Battery swapping cabinet abnormal access behavior detection system based on pattern recognition

Abstract

The application relates to the technical field of intelligent battery replacement equipment safety monitoring and mode recognition, and particularly discloses a battery replacement cabinet abnormal access behavior detection system based on mode recognition. The system comprises a battery replacement cabinet body, a six-axis force / torque sensing unit, a biomechanical modeling unit, a mode recognition analysis unit, a behavior judgment unit and a early warning response unit. The six-axis force / torque sensing unit is used for collecting force / torque vector data generated by user operation in real time, and the force / torque vector data is combined with a compliance operation reference torque curve constructed based on the biomechanical characteristics of human upper limbs. Then, a mode recognition algorithm is used to compare and identify abnormal operation characteristics, to determine whether the operation is a high-risk behavior such as violent plugging or malicious impact, and to trigger a graded early warning response. The application can perceive physical interaction force characteristics, improve the accuracy and robustness of abnormal behavior recognition, reduce the false alarm rate, and enhance the operation safety and stability of the battery replacement cabinet in complex environments.

Description

Technical Field

[0001] This invention belongs to the field of safety monitoring and pattern recognition technology for intelligent battery swapping equipment, specifically relating to a battery swapping cabinet abnormal access behavior detection system based on pattern recognition. Background Technology

[0002] With the rapid rise of the new energy industry, battery swapping stations, as the core infrastructure for energy replenishment of electric vehicles, are playing an increasingly crucial role in improving energy replenishment efficiency and ensuring urban energy security. Intelligent battery swapping systems, through automated processes, achieve rapid battery turnover and refined management, becoming an important component of modern IoT and smart city construction. In a highly automated operating environment, battery swapping stations not only need charging and discharging scheduling capabilities but also need to establish robust security mechanisms to cope with complex outdoor environments and frequent user interactions.

[0003] Monitoring user behavior during battery storage and retrieval is a critical aspect of ensuring the safe operation of equipment. Existing monitoring solutions typically employ image recognition or simple position sensors to capture user movement trajectories, aiming to identify potential violations through real-time analysis of user operation processes. These solutions primarily attempt to prevent equipment damage caused by misoperation and maintain the physical integrity and operational stability of the battery swapping station by verifying the compliance of action sequences.

[0004] Existing technologies have limitations in processing user access behavior. Traditional visual recognition solutions are extremely sensitive to external interference, easily affected by ambient lighting, user clothing thickness, and individual differences in movement habits, making it difficult for the system to accurately distinguish between unintentional clumsy operations and intentional malicious damage. Current monitoring dimensions are limited to macroscopic displacement, lacking a deep perception of the physical force vectors during interaction, making it difficult to decipher the user's true intentions through mechanical characteristics. Due to the lack of biomechanical parameters, the system cannot establish an accurate assessment model for violent insertion or malicious impact, resulting in a high false alarm rate and difficulty in providing early warning protection for critical structural components.

[0005] There is an urgent need for a pattern recognition-based system for detecting abnormal access behavior in battery swapping cabinets. Summary of the Invention

[0006] The purpose of this invention is to provide a pattern recognition-based abnormal access behavior detection system for battery swapping cabinets, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A pattern recognition-based abnormal access behavior detection system for battery swapping cabinets includes a battery swapping cabinet body, a six-axis force / torque sensing unit, a biomechanical modeling unit, a pattern recognition analysis unit, a behavior determination unit, and an early warning response unit, as follows: The battery swapping cabinet body is equipped with a standard loading module for accommodating batteries. This loading module is configured to allow users to perform battery insertion or removal operations and serves as a physical interface to bear all the user's operational force. The six-axis force / torque sensing unit is embedded inside the loading module and is configured to collect the three-dimensional force components and three-dimensional torque components acting on the loading module in real time when the user performs battery storage and retrieval operations, forming a complete force / torque vector data stream. The biomechanical modeling unit is communicatively connected to the six-axis force / torque sensing unit and is configured to construct a standard human operation mechanics model based on the biomechanical characteristics of human upper limb operation. Based on a large amount of historical operation data from normal users, a benchmark torque curve characterizing compliant operation is generated. This benchmark torque curve has stability and linearity characteristics. The pattern recognition and analysis unit is connected to the six-axis force / torque sensing unit and the biomechanical modeling unit, respectively. It is configured to receive the real-time force / torque vector data stream and compare it with the reference torque curve. The preset anomaly detection algorithm identifies data features that deviate from the normal operating mode. The behavior determination unit is connected to the pattern recognition and analysis unit and is configured to determine whether the current operation belongs to violent plugging and unplugging, malicious impact or other high-risk abnormal access behavior based on the identified abnormal data features, and output the corresponding behavior classification result. The early warning response unit is connected to the behavior determination unit and is configured to trigger a corresponding safety response mechanism after receiving the high-risk behavior classification result, including but not limited to recording operation logs, locking the loading module, sending alarm signals to the remote monitoring center, or activating the on-site audible and visual warning device.

[0008] Preferably, the six-axis force / torque sensing unit is installed at the structural connection between the loading module and the battery swapping cabinet body to ensure that the collected force / torque data can fully reflect all the interactive forces applied by the user to the battery compartment and is not affected by the weight of the battery itself.

[0009] Furthermore, the human operation mechanics model constructed by the biomechanical modeling unit comprehensively considers the user's upper limb joint range of motion, muscle force exertion pattern and typical operation path, so that the reference torque curve can cover the normal operation range of users of different genders, ages and body types, thereby improving the generalization ability of the model.

[0010] Furthermore, the support vector machine algorithm or isolated forest algorithm used by the pattern recognition analysis unit is configured to run in unsupervised or semi-supervised mode, which can automatically learn the distribution boundary of normal operating data without pre-labeling all abnormal samples, and determine real-time data that deviates from the boundary as potential anomalies.

[0011] Preferably, the behavior determination unit is equipped with multiple risk level thresholds. The risk level thresholds are dynamically set according to the amplitude change rate, direction change degree and duration of the force / torque vector. When the real-time data exceeds any risk level threshold, the behavior determination result of the corresponding level is triggered.

[0012] Furthermore, the security response mechanism triggered by the early warning response unit has a tiered response capability. For low-risk anomalies, it only logs and uploads them to the cloud analysis platform. For medium-risk anomalies, it temporarily restricts the user account's operation permissions. For high-risk anomalies, it immediately physically locks the loading module and simultaneously notifies the operation and maintenance personnel.

[0013] Furthermore, the system also includes an environmental compensation unit connected to the six-axis force / torque sensing unit, configured to dynamically correct the original force / torque data based on ambient temperature, humidity, and cabinet vibration status, in order to eliminate measurement drift caused by non-human factors and ensure the accuracy of anomaly identification.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition provided by the present invention integrates a six-axis force / torque sensing unit in the loading module and combines biomechanical modeling and pattern recognition technologies.

[0015] 2. The system no longer relies on easily disturbed visual information, but instead perceives the force vector characteristics applied by the user to the device, improving the robustness and accuracy of abnormal behavior recognition.

[0016] 3. By establishing a standard operating torque curve that conforms to ergonomics and combining it with advanced anomaly detection algorithms, the system can filter out false alarms caused by differences in clothing thickness and movement habits, thereby reducing the false alarm rate.

[0017] 4. The tiered early warning mechanism can adopt differentiated response strategies based on the degree of risk, ensuring equipment safety while avoiding unnecessary operational interruptions for normal users.

[0018] 5. The environmental compensation mechanism further enhances the stability of the system under complex outdoor conditions, extends the service life of key structural components of the battery swapping cabinet, and provides solid technical support for the safe and efficient operation of large-scale battery swapping networks. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of anomaly identification based on biomechanical modeling and torque feature comparison in this invention; Figure 3 This is a flowchart of the main stages of dynamic correction of force / torque data based on environmental compensation in this invention. Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the sensing unit, modeling unit and decision unit in this invention; Figure 5 This is a flowchart of the main stages of the abnormal behavior classification and safety response mechanism in this invention. Detailed Implementation

[0020] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0021] A pattern recognition-based battery swapping cabinet abnormal access behavior detection system includes a battery swapping cabinet body, a six-axis force / torque sensing unit, a biomechanical modeling unit, a pattern recognition analysis unit, a behavior determination unit, an early warning response unit, and an environmental compensation unit.

[0022] The battery swapping cabinet, serving as the physical platform of the system, has its internal space divided into multiple independent battery compartments and integrates a power management module, a communication control backplane, and a structural reinforcement frame. The front panel of the cabinet features a standard loading module for accommodating batteries. This module is configured as a physical interface with precise guidance for users to insert or remove batteries. The loading module not only serves as the electrical connection point for battery charging and discharging but also acts as the physical interface, bearing all operational forces applied by the user during operation, including horizontal thrust, vertical gravity, and various eccentric torques caused by operating habits. The inner wall of the loading module is made of high-strength and wear-resistant composite material to ensure structural stability during frequent battery friction and collisions, providing a stable transmission medium for force perception.

[0023] The six-axis force / torque sensing unit is embedded in the structural support point inside the loading module, specifically at the structural connection between the loading module and the battery swapping cabinet body. The six-axis force / torque sensing unit is configured to use resistance strain gauges or piezoelectric ceramic elements to sense minute structural deformations caused by user operations. During battery storage and retrieval operations, the six-axis force / torque sensing unit collects the three-dimensional force and torque components acting on the loading module in real time. The three-dimensional force components include forces along three mutually perpendicular coordinate axes: horizontal axial force, horizontal lateral force, and vertical force; the three-dimensional torque components include torsional torques around these three coordinate axes. These force and torque components together constitute a complete force / torque vector data stream. By installing the sensor at the rigid connection between the loading module and the cabinet frame, the system can isolate the interference of the battery's own gravity on the dynamic operation force recognition, allowing the collected signals to purely reflect the interaction between the user and the device.

[0024] The biomechanical modeling unit communicates with the six-axis force / torque sensing unit via a high-speed data bus. The biomechanical modeling unit is configured to construct a standard human operational mechanics model based on the biomechanical characteristics of human upper limb operations. This model deeply analyzes the range of motion of joints, muscle group force characteristics, and skeletal lever principles when the human upper limb performs actions such as pushing, pulling, lifting, and releasing. Based on a large amount of historical operation data from normal users pre-stored in the memory, the biomechanical modeling unit generates a baseline torque curve representing compliant operation through statistical analysis and dynamic simulation. The baseline torque curve is defined as the ideal trajectory of the operating torque changing with displacement or time under normal battery access physical paths. This baseline torque curve exhibits stability and linearity; that is, during normal insertion, the torque change should be continuous and with a gentle gradient, without instantaneous spikes or abrupt phase changes.

[0025] The pattern recognition and analysis unit is electrically connected to the six-axis force / torque sensing unit and the biomechanical modeling unit, respectively. The pattern recognition and analysis unit is configured to receive real-time acquired force / torque vector data streams and use a built-in high-performance processor to perform multi-dimensional comparisons with the baseline torque curve. The pattern recognition and analysis unit integrates a preset anomaly detection algorithm to identify data features that deviate from normal operating patterns. In a specific implementation, the pattern recognition and analysis unit employs a support vector machine (SVM) algorithm, which is configured to find the optimal hyperplane in the feature space that can separate normal operating vectors from abnormal operating vectors. The SVM algorithm maps the original six-axis force data to a higher-dimensional feature space through a kernel function, capturing the complex nonlinear coupling relationship between force vectors and torque vectors. The pattern recognition and analysis unit can also be configured to run an isolated forest algorithm, which recursively partitions the feature space and calculates the path length of real-time data points in the tree structure. Since abnormal data (such as malicious impacts or violent shaking) are usually in the minority and deviate from the group in the feature distribution, they are represented by shorter path lengths in the algorithm and are thus identified as potential abnormal patterns.

[0026] The behavior determination unit is connected to the pattern recognition and analysis unit via a logic control bus. The behavior determination unit is configured to perform in-depth analysis and classification of the current user's intent based on identified abnormal data features. Internally, the behavior determination unit has a complex logic judgment matrix to determine whether the current operation constitutes forced insertion / removal, malicious impact, levering, or other high-risk abnormal access behavior. The behavior determination unit not only focuses on the absolute magnitude of the force but also emphasizes the rate of change of the force vector, i.e., the derivative of force with respect to time. For example, when a step change exceeding a preset threshold is detected in the vertical direction within a very short time, accompanied by violent oscillations in the horizontal direction, the behavior determination unit classifies it as "malicious levering" behavior; while when a continuous increase in force value along the insertion direction is detected but displacement stagnates, and the torque components exhibit an asymmetrical distribution, it is classified as "forced blocking insertion / removal." The behavior determination unit outputs a corresponding behavior classification result, which includes a behavior type code, confidence score, and risk level.

[0027] The early warning response unit is connected to the output of the behavior determination unit. The early warning response unit is configured to immediately trigger a corresponding safety response mechanism upon receiving a high-risk behavior classification result to protect the hardware security and data integrity of the battery swapping cabinet. The safety response mechanism has multi-layered linkage characteristics, including but not limited to: recording detailed operation logs in the non-volatile memory of the local controller, including the original six-axis force waveform at the trigger moment; immediately locking the moving parts of the loading module via an electromagnetic locking mechanism to prevent further damage; sending real-time alarm signals to the remote monitoring center via a wireless communication interface and simultaneously pushing a behavior characteristic analysis report; and activating the on-site audible and visual warning device, using a high-decibel buzzer and high-brightness flashing red lights to psychologically deter violators.

[0028] The environmental compensation unit's hardware circuitry is connected to the signal conditioning circuitry of the six-axis force / torque sensing unit. The environmental compensation unit is configured to monitor in real-time the ambient temperature, humidity, and background vibration experienced by the battery swapping cabinet (such as ground vibrations caused by nearby heavy vehicles). The environmental compensation unit internally stores a sensor temperature drift compensation model, enabling dynamic correction of the original force / torque data based on environmental parameters. For example, in high-temperature environments, the zero point of the strain gauge may drift; the environmental compensation unit corrects this by subtracting or adding to the sampled values ​​using a preset compensation coefficient, eliminating measurement deviations caused by non-human factors. This ensures that even in extremely cold, hot, or vibrating environments near transportation hubs, the system maintains accurate anomaly detection, avoiding false alarms due to environmental fluctuations.

[0029] The detailed structure of the six-axis force / torque sensing unit includes an elastomer machined from high-strength alloy steel, on which at least three sets of mutually perpendicular strain gauge bridges are symmetrically arranged. When a user applies an operating force, the elastomer undergoes nanometer-scale micro-strain, causing a linear change in the resistance of the strain gauges. The sensing unit incorporates a high-gain, low-noise instrumentation amplifier to amplify the weak bridge imbalance voltage to the volt level, which is then converted into a digital signal by a high-bit analog-to-digital converter. To ensure real-time signal transmission, the sampling frequency is set at a level no lower than 1000 Hz, enabling the system to capture the characteristics of transient shock waves generated by violent impacts.

[0030] The biomechanical modeling unit constructs a human operational mechanics model through the simulation of the upper limb dynamics chain. The model simplifies the user's hand, wrist, forearm, and upper arm into a series of rigid bodies connected by joints, and sets an upper limit for the maximum output torque of each joint based on human physiological limits. The baseline torque curve is not a single fixed value, but rather an envelope interval based on a probability distribution. The biomechanical modeling unit is configured to use Gaussian process regression or Bayesian networks to dynamically adjust the boundaries of the envelope interval based on different time periods and user profiles (such as gender preferences obtained through app registration information), enabling the model to generalize and accommodate normal operational movements of different user groups with vastly different physical abilities.

[0031] The pattern recognition and analysis unit is also configured to perform frequency domain analysis when performing data feature recognition. By performing a Fast Fourier Transform (FFT) on the torque time series, its spectral distribution characteristics are extracted. Normal user access behavior typically concentrates in the low-frequency region (e.g., below 10 Hz), exhibiting slow energy changes; while malicious knocking or tool-induced damage will produce characteristic peaks in the high-frequency region. The pattern recognition and analysis unit weighted and fused the time-domain SVM recognition results with the frequency-domain power spectral density analysis results to further improve the identification accuracy of complex abnormal behaviors.

[0032] The behavior determination unit is equipped with multiple risk level thresholds, including low-risk, medium-risk, and high-risk thresholds. The risk level thresholds are set based on factors such as the magnitude of the force vector, the rotational vector magnitude of the torque, the rate of energy change, and the duration of the abnormal pattern. The determination logic employs a fuzzy logic reasoning system, taking multiple continuously sampled abnormal features as input and outputting a risk index between 0 and 1. When the risk index is within a first preset range, it is determined to be low-risk; within a second preset range, it is determined to be medium-risk; and exceeding a third preset threshold, it is determined to be high-risk. This multi-level determination mechanism provides a decision-making basis for subsequent differentiated early warning.

[0033] The tiered response capability of the aforementioned early warning response unit is specifically manifested as follows: For operations deemed low-risk (such as minor collisions caused by user unfamiliarity with the operation), the system only generates a maintenance log in the background, prompting maintenance personnel to pay attention to the mechanical wear of the compartment; for medium-risk operations (such as repeated attempts at forced insertion at the wrong angle), the system will pop up a warning dialog box on the battery swapping cabinet display screen and temporarily restrict the user's account's operation permissions on the cabinet, requiring them to reread the operation guide; for high-risk operations (such as identifying signals that conform to the mechanical characteristics of prying or hammering), the system will immediately cut off the power supply logic of the loading module, initiate a powerful mechanical lock, and trigger a network alarm.

[0034] Example 2: Based on Example 1, Example 2 provides an abnormal access behavior detection system for battery swapping cabinets based on a distributed edge computing architecture, which aims to further improve the real-time performance and robustness of large-scale battery swapping networks when processing massive concurrent interactive data.

[0035] The detection system based on the distributed edge computing architecture, in addition to the units in Embodiment 1 above, further integrates an edge processing node, a distributed data consistency controller, and a multi-sensor fusion bus.

[0036] In Embodiment 2, the battery swapping cabinet is configured to include multiple intelligent compartment units with independent computing capabilities. Each compartment unit is independently equipped with a set of the aforementioned six-axis force / torque sensing units, and the structure of the loading module is deeply coupled and integrated with the sensors. The load-bearing structure of the loading module is designed as a floating support frame with self-decoupling function. This frame can transform complex spatial force systems into linear components that are easier for sensors to capture, reducing inter-axis crosstalk and providing a data source with a higher signal-to-noise ratio for high-frequency sampling on the edge side.

[0037] The edge processing nodes are deployed within the internal control room of the battery swapping cabinet, physically close to each intelligent storage unit. These edge processing nodes are configured to handle some of the computational tasks originally performed by the biomechanical modeling and pattern recognition analysis units. Each edge processing node includes a high-performance neural network acceleration unit (NPU) capable of running lightweight deep learning models, such as variants of Deep Residual Networks (ResNet), for real-time processing of high-dimensional raw data streams uploaded from six-axis force / torque sensors. By performing feature extraction and preliminary pattern filtering at the edge, the system can reduce the response latency for anomaly detection to the millisecond level, which is crucial for preventing sudden acts of violence.

[0038] The distributed data consistency controller is connected to multiple edge processing nodes and the central control unit of the battery swapping cabinets. In large-scale battery swapping station scenarios, multiple battery swapping cabinets are often arranged side by side. The distributed data consistency controller is configured to share abnormal behavior feature models among multiple cabinets. When a battery swapping cabinet detects a new and previously unseen malicious operation torque pattern, the controller is responsible for encrypting the feature vector and synchronizing it to other cabinets within the local area network, enabling the entire battery swapping station cluster to have a collaborative evolutionary defense capability.

[0039] The multi-sensor fusion bus is configured as a high-speed, high-reliability industrial fieldbus for integrating auxiliary sensing data beyond the six-axis force sensor. In Embodiment 2, the system also includes an array of tactile sensors, a triaxial accelerometer, and a near-field voiceprint recognition unit connected to the multi-sensor fusion bus. The array of tactile sensors is installed in the inlet lining of the loading module to acquire pressure distribution images when the battery surface contacts the cabinet. The triaxial accelerometer is installed at the center of gravity of the battery swapping cabinet to monitor the tilt or displacement of the overall cabinet. The near-field voiceprint recognition unit is used to capture specific frequency sound waves generated by metal impacts or plastic breakage.

[0040] In Embodiment 2, the pattern recognition analysis unit is configured to execute a multimodal fusion algorithm. This algorithm employs a delayed fusion strategy, whereby the force / torque features, tactile distribution image features, and voiceprint features are evaluated independently with varying confidence levels. The final behavior determination is then derived through a decision-level fusion model (such as weighted voting or DS evidence theory). For example, when the six-axis force sensor detects a large radial torque, the tactile array shows pressure points concentrated at the edge of the guide groove, and the voiceprint unit captures a high-frequency sharp sound of metal friction, the system will determine the current behavior as "structural interference caused by forceful insertion" based on confidence level and trigger a corresponding protective termination procedure.

[0041] In Example 2, the biomechanical modeling unit is configured to possess self-learning and self-evolution capabilities. The modeling unit receives large-scale user behavior samples from a cloud-based scheduling platform and uses federated learning technology to update the human operational biomechanics model while protecting user privacy. The self-learning process includes dynamic correction of the baseline torque curve: when the system observes that the operational torque of most normal users deviates from the original baseline under specific climatic conditions (such as stiff movements due to wearing heavy clothing in winter), the modeling unit automatically widens the threshold range for stability determination to reduce the false alarm rate caused by seasonal environmental changes.

[0042] In Embodiment 2, the behavior determination unit is configured to have predictive analysis capabilities. It integrates a Long Short-Term Memory (LSTM) network specifically designed to process temporally correlated torque data streams. The LSTM network can identify precursory features of abnormal behavior; for example, unstable shaking and tentative pushing and pulling often precede actual violent impacts. By capturing these subtle precursory behaviors, the behavior determination unit can issue warnings before the destructive action is fully completed.

[0043] In Embodiment 2, the environmental compensation unit further integrates structural health monitoring functionality. By long-term tracking and recording of the reference output changes of the six-axis force sensor under no-load conditions, the environmental compensation unit can assess the fatigue level of the mechanical structure of the loaded module. When structural loosening or deformation occurs due to long-term use, causing changes in force transmission characteristics, the environmental compensation unit automatically generates compensation gain for the sensor and sends a "sub-healthy mechanical structure" maintenance suggestion to the operation and maintenance platform, achieving refined management of hardware lifespan.

[0044] In Embodiment 2, the early warning response unit is configured with interactive functionality. In addition to local locking and alarms, the early warning response unit also interacts in real-time with the user's mobile terminal. When the system determines that the user has engaged in improper operation rather than malicious damage, the early warning response unit will push a short video or 3D animation demonstration of the correct operating posture for that position to the user's mobile app, and guide the user on how to adjust the force angle. This user-friendly interactive mechanism ensures equipment safety while improving the user experience and reducing customer complaints caused by operational errors.

[0045] Example 3: In this example 3, a pattern recognition-based abnormal access behavior detection system for battery swapping cabinets is provided, optimized for extremely cold and high-dust outdoor environments. It aims to solve the problem of misjudgment caused by decreased sensor sensitivity and mechanical jamming under extreme working conditions.

[0046] The system architecture, while inheriting the core units of Embodiments 1 and 2, has been enhanced with specific improvements to the hardware implementation details.

[0047] The loading module of the battery swapping cabinet adopts a double-sealed structure, and a constant temperature heating device and a nano-dehumidification membrane are configured in the mounting chamber of the six-axis force / torque sensing unit. The constant temperature heating device is controlled by the environmental compensation unit and is configured to start when the external ambient temperature is below zero degrees Celsius, ensuring that the strain gauge inside the force sensor operates within a constant temperature compensation range and eliminating false torque signals caused by thermal expansion and contraction of materials.

[0048] The six-axis force / torque sensing unit in Embodiment 3 employs a full-bridge complementary circuit topology. In this structure, the measurement of each force component consists of four active strain gauges distributed on the symmetry plane of the elastic body. This configuration automatically cancels common-mode interference signals (such as uniform temperature rise or electromagnetic pulse noise) and amplifies only the differential-mode signal (i.e., user-interactive force) generated by asymmetric operation. The sensor surface is coated with an ultra-thin fluorocarbon protective layer to prevent chemical corrosion of the precision sensing element by outdoor high-salt spray environments.

[0049] In Example 3, the biomechanical modeling unit incorporates a mechanical impedance matching model. This model not only considers the force characteristics of the human body but also incorporates the changes in the physical properties of the battery swapping cabinet loading module under extreme environments (such as increased motion resistance due to increased viscosity of lubricating grease at low temperatures). The modeling unit is configured to adjust the slope parameter of the reference torque curve in real time, enabling it to automatically compensate for the increase in normal frictional torque caused by mechanical aging or poor lubrication. In this way, the system can distinguish between "laborious operation due to low temperature" and "forceful pushing and pulling by humans," improving the system's usability in harsh environments.

[0050] In Embodiment 3, the pattern recognition and analysis unit is configured to employ a dual-path recognition logic. The first path is a fast threshold comparison path, primarily used to capture instantaneous impacts where the peak value exceeds the physical safety limit. The second path is a deep feature extraction path, employing a convolutional neural network (CNN) to identify the spatiotemporal spectrum of the torque signal. The spatiotemporal spectrum transforms the six-dimensional force / torque signal into a two-dimensional matrix resembling a color image, with the horizontal axis representing time, the vertical axis representing sensor channels, and color intensity representing signal strength. Through this image-based processing, the system can identify extremely subtle abnormal patterns, such as attempts to crack electronic locks using vibration tools at specific frequencies. These behaviors are extremely difficult to detect using traditional numerical threshold judgments, but exhibit texture features in the spectrum.

[0051] In Embodiment 3, the behavior determination unit is configured with multi-source verification logic. It is deeply coupled with the battery management system (BMS) and access control system inside the battery swapping cabinet. When the behavior determination unit initially identifies a potential abnormal access signal, it queries the BMS to verify whether the current battery communication status is abnormal. If abnormal torque is accompanied by communication interruption, the system will determine it as "physical forced disassembly"; if abnormal torque but normal communication and the battery is undergoing a compliant unlocking process, the system will tend to determine it as "user compensation force caused by mechanical jamming," and will adopt different warning strategies.

[0052] In Embodiment 3, the early warning response unit adds a physical isolation response mode. For high-risk malicious acts, the early warning response unit is configured to control a rapidly inflating protective airbag or an emergency physical blocking pin. Upon detecting a heavy tool impact, the system can deploy the protective device within tens of milliseconds to protect the battery cells and internal high-voltage wiring from puncture, preventing secondary disasters such as fires or explosions caused by malicious damage.

[0053] In Embodiment 3, the environmental compensation unit is further expanded into a system self-healing center. It not only monitors the external environment but also monitors the impedance changes of the sensors themselves. If a poor connection or abnormally increased resistance is detected in the signal line of a force sensor, the environmental compensation unit automatically switches to a redundant backup channel or reconstructs the lost axial force information using the remaining force signals through a geometric projection algorithm. This fault-tolerant mechanism ensures that the system can maintain basic safety monitoring functions even if some components are damaged.

[0054] The installation of the six-axis force / torque sensing unit employs a pre-stress locking technique. During installation, an initial, known compressive load is applied to the sensor using specific tooling. The environmental compensation unit uses this pre-stress as the system's zero-point reference. This approach ensures that the sensor remains under stress even when the user pulls it out (generating tensile stress), avoiding mechanical dead zones during force direction switching and resulting in high symmetry and linearity in the system's sensing of thrust and tension.

[0055] In embodiment 3, the pattern recognition and analysis unit also integrates an unsupervised learning module based on an auto-encoder. This unsupervised learning module continuously learns low-dimensional representations of normal operational data streams during the daily operation of the battery swapping cabinet. Once the reconstruction error of the input force / torque data after encoding and decoding exceeds a set dynamic threshold, the system considers an "unknown type of anomaly" to have occurred. This mechanism allows the system to remain vigilant against newly emerging attack methods without needing to have seen all types of destructive tactics beforehand.

[0056] The early warning response unit is also linked to the on-site real-time video monitoring system. When the behavior determination unit confirms abnormal behavior, the early warning response unit triggers the camera's high frame rate recording mode and directs the pan-tilt unit to rotate to the specific location where the abnormality occurred for close-up capture. The captured high-definition image and the corresponding torque anomaly characteristic curve are packaged into a security event package and uploaded to the operator's evidence preservation server, providing a complete chain of evidence for subsequent legal recourse.

[0057] Example 4: This example 4 provides an abnormal access behavior detection system for battery swapping cabinets based on digital twin technology. By constructing a digital mapping of physical entities in the cloud, it achieves in-depth insight into user behavior and full lifecycle management.

[0058] In addition to the basic unit in Embodiment 1, the system also includes a cloud-based digital twin server, a real-time physical simulation engine, and a user credit evaluation module.

[0059] All physical parameters of the battery swapping cabinet, including material elastic modulus, structural stiffness, and hinge friction coefficient, are established in a high-precision three-dimensional mathematical model on a cloud-based digital twin server. The data stream collected by the six-axis force / torque sensing unit is synchronized to the cloud-based digital twin server in real time via a high-bandwidth 5G network.

[0060] The real-time physical simulation engine is configured to reproduce every user access action in virtual space. When the physical cabinet receives torque signals, the simulation engine calculates a stress distribution cloud map of the cabinet structure based on these force vectors. The pattern recognition and analysis unit performs analysis using these virtual stress features in the cloud. Unlike the local analysis that only analyzes the values ​​of sensor points, the cloud-based digital twin analysis can assess the stress health status of the entire loading module and identify long-term structural fatigue caused by incorrect force application angles, even if the instantaneous failure threshold has not been reached.

[0061] In Example 4, the biomechanical modeling unit is configured as a personalized model. Based on the user's historical battery swapping habits, the system creates a unique mechanical operation profile for each high-frequency user. The baseline torque curve is then fine-tuned to reflect the user's operating habits. For example, for a right-handed user who applies significant force, the system learns their unique torque bias and does not flag them as abnormal. This deep, customized recognition based on individual differences reduces the system's false alarm rate to a low level.

[0062] In embodiment 4, the behavior determination unit incorporates a user credit evaluation module. This module stores the user's historical compliance records. When the behavior determination unit faces ambiguous judgments (i.e., data bordering on abnormal or normal), it refers to the user's credit score. If the user has good credit, the system will prioritize identifying it as an operational error and issue a friendly reminder; if the user has multiple violation records, the system will increase the warning level and strengthen monitoring.

[0063] In Embodiment 4, the early warning response unit is configured to have remote expert intervention capabilities. Upon detecting extremely high-risk and complex abnormal behavior, the system automatically pushes real-time torque data waveforms, on-site video streams, and digital twin simulation results to the expert console at the remote operations and maintenance center. Experts can use remote VR devices to "immerse" themselves in the changes in torque applied by the user and make a final human decision. This human-machine collaborative judgment mode solves the logical blind spots that pure algorithms may have when facing complex intentions of human sabotage.

[0064] In Embodiment 4, the environmental compensation unit utilizes big data prediction technology to anticipate the impact of drastic weather changes on sensor accuracy. By accessing a weather forecast API, the environmental compensation unit can predict upcoming cold waves or heavy rainfall and activate the internal microenvironment control system in advance, adjusting the sensitivity coefficient of the algorithm to ensure the system always operates in optimal condition.

[0065] The six-axis force / torque sensing unit employs fiber optic grating (FBG) sensing technology as an alternative. In industrial areas with strong electromagnetic interference, fiber optic sensors are immune to all electrical noise, sensing force and torque by monitoring the drift of reflected light wavelengths. The environmental compensation unit achieves ultra-high precision sensing of physical interaction forces in complex electromagnetic environments through real-time demodulation of the fiber optic center wavelength.

[0066] The pattern recognition and analysis unit generates a massive amount of virtual attack samples in the cloud using Generative Adversarial Networks (GANs). By simulating various extreme and rare methods of violent destruction (such as using hydraulic shears, electric drills, etc.), GANs can train pattern recognition algorithms to identify anomalous behaviors that rarely occur in reality but have serious consequences. This adversarial training based on generative AI enriches the system's anomalous feature library.

[0067] The behavior determination unit is configured to output a "device health loss score". For each abnormal access behavior, regardless of whether an alert is triggered, the system calculates the degree of wear and tear on the physical lifespan of the loading module. When the accumulated wear and tear reaches a critical value, the alert response unit automatically generates a repair request and assigns maintenance personnel to perform preventative component replacement, thus avoiding accidents caused by structural fatigue.

[0068] The early warning response unit also has social governance functions. After determining that the behavior is malicious, the system will automatically synchronize the user's violation to the industry blacklist system, restricting their battery swapping service privileges throughout the city and even the country, thus establishing an industry constraint mechanism through technical means.

[0069] In summary, the pattern recognition-based abnormal access behavior detection system for battery swapping cabinets provided by this invention achieves deep perception of user access intentions by integrating a high-precision six-axis force / torque sensing unit into the loading module and combining advanced biomechanical modeling, multimodal pattern recognition, and edge computing technologies. The system eliminates the dependence on ambient lighting and macroscopic displacement found in traditional solutions, reaching the essential mechanical characteristics of physical interaction. Through the collaboration of an environmental compensation unit and a self-learning modeling unit, this invention can distinguish between normal operating habits, environmental disturbances, and genuine malicious damage, reducing the false alarm rate. The tiered early warning and safety response mechanism, combined with digital twins and predictive analytics, provides comprehensive safety protection for battery swapping cabinets, ensuring the long-term, stable, and efficient operation of battery swapping infrastructure in complex outdoor environments, demonstrating engineering application value and socio-economic benefits.

[0070] Although the present invention has been described in detail above with reference to the embodiments, the present invention is not limited to the specific embodiments described above. Those skilled in the art will be able to make several improvements and modifications without departing from the spirit and scope of the present invention. These improvements and modifications should also be considered within the scope of protection of the present invention. For example, the six-axis force sensing unit can be further combined with other biometric technologies (such as fingerprint and facial recognition), or the detection system can be applied to automated battery replacement robots. All these structural transformations and functional extensions based on the core concept of the present invention fall within the scope of protection of the claims of the present invention.

Claims

1. A pattern recognition-based system for detecting abnormal access behavior in battery swapping cabinets, characterized in that, include: The battery swapping cabinet body is equipped with a standard loading module for accommodating batteries. The standard loading module is configured to allow users to perform battery insertion or removal operations and serves as a physical interface to support the user's operating force. A six-axis force / torque sensing unit is embedded inside the standard loading module and is configured to collect the three-dimensional force components and three-dimensional torque components acting on the standard loading module in real time when the user performs battery storage and retrieval operations, forming a force / torque vector data stream; The biomechanical modeling unit is communicatively connected to the six-axis force / torque sensing unit and is configured to construct a human operation mechanics model based on the biomechanical characteristics of human upper limb operation, and generate a reference torque curve representing compliant operation based on historical operation data. The pattern recognition and analysis unit is connected to the six-axis force / torque sensing unit and the biomechanical modeling unit, respectively, and is configured to receive the force / torque vector data stream, compare it with the reference torque curve, and identify data features that deviate from the normal operating mode. The behavior determination unit is connected to the pattern recognition and analysis unit and is configured to determine the behavior classification result of the current operation based on the identified data features. The early warning response unit is connected to the behavior determination unit and is configured to trigger the corresponding security response mechanism after receiving the high-risk behavior classification result.

2. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 1, characterized in that, The six-axis force / torque sensing unit is installed at the structural connection between the standard loading module and the battery swapping cabinet body, and the six-axis force / torque sensing unit includes an elastic body made of high-strength alloy steel material, on which at least three sets of mutually perpendicular strain gauge bridges are symmetrically arranged. The six-axis force / torque sensing unit also includes a built-in instrumentation amplifier and an analog-to-digital converter, wherein the instrumentation amplifier is configured to amplify the bridge unbalance voltage generated by the strain gauge bridge circuit, and the analog-to-digital converter is configured to convert the amplified voltage signal into a digital signal. The three-dimensional force components include horizontal axial force, horizontal transverse force, and vertical force distributed along three mutually perpendicular coordinate axes; the three-dimensional torque components include torsional torque distributed around the three coordinate axes. The sampling frequency of the six-axis force / torque sensing unit is set to no less than 1000 Hz to capture the characteristics of transient shock waves generated by violent impacts.

3. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 2, characterized in that, The biomechanical modeling unit constructs the human operation mechanics model by simulating the human upper limb dynamic chain. The human upper limb dynamic chain simplifies the user's hand, wrist, forearm and upper arm into a series of rigid bodies connected by joints, and sets the upper limit of the maximum output torque of each joint according to the physiological limits of the human body. The baseline torque curve is defined as an envelope interval based on a probability distribution. The biomechanical modeling unit is configured to use Gaussian process regression or Bayesian network to dynamically adjust the boundary of the envelope interval according to different time periods and different user profiles, so that the baseline torque curve can be compatible with the normal operating actions of different user groups. The reference torque curve has the characteristics of stability and linearity. During normal battery insertion or removal, the trajectory of torque change over time is limited to a continuous and gently sloping curve, without any instantaneous spikes or phase abrupt changes.

4. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 3, characterized in that, The pattern recognition analysis unit integrates a support vector machine algorithm, which is configured to map the original six-axis force data to a high-dimensional feature space through a kernel function, and construct an optimal hyperplane in the feature space to separate normal operation vectors from abnormal operation vectors in order to capture the nonlinear coupling relationship between force vectors and torque vectors. The pattern recognition analysis unit also integrates the isolated forest algorithm, which is configured to construct a tree structure by recursively partitioning the feature space and calculate the path length of real-time data points in the tree structure, and determine data points with path lengths less than a preset length threshold as potential abnormal patterns. The pattern recognition and analysis unit is also configured to perform a fast Fourier transform on the torque time series, extract its spectral distribution features, and identify high-frequency characteristic peaks generated by malicious knocking or tool damage by weighted fusion of the time-domain recognition results and the frequency-domain power spectral density analysis results.

5. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 4, characterized in that, The behavior determination unit is equipped with a logic judgment matrix, which is used to determine whether the current operation is a violent plugging or unplugging, malicious impact, lever prying or violent blocking based on the magnitude of the force vector, the rotational vector magnitude of the torque, the rate of energy change and the duration of the abnormal mode. The behavior determination unit is configured to analyze the derivative of force with respect to time. When the torque in the vertical direction changes beyond a step threshold within a preset time interval and is accompanied by an oscillation signal in the horizontal direction, it is determined to be a malicious prying behavior. When the force value along the insertion direction continues to increase and the displacement stops, and the torque components show an asymmetrical distribution, it is determined to be a violent blocking insertion / removal. The behavior determination unit uses a fuzzy logic reasoning system to output a risk index between 0 and 1 based on multiple continuously sampled abnormal features, and compares the risk index with multiple preset intervals to determine the risk level of the current behavior.

6. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 5, characterized in that, The safety response mechanism triggered by the early warning response unit has multi-level linkage characteristics. The early warning response unit includes an operation log recording module for recording the original force / torque waveform, an electromagnetic locking mechanism for locking the moving parts of the standard loading module, a wireless communication interface for sending alarm signals to the remote monitoring center, and a buzzer and red flashing light for activating on-site audible and visual warnings. The early warning response unit is configured with hierarchical response logic, and for low-risk anomalies, only the operation log recording module generates maintenance logs. For medium-risk anomalies, a warning dialog box will pop up on the display interface of the battery swapping cabinet and temporarily restrict user operation permissions; for high-risk anomalies, the electromagnetic locking mechanism will immediately perform a physical locking operation and simultaneously notify the maintenance personnel.

7. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 6, characterized in that, The abnormal access behavior detection system for the battery swapping cabinet also includes an environmental compensation unit, which is connected to the signal conditioning circuit of the six-axis force / torque sensing unit and is configured to monitor the ambient temperature, humidity and background vibration of the cabinet. The environmental compensation unit stores a sensor temperature drift compensation model, which is configured to dynamically correct the original force / torque data according to environmental parameters. It corrects the sampled values ​​by addition or subtraction through preset compensation coefficients to eliminate false torque signals caused by thermal expansion and contraction of materials. The six-axis force / torque sensing unit is installed using prestressed locking technology. During installation, an initial compressive load is applied to the six-axis force / torque sensing unit using a tooling. The environmental compensation unit uses the initial compressive load as the zero-point reference of the system to eliminate the mechanical dead zone when the force direction changes.

8. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 7, characterized in that, The abnormal access behavior detection system for the battery swapping cabinet also includes an edge processing node and a distributed data consistency controller. The battery swapping cabinet body includes multiple intelligent compartment units with independent computing capabilities, and each intelligent compartment unit is independently configured with a set of the six-axis force / torque sensing units. The edge processing node is deployed inside the battery swapping cabinet body and includes a neural network acceleration unit. It is configured to run a lightweight deep residual network model to complete feature extraction and preliminary pattern filtering at the edge. The distributed data consistency controller is connected to multiple edge processing nodes and configured to share abnormal behavior feature models among multiple battery swapping cabinets. After encrypting the feature vector of newly detected malicious operation torque patterns, it is synchronized to other cabinets in the local area network.

9. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 8, characterized in that, The abnormal access behavior detection system for the battery swapping cabinet also includes an array of tactile sensors, a triaxial accelerometer, and a near-field voiceprint recognition unit connected via a multi-sensor fusion bus. The array-type tactile sensor is laid in the inlet liner of the standard loading module to acquire pressure distribution images when the battery surface contacts the cabinet. The triaxial accelerometer is installed at the center of gravity of the battery swapping cabinet body to monitor the overall displacement of the cabinet. The near-field voiceprint recognition unit is used to capture frequency sound waves generated by metal impact; the pattern recognition analysis unit is configured to execute a multimodal fusion algorithm, and evaluate the confidence of force / torque features, pressure distribution image features and voiceprint features through a decision layer fusion model to obtain the final behavior judgment result.

10. The abnormal access behavior detection system for battery swapping cabinets based on pattern recognition according to claim 9, characterized in that, The abnormal access behavior detection system for the battery swapping cabinet also includes a cloud-based digital twin server and a user credit evaluation module. The cloud-based digital twin server stores a three-dimensional mathematical model of the battery swapping cabinet body, including the material's elastic modulus, structural stiffness, and coefficient of friction, and integrates a real-time physical simulation engine. The real-time physical simulation engine is configured to calculate the stress distribution cloud map of the cabinet structure based on the real-time collected torque signals, and to evaluate the long-term structural fatigue state of the standard loading module. The user credit evaluation module is configured to store the user's historical compliance records, and the behavior determination unit is configured to call the credit score in the user credit evaluation module and dynamically adjust the warning level when the behavior characteristics are in a critical state. The pattern recognition and analysis unit uses a generative adversarial network in the cloud to simulate virtual attack samples and conduct adversarial training on the abnormal feature database.

Images