Intelligent charging pile based on user behavior analysis and control method thereof

Abstract

The invention discloses an intelligent charging pile based on user behavior analysis and a control method thereof. The method comprises the following steps: acquiring electric power parameters, equipment states and user voiceprint features through an edge end; the cloud end performs load prediction by using a space-time diagram neural network, and fuses multi-source data to update a fault model; according to the method, the charging power can be intelligently adjusted based on the real-time power grid state, the load fluctuation can be effectively stabilized, differentiated electricity price and service feedback can be provided according to the historical behaviors of the user, and the user experience is improved. Therefore, the user is guided to select an idle charging pile or off-peak charging by a non-mandatory economic means, the independent selection right of the user is respected while stable operation of a power grid is guaranteed, the enthusiasm of the user in participating in demand side response is improved, and meanwhile, the two-factor security authentication process combining voiceprint verification and token unlocking has a good security authentication effect. And higher safety and reliability are provided in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of intelligent management and control of electric vehicle charging facilities, specifically to an intelligent charging pile based on user behavior analysis and its control method. Background Technology

[0002] With the rapid increase in the popularity of electric vehicles, the operation and management of charging infrastructure faces increasingly complex challenges. Current charging pile management often suffers from the following prominent problems: First, resource allocation and grid load balancing are challenging. Most charging station systems lack real-time forecasting and response capabilities for the overall regional load, making them prone to congestion during peak hours, resulting in excessively long waiting times for users and potentially impacting the local power grid. Traditional centralized control methods struggle to efficiently process massive amounts of heterogeneous data and lack multi-objective optimization capabilities.

[0003] Secondly, the user experience needs improvement. Users often struggle to find the optimal charging option, and common charging guidance strategies frequently overlook individual user charging habits and price sensitivity, failing to provide recommendations that truly meet their needs and thus reducing user satisfaction.

[0004] Furthermore, operational efficiency and sophistication are insufficient. Many systems fail to fully utilize historical user behavior data to build sophisticated user profiles and credit systems, making it impossible to implement differentiated services and incentives based on credit ratings, and thus hindering the effective guidance of user behavior to improve overall operational efficiency.

[0005] Furthermore, security and identity authentication methods are limited. Traditional methods for controlling access to charging stations, such as IC cards or simple password verification, are susceptible to theft or forgetting, and lack effective and convenient emergency solutions.

[0006] Therefore, there is an urgent need in this field for an intelligent control method that can deeply integrate user behavior analysis, real-time power grid status perception, intelligent decision-making and security authentication technologies to achieve collaborative optimization of charging pile clusters, personalized service guidance and safe and reliable operation.

[0007] Chinese patent literature discloses an electric vehicle charging guidance method and system that considers users' charging station selection habits [Application No.: 202411182464.9, Publication No.: CN119066556A]. This includes: using SOFM and SHAP algorithms to cluster and analyze users' historical charging data to construct user profiles and divide users into different groups; and employing entropy weighting to calculate the weights of time and cost to construct a user satisfaction model. This comparative patent focuses on a macro-level charging station selection guidance strategy. Its core is to segment users into groups and model their preferences based on historical data, thereby maximizing satisfaction through recommendations. Its technology focuses on decision optimization in the guidance phase.

[0008] In contrast, this invention focuses more on the real-time, refined control and safe execution of individual charging piles or charging pile clusters. It deeply integrates edge computing, cloud-based intelligent analysis, dynamic credit mechanisms, and two-factor authentication. This invention achieves closed-loop management across the entire chain from "guidance" to "control," "authentication," and "execution," and proposes more in-depth and comprehensive solutions, especially in real-time response, credit system linkage, and security control. Summary of the Invention

[0009] A smart charging pile control method based on user behavior analysis, characterized by comprising: S1: Real-time data collected from edge devices, including power parameters, device status data, and voiceprint feature vectors of user voice commands; S2: Encrypt and upload the data to the cloud; S3: The cloud uses a spatiotemporal graph neural network to process regional charging pile cluster data and generate a power grid load prediction model; S4: Integrates multi-source real-time data to update the fault prediction model, including real-time temperature curves, user operation frequency, and confidence weights calculated based on regional fault history data; S5: Calculate dynamic credit scores based on payment timeliness and equipment operation records, and generate tiered electricity pricing strategies by linking load forecast results; S6: Encrypt and transmit the model and strategy to the edge; S7: The edge performs power adjustment and fault protection. If the voiceprint verification is successful, the operation is authorized. If it fails, a preset token is used to unlock.

[0010] Preferably, step S1 includes: DMFCC and ALPCC features of speech signals were extracted using an improved multi-scale CNN. An adversarial autoencoder is used to suppress environmental noise and generate voiceprint feature vectors. After encryption, the data is uploaded to the cloud-based user database.

[0011] The above technical solutions can efficiently and accurately acquire biometric information from user voice commands and effectively suppress the interference of environmental noise on voiceprint recognition, thereby improving the reliability of user authentication and system security.

[0012] Specifically, an improved multi-scale CNN is used to extract deep features from the speech signal, simultaneously capturing dynamic Mel-frequency coherence spectroscopy (DMFCC) and linear predictive coding (ALPCC) features, resulting in richer and more comprehensive voiceprint representation. Combined with an adversarial autoencoder, external noise and the user's biometric features are effectively separated, ensuring the purity and accuracy of voiceprint features even in complex environments. Therefore, the generated voiceprint feature vector not only has good noise resistance but also accurately reflects the user's individual characteristics.

[0013] Furthermore, encrypting voiceprint features and uploading them to the cloud-based user database further ensures the privacy and security of user data, laying a solid foundation for subsequent identity authentication and intelligent control. In practical applications, this process enhances the charging pile's ability to accurately identify users, improves the system's flexibility and security level, and is particularly suitable for intelligent charging scenarios with multiple concurrent users and complex outdoor environments. It is of great significance for improving the overall intelligent management of charging piles and the user experience.

[0014] Preferably, step S5 includes: A dynamic credit score is calculated using an algorithm, with the input parameters being payment timeliness rate and device operation compliance rate. A tiered electricity price table is generated from the associated power grid load forecast results; Settlement is executed via blockchain smart contracts.

[0015] The aforementioned technical solutions enable dynamic assessment of user credit and personalized electricity pricing, while ensuring the security and transparency of the settlement process. By introducing algorithms, key parameters such as user payment timeliness and equipment operation compliance are comprehensively analyzed, effectively improving the accuracy and fairness of credit scoring, thereby providing a basis for subsequent pricing.

[0016] Specifically, the dynamic credit score system not only reflects users' actual behavior in a timely manner but also automatically generates tiered electricity pricing tables based on load forecasts, enabling differentiated charging. Combining credit scores with grid load conditions effectively guides users to use charging stations rationally, optimizes grid resource allocation, and promotes the intelligent and efficient operation of the system. Meanwhile, using blockchain smart contracts for settlement ensures the decentralization and immutability of the transaction process, guaranteeing data security and transparency.

[0017] In practical applications, this technical solution can meet the needs for fair billing and privacy protection in multi-user scenarios, improve the automation and intelligence level of charging pile management, and is of great significance for promoting the healthy development of the smart charging pile industry.

[0018] Preferably, step S7 includes: Decrypt the TEE strategy and run the CNN-LSTM model; When the credit score is below the threshold, the charging current is limited and advanced voice features are turned off; If voiceprint verification fails, a preset static token authorization operation can be accepted.

[0019] The above technical solution enables secure, controllable, and dynamically adjustable user authorization and management processes for charging stations. By decrypting the control policy and running a CNN-LSTM model within a Trusted Execution Environment (TEE), not only is the security of the user authentication process ensured, but user behavior data can also be analyzed in real time to achieve intelligent judgment and authorization.

[0020] Specifically, by setting a credit score threshold, the system can automatically limit the charging current and disable advanced voice functions when a user's credit score falls below a predetermined level, thereby reducing risk and resource waste and protecting device security. If voiceprint verification fails, the system allows users to authorize operations using a preset static token, improving the system's flexibility and user-friendliness and ensuring users' normal usage rights in special circumstances.

[0021] In practical applications, the above measures can dynamically and automatically adjust charging parameters and function permissions based on user credit, effectively reducing equipment violations and abuse, and improving overall operational efficiency and security capabilities. This design is of great significance for charging pile identity authentication, access control, and intelligent protection in multi-user scenarios.

[0022] Preferably, the system includes: Edge execution device: includes a multi-source data acquisition module and a TEE strategy execution module; Cloud-based analytics device: includes a load forecasting engine, a user behavior analysis engine, and a blockchain node; The credit score calculation of the user behavior analysis engine is only associated with payment records and device operation logs.

[0023] The above technical solutions enable distributed intelligent management and safe, efficient operation of charging pile systems. Edge-end execution devices are responsible for real-time collection of multi-source data, including user operations, environmental data, and charging equipment data, and securely execute policies within a protected TEE environment, ensuring the privacy and security of sensitive on-site operations. Cloud-based analytics devices integrate load forecasting and user behavior analysis engines, enabling real-time load forecasting and behavior analysis based on large-scale historical data. Combined with blockchain nodes, this ensures transparency and tamper-proof characteristics for all data and settlement processes.

[0024] Specifically, the user behavior analysis engine calculates credit scores only by linking user payment records and device operation logs, avoiding interference from other irrelevant information and thus improving the objectivity and fairness of the assessment results. The load forecasting engine utilizes spatiotemporal correlation analysis to accurately predict grid load changes, optimize charging strategies and resource scheduling, and, in conjunction with blockchain nodes, achieve secure data sharing and automated settlement, enhancing the overall system's intelligent management level.

[0025] In practical applications, this distributed system architecture can simultaneously meet the requirements of data security, scalability, and intelligent operation, which is of great significance for improving the automated operation capabilities and user experience of the charging pile industry.

[0026] Preferably, the voiceprint processing unit performs: Voiceprint vectors are generated using Bi-LSTM; The APP token verification process is triggered when voiceprint matching fails.

[0027] The above technical solutions enable efficient and intelligent user identification, and improve the overall system's security and ease of use. By utilizing a bidirectional long short-term memory network (Bi-LSTM) to perform deep modeling and feature extraction on user voice data through the voiceprint processing unit, accurate voiceprint vectors reflecting individual biometric characteristics can be generated, thereby effectively ensuring the accuracy and robustness of voiceprint recognition.

[0028] Specifically, the Bi-LSTM model can fully capture the temporal information and bidirectional features in the speech signal, ensuring that the voiceprint recognition process has a stronger ability to adapt to various speech variations and environmental interference. When voiceprint matching fails, the system automatically triggers the APP token verification process, providing users with an alternative secondary authentication method. This not only increases the security level of the authentication process but also enhances the system's user-friendliness and continuous availability, allowing users to complete operations smoothly even when voiceprint recognition fails.

[0029] In practical applications, the above mechanism can better adapt to diverse identity authentication needs, ensuring that the system has both high security and a convenient operating experience. It is of great significance for improving the overall security and user satisfaction of identity-sensitive devices such as smart charging piles.

[0030] Preferably, the strategy execution module: Authorize the operation directly upon successful voiceprint verification; The execution record is uploaded to the blockchain for evidence storage in real time.

[0031] The above technical solution enables efficient authorization and full-process traceability management of user operation flows. Through the policy execution module, when voiceprint verification is successful, the system can immediately complete the authorization operation for the user, significantly shortening the verification and authorization latency, improving user experience convenience, and enhancing system response speed.

[0032] Specifically, all authorized operations and related execution records are uploaded to the blockchain system in real time for evidence storage. Leveraging the immutability and distributed nature of blockchain, the forgery and tampering of operation records are effectively prevented, ensuring that every authorized action has complete traceability and a reliable audit trail. This not only enhances the system's security and transparency but also meets the high standards of data management required for compliance and accountability in scenarios such as finance, electricity, and shared mobility.

[0033] In practical applications, the above design can support large-scale distributed identity authentication and access control, ensuring that every operation can be traced securely, greatly improving the compliance and reliability of equipment operation and maintenance, and is of great significance for various application scenarios with extremely high requirements for security and data trustworthiness.

[0034] Preferably, the multi-source data acquisition module includes: Dual microphone array and noise reduction chip capture voice stream; The temperature / vibration sensor samples the device status at 10Hz. Power sensors monitor the harmonic distortion rate of the power grid.

[0035] The above technical solution enables high-precision, all-around, real-time acquisition of charging equipment operating status and user voice information, providing a solid data foundation for intelligent system management and fault protection. The multi-source data acquisition module integrates a dual-microphone array and a noise reduction chip, efficiently and clearly acquiring user voice streams, effectively suppressing environmental noise, and improving the accuracy of voice recognition and voiceprint authentication.

[0036] Specifically, temperature and vibration sensors dynamically monitor the equipment's operating status at a high frequency of 10Hz, enabling timely detection of abnormal vibrations or temperature changes and providing reliable support for equipment condition assessment and fault early warning. Simultaneously, power sensors monitor the grid's harmonic distortion rate in real time, ensuring power quality, preventing harmonic pollution from damaging the equipment, and providing a scientific basis for subsequent safety strategies and charging load management.

[0037] In practical applications, this multi-source parallel high-precision acquisition design can meet the real-time data requirements of smart charging piles in complex operating environments, significantly improve user interaction experience, equipment safety and system intelligent perception capabilities, and is of great significance for promoting the sustainable development of smart energy and smart IoT platforms.

[0038] Compared with the prior art, the present invention has the following advantages: 1. This invention deeply integrates user credit assessment, grid load forecasting, and dynamic pricing strategies to construct a win-win incentive mechanism. This method not only intelligently adjusts charging power based on real-time grid conditions to effectively mitigate load fluctuations, but also provides differentiated electricity prices and service feedback based on users' historical behavior. This non-mandatory economic approach guides users to choose idle charging stations or off-peak charging, ensuring stable grid operation while respecting users' right to choose and enhancing their enthusiasm for participating in demand-side response.

[0039] 2. This invention designs a two-factor authentication process combining voiceprint verification and token unlocking, providing higher security and reliability in complex environments. When biometric recognition fails due to environmental noise or identity theft, this method can seamlessly switch to an emergency verification channel based on a mobile application. This ensures the security of charging operation permissions, effectively prevents device theft, and avoids business interruptions caused by the failure of a single authentication method, guaranteeing a smooth user experience for legitimate users in various scenarios. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the process of this invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] The flowchart of this invention is shown below. Figure 1 As shown.

[0043] Edge data acquisition: Step 1.1 Speech signal acquisition and noise reduction: A dual-microphone array is used, with phase matching accuracy ≤ ±2° and sensitivity -38dBV / Pa, to acquire raw speech signals; Environmental noise suppression is performed using a digital noise reduction module; Noise reduction principle: Eliminate steady-state noise (such as fan hum) based on adaptive filters; Step 1.2 Voiceprint Feature Extraction: DMFCC Feature Extraction: ; Parameter description: M: Mel spectrum matrix (size 64 frames × 40 bands), which is the energy distribution of the speech signal on the Mel scale; W: 3×3 convolution kernel, which extracts local gradient features of the spectrum; ALPCC Feature Extraction: ; Parameter description: Linear predictive coding coefficients, simulating the resonance characteristics of the human vocal tract; Feature enhancement: The features of DMFCC and ALPCC are concatenated and input into the adversarial autoencoder (AAE) to generate a noise-resistant voiceprint vector. The environmental noise and biological features are separated through adversarial training.

[0044] Step 1.3 Equipment Status Monitoring: Temperature sensor: 10Hz sampling rate, records the charging module temperature in real time; Vibration sensor: monitors vibration energy in the 500-800Hz frequency band.

[0045] 2. Cloud-based intelligent analysis: Step 2.1 Power Grid Load Forecasting Spatial dependency modeling ; Parameter description: Euclidean distance between charging stations; Gaussian kernel scale parameters; Step 2.2 Electricity Price Decision Formula: ; Parameter description: Basic electricity price; Price sensitivity coefficient; : Power grid load factor; Load threshold; 3. Credit score-linked control: Step 3.1 Dynamic Credit System: ; Step 3.2 Hierarchical Execution Strategy: ; When the credit score is greater than or equal to 70, the function can be used in its entirety; when the credit score is less than 70, the rated current will be reduced proportionally. Step 3.3 Voiceprint Verification Process Bi-LSTM dynamic voiceprint matching formula: Voiceprint vector synthesis formula: Average the hidden state sequence of the entire speech segment to obtain a fixed-length voiceprint vector. ; Voiceprint matching formula: ; in To collect voiceprint vectors on site, Pre-store voiceprint vectors in the cloud. To verify the threshold.

[0046] Example 1: Dynamic load splitting recommendation based on real-time load forecasting Scene Description During the evening rush hour, a user initiates a charging request to their frequently visited charging station A via a mobile application. The system detects that the station's current load rate has reached 92%, and there are several charging stations in the surrounding area with significantly different load rates. To improve grid operating efficiency and optimize user experience, the system activates a dynamic load diversion recommendation mechanism based on real-time load forecasting.

[0047] Technical process 1. Edge data acquisition: The edge-end data acquisition module of the charging pile collects multiple operational data in real time: Power parameters: Three-phase voltage 380V±5%, current harmonic distortion rate ≤3%, real-time power 158kW; Equipment status data: Charging module temperature 48℃, cooling fan speed 3200rpm, contactor status normal; Voiceprint feature vector of user voice commands: acquired through dual microphone array, signal-to-noise ratio 42dB; 2. Cloud load forecasting: The spatiotemporal graph neural network processes regional charging pile cluster data, and the calculation formula is as follows: ; Parameter description: =1500m (Euclidean distance between stations A and B); =0.5 (Gaussian kernel scale parameter); 3. User behavior analysis: Credit scores are calculated using a dynamic credit system. Input parameters: Payment on time rate 93%, equipment operation standardization rate 98%; Output credit score: Sc=85; 4. Tiered electricity pricing strategy: Electricity price calculation for station A (high load station): ; Parameter description: =1.2 yuan / kWh (basic electricity price); =0.95 (Price sensitivity coefficient for high-load sites). =92 (Real-time load rate of station A); =40 (load threshold); Electricity price calculation for Station B (low load station): ; Parameter description: =1.2 yuan / kWh (basic electricity price); =0.2 (Price sensitivity coefficient for high-load sites). =28 (Real-time load rate of station A); =40 (load threshold); Implementation results: The system successfully achieved load balancing, effectively reducing the load rate of station A, increasing the utilization rate of station B, reducing the average waiting time for users, lowering charging costs, and improving the efficiency of power grid operation.

[0048] Example 2: Credit Incentive Guidance Based on Usage Habits Scene Description The user consistently charges in a fixed area, demonstrating high location stability (location standard deviation ≤ 0.3km over the last 30 days). The system detected an excellent credit score and consistent charging behavior, and to further optimize the allocation of regional charging resources, a credit incentive mechanism was activated.

[0049] Technical process 1. Edge data acquisition: The edge-end data acquisition module of the charging pile collects multiple operational data in real time: Power parameters: Three-phase voltage 380V±5%, current harmonic distortion rate ≤3%, real-time power 158kW; Equipment status data: Charging module temperature 48℃, cooling fan speed 3200rpm, contactor status normal; Voiceprint feature vector of user voice commands: acquired through dual microphone array, signal-to-noise ratio 42dB; 2. User behavior analysis: Credit scores are calculated using a dynamic credit system. Input parameters: Payment on time rate 100%, Equipment operation standardization rate 100%; Output credit score: Sc=95; User behavior analysis engine extracts features: Location stability index: 0.98 (based on the variance of charging location over the past 30 days); Regular timing: 89% of charging occurs between 7:00 PM and 10:00 PM on weekdays; According to the tiered execution strategy, full-power authorization permissions can be granted; 3. Credit Incentive Feedback: Based on the user's excellent credit performance, the system provides: Priority charging: Guarantee charging power during peak hours; Rate discount: Enjoy an additional 5% discount on top of the tiered electricity pricing; Service Upgrade: Priority handling channel for abnormal situations.

[0050] Example 3: Emergency Handling of Voiceprint Verification Failure Scene Description An unregistered user attempts to operate a charging station using a registered user's voiceprint. The system detects a mismatch in voiceprint features and triggers a security protection mechanism. This scenario simulates one of the main security threats faced by biometric identification systems.

[0051] Technical process 1. Voiceprint feature extraction: Perform voiceprint processing at the edge: DMFCC and ALPCC features of speech signals were extracted using an improved multi-scale CNN. An anti-autoencoder is used to suppress environmental noise; Generate voiceprint feature vector: =[0.12,-0.08,0.25,...]; 2. Voiceprint feature comparison: Based on the voiceprint matching formula: ; Comparison result: 0.67 =0.85; The system triggers the APP token verification process.

[0052] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart charging pile control method based on user behavior analysis, characterized in that, include: S1: Real-time data collected from edge devices, including power parameters, device status data, and voiceprint feature vectors of user voice commands; S2: Encrypt and upload the data to the cloud; S3: The cloud uses a spatiotemporal graph neural network to process regional charging pile cluster data and generate a power grid load prediction model; S4: Integrates multi-source real-time data to update the fault prediction model, including real-time temperature curves, user operation frequency, and confidence weights calculated based on regional fault history data; S5: Calculate dynamic credit scores based on payment timeliness and equipment operation records, and generate tiered electricity pricing strategies by linking load forecast results; S6: Encrypt and transmit the model and strategy to the edge; S7: The edge performs power adjustment and fault protection. If the voiceprint verification is successful, the operation is authorized. If it fails, a preset token is used to unlock.

2. The method according to claim 1, characterized in that, Step S1 includes: DMFCC and ALPCC features of speech signals were extracted using an improved multi-scale CNN. An adversarial autoencoder is used to suppress environmental noise and generate voiceprint feature vectors. After encryption, the data is uploaded to the cloud-based user database.

3. The method according to claim 1, characterized in that, Step S5 includes: A dynamic credit score is calculated using an algorithm, with the input parameters being payment timeliness rate and device operation compliance rate. A tiered electricity price table is generated from the associated power grid load forecast results; Settlement is executed via blockchain smart contracts.

4. The method according to claim 1, characterized in that, Step S7 includes: Decrypt the TEE strategy and run the CNN-LSTM model; When the credit score is below the threshold, the charging current is limited and advanced voice features are turned off; If voiceprint verification fails, a preset static token authorization operation can be accepted.

5. A smart charging pile system implementing claims 1-4, characterized in that, The system includes: Edge execution device: includes a multi-source data acquisition module and a TEE strategy execution module; Cloud-based analytics device: includes a load forecasting engine, a user behavior analysis engine, and a blockchain node; The credit score calculation of the user behavior analysis engine is only associated with payment records and device operation logs.

6. The system according to claim 5, characterized in that, The voiceprint processing unit performs the following: Voiceprint vectors are generated using Bi-LSTM; The APP token verification process is triggered when voiceprint matching fails.

7. The system according to claim 5, characterized in that, The strategy execution module mentioned above: Authorize the operation directly upon successful voiceprint verification; Execution records are uploaded to the blockchain for evidence storage in real time.

8. The system according to claim 5, characterized in that, The multi-source data acquisition module includes: Dual microphone array and noise reduction chip capture voice stream; The temperature / vibration sensor samples the device status at 10Hz. Power sensors monitor the harmonic distortion rate of the power grid.

Images