fields, such as next usage time, planned travel mileage, and itinerary type label) for the charging urgency label, price sensitivity coefficient, and usage frequency. This includes at least digital encoding (converting labels such as "high", "medium", and "low" into numerical values ​​(e.g., 2, 1, 0)) and normalization (mapping numerical values ​​such as price sensitivity coefficient and usage frequency to a unified range (e.g., [0, 1]) to eliminate the influence of units and facilitate model processing).For example, preprocessing includes normalizing the frequency of vehicle use by the highest-frequency user to 1 and the lowest to 0; preprocessing the real-time load and time-of-use electricity price curves of the power grid by at least parsing and alignment (parsing the current electricity price, future electricity price prediction, etc. from the API data package provided by the power grid (possibly in JSON / XML format), resampling (power grid data may be at 15-minute or 1-hour intervals, requiring resampling (interpolation) to a higher frequency (e.g., 1 minute) timeline to maintain timestamp consistency); preprocessing the predicted power output of the household photovoltaic system by at least uncertainty handling (predicted values ​​usually have uncertainty (e.g., "predicted power generation of 3kW at 2 PM, possible error ±0.5kW"); preprocessing can process it into a range or retain its probability distribution characteristics for robust optimization by the optimization module), and unit unification (ensuring power is in kW and energy is in kWh); and preprocessing the remaining battery capacity by at least unit conversion (converting the original data (possibly voltage, energy Ah) to kWh). The preprocessing for the battery health status includes at least normalization; the preprocessing for the expected charging time includes at least logical verification (checking whether the expected charging time reported by the vehicle is reasonable (e.g., calculating a theoretical value based on the remaining capacity and charging pile power for cross-validation); if unreasonable, the calculated value is used instead of the reported value); the preprocessing for the real-time total power load includes at least filtering and noise reduction (current / power data may contain high-frequency noise, which needs to be filtered (e.g., moving average filtering) to obtain a smooth power curve); and the preprocessing for the operating status of key electrical appliances includes at least event detection (using load identification algorithms to determine which high-power appliances are running (e.g., air conditioners, washing machines, electric water heaters, etc.)) and status encoding (creating a binary feature (e.g., air conditioner status: 1 = on, 0 = off) or power level feature) for each key electrical appliance.

[0071] The user charging dataset is obtained by fusing the preprocessed multi-source heterogeneous data with the following steps: determining the baseline time axis (selecting the most core dimension common to all data as the baseline axis, usually time; determining the start point, end point and time resolution of a time series (e.g., one record every 1 minute or 5 minutes); this resolution is usually determined by the most frequently updated data source (e.g., vehicle status or household load)); data association and alignment (timestamp alignment, spatial association).

[0072] "Multi-source heterogeneous data" means that the data comes from different sources (users, power grids, vehicles, home appliances) and has different formats, structures, update frequencies and units. The goal of preprocessing is to clean, transform and standardize this messy raw data into a clean, uniform format that can be used for machine learning. The goal of fusion is to organically integrate this processed data to form a comprehensive "user charging dataset" oriented towards charging decision-making.

[0073] In step S2, the demand prediction agent module is used to infer the user charging dataset through the LSTM network, predict the user's travel demand and charging urgency in the next 24 hours, and output the demand priority weight.

[0074] The energy optimization agent module is used to infer the user charging dataset through the DQN network and calculate the optimal charging time window; the objective function of the DQN network integrates three optimization objectives: the lowest electricity cost, the highest clean energy utilization rate, and peak avoidance of household load.

[0075] The decision output module is used to output personalized charging strategies based on demand priority weights and the optimal charging time window.

[0076] The demand prediction agent module is constructed from an LSTM network including input gates, forget gates, output gates, and memory cells. It directly receives user profile data and vehicle status data as input, processes them, and outputs demand priority weights (a multi-dimensional vector, such as [time 1 weight, time 2 weight, ...], representing the charging priority at different time periods). This is used to capture the time-series dependence of user behavior, predict future travel demand and charging urgency, and generate demand priority weights to guide charging decisions. The LSTM network can effectively process time-series data, reduce prediction errors, and improve adaptability to personalized user needs.

[0077] The energy optimization agent module is constructed using a DQN network (Deep Q-Network). The DQN network includes convolutional or fully connected layers for state feature extraction, as well as Q-value estimation layers. The DQN network directly receives energy dynamic data and household load data as input, and outputs the optimal charging time window (e.g., a time interval, such as [start time, end time]) after inference. It is used to learn the optimal charging strategy in a dynamic environment, with the goal of minimizing electricity costs, maximizing clean energy utilization, and avoiding household load peaks. Employing a reinforcement learning mechanism, it can adapt to environmental changes, achieve multi-objective optimization, and improve energy utilization efficiency.

[0078] The decision output module consists of a rule engine or a simple fully connected layer. It receives the demand priority weights output by the demand forecasting agent module and the optimal charging time window output by the energy optimization agent module as input. Through fusion logic, it outputs personalized charging strategies (such as structured output including personalized charging strategies with charging time periods, suggested power, and priority prompts) to ensure that the strategy meets both the user's urgent needs and energy efficiency.

[0079] In practice, a Markov Decision Process (MDP) is constructed based on historical datasets to pre-train the dual-agent decision-making model; the state space of the Markov Decision Process is defined as follows:

[0080] S t =[U profile E grid ,P solar SOC t H load ];

[0081] Among them, S t U represents the state vector; profile E represents the feature vector of a user profile. grid Indicates the real-time load of the power grid; P solar Indicates the predicted photovoltaic power; SOC t Indicates the current battery capacity; H load Indicates the total household load;

[0082] The reward function of the dual-agent decision-making model is:

[0083] R = α·(1-Cost) ratio )+β·Green ratio +γ·(1-Load peak );

[0084] Where R represents the reward value of the reward function; Cost ratio Indicates the ratio of actual electricity cost to benchmark electricity cost; Green ratio Indicates the proportion of clean energy; Load peak This represents the peak household load coefficient during charging periods; α, β, and γ are all weight parameters that are dynamically adjusted based on user profiles.

[0085] Step S3 specifically involves:

[0086] The server obtains the current timestamp and the local MAC address, XORs the personalized charging policy with the MAC address to obtain first-level encrypted data, encrypts the first-level encrypted data and the timestamp using the AES algorithm to obtain second-level encrypted data, encrypts the second-level encrypted data and the MAC address using the RSA algorithm to obtain third-level encrypted data, calculates the hash value of the personalized charging policy using the SHA-256 algorithm, encrypts the third-level encrypted data and the hash value using the RC6 algorithm to obtain an encrypted charging policy, and pushes the encrypted charging policy to the corresponding mobile terminal in real time via the TLS protocol.

[0087] The combination of multi-layered encryption and dynamic parameters (such as timestamps and MAC addresses) can effectively resist replay attacks and unauthorized access. The encrypted charging strategy can be used for secure storage or transmission.

[0088] Step S4 specifically involves:

[0089] The mobile terminal receives the encrypted charging strategy in real time, decrypts the encrypted charging strategy into three-level encrypted data and a hash value using the RC6 algorithm, decrypts the three-level encrypted data into two-level encrypted data and a MAC address using the RSA algorithm, decrypts the two-level encrypted data into one-level encrypted data and a timestamp using the AES algorithm, performs timeliness verification using the timestamp, XORs the one-level encrypted data with the MAC address to obtain a personalized charging strategy, performs integrity verification of the personalized charging strategy using the hash value, and displays the personalized charging strategy on the display screen.

[0090] Based on the input execution command, the mobile terminal pushes the personalized charging strategy to the charging pile via Bluetooth for execution.

[0091] A preferred embodiment of the intelligent charging strategy push system of the present invention includes the following modules:

[0092] The user charging dataset construction module is used by the server to collect multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data and household load data, and to preprocess and fuse the multi-source heterogeneous data to obtain the user charging dataset.

[0093] The personalized charging strategy generation module is used by the server to input the user charging dataset into a dual-agent decision model constructed based on the demand prediction agent module, the energy optimization agent module, and the decision output module to obtain a personalized charging strategy.

[0094] A personalized charging strategy encrypted push module is used by the server to encrypt the personalized charging strategy into an encrypted charging strategy and push the encrypted charging strategy to the corresponding mobile terminal in real time.

[0095] The personalized charging strategy execution module is used by the mobile terminal to decrypt the received encrypted charging strategy in real time, obtain and display the personalized charging strategy, and push the personalized charging strategy to the charging pile for execution based on the input execution command.

[0096] The feedback optimization module is used by the server to collect the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the module updates the user profile data and the model parameters of the dual-agent decision model through an online learning mechanism.

[0097] The user charging dataset construction module is specifically used for:

[0098] The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. The user profile data includes at least historical charging records, travel schedules, charging urgency tags, price sensitivity coefficients, and vehicle usage frequency. The energy dynamic data includes at least real-time grid load, time-of-use electricity price curves, and predicted household photovoltaic power generation. The vehicle status data includes at least remaining battery capacity, battery health status, and expected charging time. The household load data includes at least real-time total electricity load and the operating status of key appliances.

[0099] The multi-source heterogeneous data is preprocessed, specifically by: processing the historical charging records for at least missing values ​​(for occasionally missing records, data interpolation between previous and subsequent time points can be used, or they can be directly marked as "no record"), and time alignment (converting all charging start times, end times, charging amounts, etc., into standard timestamps (such as Unix timestamps or ISO timestamps). The preprocessing of the travel itinerary includes at least natural language parsing (if the itinerary is user-input text (e.g., "go to the airport at 9 am tomorrow"), and feature extraction (deriving more useful features from the original records, such as average charging time, common charging periods, weekly charging frequency, average charging capacity, etc.); and structuring (converting the parsed information into structured fields, such as next usage time, planned travel mileage, and itinerary type label) for the charging urgency label, price sensitivity coefficient, and usage frequency. This includes at least digital encoding (converting labels such as "high", "medium", and "low" into numerical values ​​(e.g., 2, 1, 0)) and normalization (mapping numerical values ​​such as price sensitivity coefficient and usage frequency to a unified range (e.g., [0, 1]) to eliminate the influence of units and facilitate model processing).For example, preprocessing includes normalizing the frequency of vehicle use by the highest-frequency user to 1 and the lowest to 0; preprocessing the real-time load and time-of-use electricity price curves of the power grid by at least parsing and alignment (parsing the current electricity price, future electricity price prediction, etc. from the API data package provided by the power grid (possibly in JSON / XML format), resampling (power grid data may be at 15-minute or 1-hour intervals, requiring resampling (interpolation) to a higher frequency (e.g., 1 minute) timeline to maintain timestamp consistency); preprocessing the predicted power output of the household photovoltaic system by at least uncertainty handling (predicted values ​​usually have uncertainty (e.g., "predicted power generation of 3kW at 2 PM, possible error ±0.5kW"); preprocessing can process it into a range or retain its probability distribution characteristics for robust optimization by the optimization module), and unit unification (ensuring power is in kW and energy is in kWh); and preprocessing the remaining battery capacity by at least unit conversion (converting the original data (possibly voltage, energy Ah) to kWh). The preprocessing for the battery health status includes at least normalization; the preprocessing for the expected charging time includes at least logical verification (checking whether the expected charging time reported by the vehicle is reasonable (e.g., calculating a theoretical value based on the remaining capacity and charging pile power for cross-validation); if unreasonable, the calculated value is used instead of the reported value); the preprocessing for the real-time total power load includes at least filtering and noise reduction (current / power data may contain high-frequency noise, which needs to be filtered (e.g., moving average filtering) to obtain a smooth power curve); and the preprocessing for the operating status of key electrical appliances includes at least event detection (using load identification algorithms to determine which high-power appliances are running (e.g., air conditioners, washing machines, electric water heaters, etc.)) and status encoding (creating a binary feature (e.g., air conditioner status: 1 = on, 0 = off) or power level feature) for each key electrical appliance.

[0100] The user charging dataset is obtained by fusing the preprocessed multi-source heterogeneous data with the following steps: determining the baseline time axis (selecting the most core dimension common to all data as the baseline axis, usually time; determining the start point, end point and time resolution of a time series (e.g., one record every 1 minute or 5 minutes); this resolution is usually determined by the most frequently updated data source (e.g., vehicle status or household load)); data association and alignment (timestamp alignment, spatial association).

[0101] "Multi-source heterogeneous data" means that the data comes from different sources (users, power grids, vehicles, home appliances) and has different formats, structures, update frequencies and units. The goal of preprocessing is to clean, transform and standardize this messy raw data into a clean, uniform format that can be used for machine learning. The goal of fusion is to organically integrate this processed data to form a comprehensive "user charging dataset" oriented towards charging decision-making.

[0102] In the personalized charging strategy generation module, the demand prediction agent module is used to infer the user charging dataset through the LSTM network, predict the user's travel demand and charging urgency in the next 24 hours, and output the demand priority weight.

[0103] The energy optimization agent module is used to infer the user charging dataset through the DQN network and calculate the optimal charging time window; the objective function of the DQN network integrates three optimization objectives: the lowest electricity cost, the highest clean energy utilization rate, and peak avoidance of household load.

[0104] The decision output module is used to output personalized charging strategies based on demand priority weights and the optimal charging time window.

[0105] The demand prediction agent module is constructed from an LSTM network including input gates, forget gates, output gates, and memory cells. It directly receives user profile data and vehicle status data as input, processes them, and outputs demand priority weights (a multi-dimensional vector, such as [time 1 weight, time 2 weight, ...], representing the charging priority at different time periods). This is used to capture the time-series dependence of user behavior, predict future travel demand and charging urgency, and generate demand priority weights to guide charging decisions. The LSTM network can effectively process time-series data, reduce prediction errors, and improve adaptability to personalized user needs.

[0106] The energy optimization agent module is constructed using a DQN network (Deep Q-Network). The DQN network includes convolutional or fully connected layers for state feature extraction, as well as Q-value estimation layers. The DQN network directly receives energy dynamic data and household load data as input, and outputs the optimal charging time window (e.g., a time interval, such as [start time, end time]) after inference. It is used to learn the optimal charging strategy in a dynamic environment, with the goal of minimizing electricity costs, maximizing clean energy utilization, and avoiding household load peaks. Employing a reinforcement learning mechanism, it can adapt to environmental changes, achieve multi-objective optimization, and improve energy utilization efficiency.

[0107] The decision output module consists of a rule engine or a simple fully connected layer. It receives the demand priority weights output by the demand forecasting agent module and the optimal charging time window output by the energy optimization agent module as input. Through fusion logic, it outputs personalized charging strategies (such as structured output including personalized charging strategies with charging time periods, suggested power, and priority prompts) to ensure that the strategy meets both the user's urgent needs and energy efficiency.

[0108] In practice, a Markov Decision Process (MDP) is constructed based on historical datasets to pre-train the dual-agent decision-making model; the state space of the Markov Decision Process is defined as follows:

[0109] S t =[U profile E grid ,P solar SOC t H load ];

[0110] Among them, S t U represents the state vector; profile E represents the feature vector of a user profile. grid Indicates the real-time load of the power grid; P solar Indicates the predicted photovoltaic power; SOC t Indicates the current battery capacity; H load Indicates the total household load;

[0111] The reward function of the dual-agent decision-making model is:

[0112] R = α·(1-Cost) ratio )+β·Green ratio +γ·(1-Load peak );

[0113] Where R represents the reward value of the reward function; Cost ratio Indicates the ratio of actual electricity cost to benchmark electricity cost; Green ratio Indicates the proportion of clean energy; Load peak This represents the peak household load coefficient during charging periods; α, β, and γ are all weight parameters that are dynamically adjusted based on user profiles.

[0114] The personalized charging strategy encrypted push module is specifically used for:

[0115] The server obtains the current timestamp and the local MAC address, XORs the personalized charging policy with the MAC address to obtain first-level encrypted data, encrypts the first-level encrypted data and the timestamp using the AES algorithm to obtain second-level encrypted data, encrypts the second-level encrypted data and the MAC address using the RSA algorithm to obtain third-level encrypted data, calculates the hash value of the personalized charging policy using the SHA-256 algorithm, encrypts the third-level encrypted data and the hash value using the RC6 algorithm to obtain an encrypted charging policy, and pushes the encrypted charging policy to the corresponding mobile terminal in real time via the TLS protocol.

[0116] The combination of multi-layered encryption and dynamic parameters (such as timestamps and MAC addresses) can effectively resist replay attacks and unauthorized access. The encrypted charging strategy can be used for secure storage or transmission.

[0117] The personalized charging strategy execution module is specifically used for:

[0118] The mobile terminal receives the encrypted charging strategy in real time, decrypts the encrypted charging strategy into three-level encrypted data and a hash value using the RC6 algorithm, decrypts the three-level encrypted data into two-level encrypted data and a MAC address using the RSA algorithm, decrypts the two-level encrypted data into one-level encrypted data and a timestamp using the AES algorithm, performs timeliness verification using the timestamp, XORs the one-level encrypted data with the MAC address to obtain a personalized charging strategy, performs integrity verification of the personalized charging strategy using the hash value, and displays the personalized charging strategy on the display screen.

[0119] Based on the input execution command, the mobile terminal pushes the personalized charging strategy to the charging pile via Bluetooth for execution.

[0120] In summary, the advantages of this invention are as follows:

[0121] 1. The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. This data is preprocessed and fused to obtain a user charging dataset. This dataset is then input into a dual-agent decision-making model constructed based on a demand prediction agent module, an energy optimization agent module, and a decision output module to obtain a personalized charging strategy. The server encrypts the personalized charging strategy and pushes it to the corresponding mobile terminal. The mobile terminal decrypts the encrypted charging strategy in real time, obtains and displays the personalized charging strategy, and pushes it to the charging pile for execution based on the input execution command. The server collects the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the user profile data is updated through an online learning mechanism. The system also includes model parameters for a dual-agent decision-making model. This involves real-time collection and integration of multi-source data, such as user profiles, energy dynamics, vehicle status, and household load, by the server. This data is then input into a dual-agent decision-making model that combines demand forecasting and energy optimization agents. The model dynamically generates personalized charging strategies that accurately match users' individual charging needs (e.g., travel plans, cost sensitivity) while responding in real-time to grid conditions (e.g., time-of-use pricing, clean energy supply) and household electricity load. These strategies are encrypted and pushed to the user's mobile terminal in real-time. After decryption and display, the user authorizes execution. Simultaneously, the server continuously optimizes user profiles and model parameters using an online learning mechanism based on strategy adoption rates and feedback data from actual charging behavior, forming a closed-loop system. This ultimately enables the dynamic generation and delivery of personalized charging strategies, significantly improving user experience and energy efficiency.

[0122] 2. By collecting real-time user profile data (such as travel plans and price sensitivity), energy dynamic data (such as real-time electricity prices and photovoltaic power generation), vehicle status, and household load information from multiple dimensions, the data is preprocessed and integrated before being input into a dual-agent decision-making model. The LSTM network predicts the user's charging needs and urgency, while the DQN network integrates electricity cost, clean energy utilization rate, and household load peak-shaving targets to dynamically generate the optimal charging time window. This ultimately forms a personalized charging strategy that balances economy, environmental protection, and electricity safety. The strategy is securely pushed to the user's mobile terminal through multi-layer encryption and timeliness verification mechanisms. After manual confirmation, the strategy is executed. At the same time, the system continuously tracks the strategy adoption rate and user feedback, and uses an online learning mechanism to dynamically optimize the model and user profile, achieving continuous and accurate iteration of the strategy, thereby improving user experience and energy efficiency.

[0123] 3. The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. It then performs preprocessing (such as missing value handling, time alignment, and feature extraction) and fusion (by determining and associating data through a baseline time axis). This comprehensive data processing ensures data quality, consistency, and integrity, providing a reliable foundation for subsequent decision-making, improving the robustness and accuracy of the system, reducing decision-making errors caused by data noise or inconsistency, and demonstrating innovation in data integration.

[0124] 4. A dual-agent decision-making model is adopted, which combines a demand forecasting agent module (using an LSTM network to predict user travel demand and charging urgency) and an energy optimization agent module (using a DQN network to calculate the optimal charging time window, integrating multiple objectives such as lowest electricity cost, highest clean energy utilization, and peak avoidance of household load). This method, which combines deep learning and reinforcement learning, enables the generation of highly personalized and optimized charging strategies, improves the intelligence and efficiency of decision-making, and can dynamically balance user demand and energy constraints.

[0125] 5. Employing multiple encryption algorithms (such as AES, RSA, SHA-256, RC6) and protocols (TLS), including the integration of timestamps and MAC addresses for verification, this multi-layered encryption mechanism ensures the security of data during transmission and storage, prevents unauthorized access and data tampering, and effectively enhances user privacy protection.

[0126] 6. The server collects policy adoption rates and charging behavior feedback in real time, and updates user profile data and dual-agent decision-making model parameters through an online learning mechanism. It can continuously learn and improve, adapt to changing user behavior and external conditions (such as energy market fluctuations), realize the system's self-optimization and long-term performance improvement, reduce manual intervention, lower maintenance costs, and demonstrate the advanced application of artificial intelligence in energy management.

[0127] 7. The objective function of the energy optimization agent module integrates multiple optimization objectives such as the lowest electricity cost, the highest clean energy utilization rate, and peak load avoidance for households. It not only reduces users' charging costs, but also encourages the use of clean energy (such as photovoltaic power generation) and reduces peak load on the power grid.

[0128] 8. Through real-time acquisition and efficient processing and fusion of multi-source heterogeneous data, data quality and integrity are ensured. An innovative dual-agent decision-making model (combining LSTM and DQN networks) enables intelligent generation of personalized charging strategies, significantly improving decision-making accuracy and optimization capabilities. Meanwhile, multi-layer encryption mechanisms and real-time push ensure data security and privacy, while online learning mechanisms enable the system to adaptively optimize and continuously improve performance. In addition, multi-objective optimization promotes energy efficiency and sustainability, and a user-friendly interface and reliable technical details enhance practicality and ease of operation.

[0129] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligently pushing charging strategies, characterized in that: Includes the following steps: Step S1: The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data and household load data, and preprocesses and merges the multi-source heterogeneous data to obtain the user charging dataset. Step S2: The server inputs the user charging dataset into a dual-agent decision model constructed based on the demand prediction agent module, the energy optimization agent module, and the decision output module to obtain a personalized charging strategy. Step S3: The server encrypts the personalized charging strategy into an encrypted charging strategy and pushes the encrypted charging strategy to the corresponding mobile terminal in real time. Step S4: The mobile terminal decrypts the received encrypted charging strategy in real time, obtains and displays the personalized charging strategy, and pushes the personalized charging strategy to the charging pile for execution based on the input execution command. Step S5: The server collects the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the server updates the user profile data and the model parameters of the dual-agent decision model through an online learning mechanism.

2. The intelligent charging strategy push method as described in claim 1, characterized in that: Step S1 specifically involves: The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. The user profile data includes at least historical charging records, travel schedules, charging urgency tags, price sensitivity coefficients, and vehicle usage frequency. The energy dynamic data includes at least real-time grid load, time-of-use electricity price curves, and predicted household photovoltaic power generation. The vehicle status data includes at least remaining battery capacity, battery health status, and expected charging time. The household load data includes at least real-time total electricity load and the operating status of key appliances. The multi-source heterogeneous data are preprocessed, specifically by performing preprocessing on the historical charging records, including at least missing value processing, time alignment, and feature extraction. The travel itinerary undergoes at least natural language parsing and structured preprocessing. The charging urgency label, price sensitivity coefficient, and vehicle usage frequency are preprocessed, including at least digital encoding and normalization. The real-time load and time-of-use electricity price curves of the power grid are preprocessed, including at least parsing and alignment, and resampling; the predicted power generation value of the household photovoltaic system is preprocessed, including at least uncertainty handling and unit unification; and the remaining battery capacity is preprocessed, including at least unit conversion and range calculation. The battery health status is preprocessed, including at least normalization; the expected charging time is preprocessed, including at least logic verification. The real-time total power load is subjected to preprocessing including at least filtering and noise reduction; The operating status of the key electrical appliances is preprocessed, including at least event detection and status coding. The user charging dataset is obtained by fusing the preprocessed multi-source heterogeneous data by determining the reference time axis, as well as data association and alignment.

3. The intelligent charging strategy push method as described in claim 1, characterized in that: In step S2, the demand prediction agent module is used to infer the user charging dataset through the LSTM network, predict the user's travel demand and charging urgency in the next 24 hours, and output the demand priority weight. The energy optimization agent module is used to infer the user charging dataset through the DQN network and calculate the optimal charging time window; the objective function of the DQN network integrates three optimization objectives: the lowest electricity cost, the highest clean energy utilization rate, and peak avoidance of household load. The decision output module is used to output personalized charging strategies based on demand priority weights and the optimal charging time window.

4. The intelligent charging strategy push method as described in claim 1, characterized in that: Step S3 specifically involves: The server obtains the current timestamp and the local MAC address, XORs the personalized charging policy with the MAC address to obtain first-level encrypted data, encrypts the first-level encrypted data and the timestamp using the AES algorithm to obtain second-level encrypted data, encrypts the second-level encrypted data and the MAC address using the RSA algorithm to obtain third-level encrypted data, calculates the hash value of the personalized charging policy using the SHA-256 algorithm, encrypts the third-level encrypted data and the hash value using the RC6 algorithm to obtain an encrypted charging policy, and pushes the encrypted charging policy to the corresponding mobile terminal in real time via the TLS protocol.

5. The intelligent charging strategy push method as described in claim 1, characterized in that: Step S4 specifically involves: The mobile terminal receives the encrypted charging strategy in real time, decrypts the encrypted charging strategy into three-level encrypted data and a hash value using the RC6 algorithm, decrypts the three-level encrypted data into two-level encrypted data and a MAC address using the RSA algorithm, decrypts the two-level encrypted data into one-level encrypted data and a timestamp using the AES algorithm, performs timeliness verification using the timestamp, XORs the one-level encrypted data with the MAC address to obtain a personalized charging strategy, performs integrity verification of the personalized charging strategy using the hash value, and displays the personalized charging strategy on the display screen. Based on the input execution command, the mobile terminal pushes the personalized charging strategy to the charging pile via Bluetooth for execution.

6. A charging strategy intelligent push system, characterized in that: Includes the following modules: The user charging dataset construction module is used by the server to collect multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data and household load data, and to preprocess and fuse the multi-source heterogeneous data to obtain the user charging dataset. The personalized charging strategy generation module is used by the server to input the user charging dataset into a dual-agent decision model constructed based on the demand prediction agent module, the energy optimization agent module, and the decision output module to obtain a personalized charging strategy. A personalized charging strategy encrypted push module is used by the server to encrypt the personalized charging strategy into an encrypted charging strategy and push the encrypted charging strategy to the corresponding mobile terminal in real time. The personalized charging strategy execution module is used by the mobile terminal to decrypt the received encrypted charging strategy in real time, obtain and display the personalized charging strategy, and push the personalized charging strategy to the charging pile for execution based on the input execution command. The feedback optimization module is used by the server to collect the adoption rate of the personalized charging strategy and charging behavior feedback in real time. Based on the adoption rate and charging behavior feedback, the module updates the user profile data and the model parameters of the dual-agent decision model through an online learning mechanism.

7. The intelligent charging strategy push system as described in claim 6, characterized in that: The user charging dataset construction module is specifically used for: The server collects multi-source heterogeneous data in real time, including user profile data, energy dynamic data, vehicle status data, and household load data. The user profile data includes at least historical charging records, travel schedules, charging urgency tags, price sensitivity coefficients, and vehicle usage frequency. The energy dynamic data includes at least real-time grid load, time-of-use electricity price curves, and predicted household photovoltaic power generation. The vehicle status data includes at least remaining battery capacity, battery health status, and expected charging time. The household load data includes at least real-time total electricity load and the operating status of key appliances. The multi-source heterogeneous data are preprocessed, specifically by performing preprocessing on the historical charging records, including at least missing value processing, time alignment, and feature extraction. The travel itinerary undergoes at least natural language parsing and structured preprocessing. The charging urgency label, price sensitivity coefficient, and vehicle usage frequency are preprocessed, including at least digital encoding and normalization. The real-time load and time-of-use electricity price curves of the power grid are preprocessed, including at least parsing and alignment, and resampling; the predicted power generation value of the household photovoltaic system is preprocessed, including at least uncertainty handling and unit unification; and the remaining battery capacity is preprocessed, including at least unit conversion and range calculation. The battery health status is preprocessed, including at least normalization; the expected charging time is preprocessed, including at least logic verification. The real-time total power load is subjected to preprocessing including at least filtering and noise reduction; The operating status of the key electrical appliances is preprocessed, including at least event detection and status coding. The user charging dataset is obtained by fusing the preprocessed multi-source heterogeneous data by determining the reference time axis, as well as data association and alignment.

8. The intelligent charging strategy push system as described in claim 6, characterized in that: In the personalized charging strategy generation module, the demand prediction agent module is used to infer the user charging dataset through the LSTM network, predict the user's travel demand and charging urgency in the next 24 hours, and output the demand priority weight. The energy optimization agent module is used to infer the user charging dataset through the DQN network and calculate the optimal charging time window; the objective function of the DQN network integrates three optimization objectives: the lowest electricity cost, the highest clean energy utilization rate, and peak avoidance of household load. The decision output module is used to output personalized charging strategies based on demand priority weights and the optimal charging time window.

9. The intelligent charging strategy push system as described in claim 6, characterized in that: The personalized charging strategy encrypted push module is specifically used for: The server obtains the current timestamp and the local MAC address, XORs the personalized charging policy with the MAC address to obtain first-level encrypted data, encrypts the first-level encrypted data and the timestamp using the AES algorithm to obtain second-level encrypted data, encrypts the second-level encrypted data and the MAC address using the RSA algorithm to obtain third-level encrypted data, calculates the hash value of the personalized charging policy using the SHA-256 algorithm, encrypts the third-level encrypted data and the hash value using the RC6 algorithm to obtain an encrypted charging policy, and pushes the encrypted charging policy to the corresponding mobile terminal in real time via the TLS protocol.

10. The intelligent charging strategy push system as described in claim 6, characterized in that: The personalized charging strategy execution module is specifically used for: The mobile terminal receives the encrypted charging strategy in real time, decrypts the encrypted charging strategy into three-level encrypted data and a hash value using the RC6 algorithm, decrypts the three-level encrypted data into two-level encrypted data and a MAC address using the RSA algorithm, decrypts the two-level encrypted data into one-level encrypted data and a timestamp using the AES algorithm, performs timeliness verification using the timestamp, XORs the one-level encrypted data with the MAC address to obtain a personalized charging strategy, performs integrity verification of the personalized charging strategy using the hash value, and displays the personalized charging strategy on the display screen. Based on the input execution command, the mobile terminal pushes the personalized charging strategy to the charging pile via Bluetooth for execution.

Images