An AI-based battery anti-theft system for battery swapping cabinets

Abstract

The present invention discloses a battery anti-theft system for a battery swapping cabinet based on artificial intelligence, which relates to the technical field of anti-theft. In the present invention, the user identity is verified through a face recognition module, and the operation legitimacy is dynamically evaluated in combination with behavior analysis technology, and a unique user operation identifier is generated, thus preventing unauthorized use from the source. When an abnormality is recognized, the system immediately restricts the user's permissions or triggers multi-factor authentication to further strengthen the security protection. The sensor module collects vibration, displacement and ambient light data during the process of battery removal and placement in real time, and transmits them to the central control unit through the Internet of Things module. The central control unit uses an artificial intelligence model to jointly analyze the sensor data and the surrounding environment features obtained by the camera, extracts the dynamic features of multi-modal time-series data through a bidirectional recurrent neural network, and accurately evaluates the behavior legitimacy in combination with a classification algorithm. If an abnormal behavior is confirmed, an emergency response is triggered according to the threat level.

Description

Technical Field

[0001] The present invention relates to the technical field of anti-theft, and in particular to an anti-theft system for the batteries of battery swapping cabinets based on artificial intelligence. Background Art

[0002] In high-density urban areas, in order to cover more electric vehicles, the distribution of user battery swapping cabinets is relatively compact; in high-density urban areas, the flow of people is large and complex, the usage frequency of battery swapping cabinets is high, and the operator needs to cope with high-intensity equipment requirements and potential security threats at the same time. Since battery swapping cabinets are mostly deployed in open environments, such as community entrances and exits, business districts or street locations, the anti-theft pressure is particularly prominent.

[0003] The traditional anti-theft methods for the batteries of battery swapping cabinets mainly rely on mechanical locks, electronic locks and RFID identity recognition systems. The original intention of such methods is to restrict unauthorized personnel from operating and to ensure the safety of the batteries as much as possible. However, in the high-density area usage environment, mechanical locks and electronic locks are easily damaged by violence, and there are also risks such as forgery and cracking in RFID technology; in addition, high-frequency equipment operations will lead to an increase in the workload of maintenance. When the equipment fails to be repaired in time due to a fault, it is more likely to become a security vulnerability.

[0004] Some traditional solutions address these problems by strengthening the lock material, increasing the complexity of the electronic lock, and enhancing the inspection frequency. However, these countermeasures are essentially "passive defenses", lacking pertinence and initiative. The inspections require a large amount of human resources and cannot achieve real-time response. In addition, increasing the complexity of the lock still has limitations in the face of malicious damage. Therefore, there is an urgent need for an anti-theft solution for the batteries of battery swapping cabinets based on artificial intelligence to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an anti-theft system for the batteries of battery swapping cabinets based on artificial intelligence to solve the problems that in high-density urban areas, the distribution of battery swapping cabinets is compact but the theft risk is high, the number of battery swapping cabinets in urban intensive areas is large, the mobile population is complex, and there may be situations of rapid battery theft or malicious damage.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] An embodiment of the present invention provides an anti-theft system for the batteries of battery swapping cabinets based on artificial intelligence, which includes,

[0009] A face recognition module, with a built-in camera, is used to verify the identity of the user, and dynamically evaluate the legality of the operation in combination with behavior analysis technology, and generate a user operation identifier;

[0010] The sensor module is used to collect vibration, displacement and ambient light data during battery placement in real time;

[0011] The IoT module is used to transmit the data collected by the sensors to the central control unit.

[0012] The central control unit receives sensor data and uses artificial intelligence models to perform multi-dimensional analysis, identifying abnormal behavior by combining surrounding environmental characteristics and triggering emergency response;

[0013] The management center is used to receive abnormal event data, coordinate with the security department to respond, and store all battery replacement operation records and abnormal events through blockchain technology.

[0014] As a preferred solution of the battery anti-theft system for battery swap cabinet based on artificial intelligence described in the present invention, the battery swapping method of the battery anti-theft system for battery swap cabinet based on artificial intelligence is:

[0015] Step S1, obtain user facial data through the facial recognition module embedded in the battery swap cabinet, use the convolutional neural network CNN to perform real-time facial feature extraction and comparison based on the user facial data, verify the user identity, and authorize after successful verification, and generate a unique user operation identifier;

[0016] Step S2: a sensor module is integrated inside the battery swap cabinet, including a vibration sensor, a displacement sensor and a light sensor, to collect battery access process data, including vibration, displacement and ambient light data, and the collected access process data is transmitted to the central control unit through the Internet of Things module;

[0017] Step S3, using the camera to obtain the surrounding environment characteristics, and using the recurrent neural network (RNN) to jointly analyze the pick-and-place process data and the surrounding environment characteristics to identify abnormal behaviors, including illegal tool operation, multi-person gathering, and unauthorized operation;

[0018] Step S4, after confirming the existence of abnormal behavior, trigger the emergency response mechanism, and classify the abnormal behavior at the same time, and different threat levels trigger different loudness alarms.

[0019] As a preferred solution of the artificial intelligence-based battery anti-theft system for battery swap cabinets described in the present invention, in which: in step S1, when facial recognition detects an identity mismatch, user permissions are immediately restricted or additional multi-factor verification is required.

[0020] As a preferred solution of the battery anti-theft system for battery swap cabinets based on artificial intelligence described in the present invention, the steps of extracting and comparing facial features in real time using a convolutional neural network CNN based on user facial data and verifying the user's identity are as follows:

[0021] Obtain the user's real-time facial image through the built-in camera, and set the image as:

[0022] ,

[0023] where, represents the original image collected, represents the real number field, represents the image height, represents the image width, represents the number of image channels;

[0024] Perform grayscale processing on to generate a grayscale image , and perform normalization. The normalization formula is:

[0025] ,

[0026] where, represents the grayscale image, represents the pixel mean of the grayscale image, represents the pixel standard deviation of the grayscale image, represents the normalized grayscale image;

[0027] Input the normalized grayscale image into the convolutional neural network CNN. The used CNN consists of three convolutional layers and two fully connected layers. Among them, the output feature map of each layer of convolution is:

[0028] ,

[0029] where, represents the feature map after the th layer of convolution, represents the convolutional kernel of the th layer, represents the bias of the th layer, represents the convolution operation, represents the activation function;

[0030] The convolution output feature is dimension-reduced through the max pooling operation. Set the pooling window as . The pooling operation formula is:

[0031] ,

[0032] where, represents the feature after pooling, represents the feature map of the last layer of convolution, represents the pooling window size,

[0033] The pooled features are input into a fully connected layer for classification processing to generate a feature vector :

[0034] ,

[0035] Among them, represents the weight matrix of the fully connected layer, represents the bias of the fully connected layer, represents the finally extracted feature vector;

[0036] The extracted feature vector is calculated for cosine similarity with the feature vectors of registered users in the database The calculation formula is:

[0037] ,

[0038] Among them, represents the similarity between the extracted feature and the reference feature, represents the reference user feature vector, and are the norms of the feature vectors respectively;

[0039] If the similarity exceeds the set threshold , the verification passes and a unique user identifier :

[0040] ,

[0041] Among them, represents the unique operation identifier of the user, represents the hash generation function.

[0042] As a preferred solution of the battery anti-theft system for battery swapping cabinets based on artificial intelligence according to the present invention, wherein: the surrounding environment features include the density of people flow and the usage of tools.

[0043] As a preferred solution of the battery anti-theft system for battery swapping cabinets based on artificial intelligence according to the present invention, wherein: all battery swapping operation and abnormal event data are encrypted and stored through blockchain technology.

[0044] As a preferred solution of the battery anti-theft system for battery swapping cabinets based on artificial intelligence according to the present invention, wherein: the steps of using a recurrent neural network (RNN) to jointly analyze the data of the taking and placing process and the surrounding environment features to identify abnormal behaviors are as follows,

[0045] Obtain real-time data of the taking and placing process through the sensor module, including vibration data , displacement data , ambient light data , and surrounding environment features collected by the camera, including crowd density and tool usage , combine all the data into a joint time series input :

[0046] ,

[0047] Among them, represents the vibration data value at time step , represents the displacement data value at time step , represents the ambient light data value at time step , represents the crowd density feature at time step , represents the tool usage feature at time step , represents the total length of the time series,

[0048] Use a bidirectional recurrent neural network Bi - RNN to extract features from the sequence data. Let the hidden state of the forward RNN be , and the hidden state of the backward RNN be , and the joint output is:

[0049] ,

[0050] Among them, represents the hidden state of the forward RNN at time step , represents the hidden state of the backward RNN at time step , represents the joint hidden state, containing forward and backward feature information,

[0051] The update formula for the single - step hidden state is:

[0052] ,

[0053] Among them, represents the hidden state of the forward RNN at time step , represents the joint input feature at time step , represents the weight matrix of the forward hidden state, represents the weight matrix of the forward input feature, represents the forward bias term, Denote the hidden state of the backward RNN at time step , Denote the weight matrix of the backward hidden state, Denote the weight matrix of the backward input features, Denote the backward bias term, Denote the tanh activation function;

[0054] Average and aggregate the hidden states for all time steps to obtain the global feature representation :

[0055] ,

[0056] where, Denote the global representation vector of the joint features, Denote the total length of the time series, Denote the joint hidden state at time step ;

[0057] Input the global feature representation into the fully connected layer and use the Softmax function for classification:

[0058] ,

[0059] where, Denote the probability distribution of the abnormal behavior classification result, Denote the weight matrix of the fully connected layer, Denote the bias term of the fully connected layer;

[0060] According to the classification result , select the category with the highest probability as the prediction result. If the probability exceeds the preset threshold, it is determined as an abnormal behavior.

[0061] As a preferred solution of the battery anti-theft system for battery swapping cabinets based on artificial intelligence according to the present invention, wherein: the emergency response mechanism includes remotely locking the battery compartment door and activating a high-pitched alarm, and at the same time pushing the abnormal event data to the management center in real time.

[0062] As a preferred solution of the battery anti-theft system for battery swapping cabinets based on artificial intelligence according to the present invention, wherein: after confirming the existence of abnormal behavior, the step of triggering the emergency response mechanism and grading the abnormal behavior is,

[0063] According to the probability value of the classification result in step S3 and the threat level corresponding to the category, define the abnormal behavior threat value :

[0064] ,

[0065] Among them, represents the threat value of abnormal behavior, represents the probability value of the th type of behavior in the classification result, represents the predefined threat weight of the th type of behavior,

[0066] Set the grading threshold of the threat value , and map to different threat levels :

[0067] If , then ,

[0068] If , then ,

[0069] …

[0070] If , then ,

[0071] Among them, represents the threat level, represents the threshold sequence of threat grading;

[0072] According to the threat level , select the corresponding emergency response measures. Let the emergency response matrix be , where the rd row represents the set of emergency actions with the threat level of :

[0073] ,

[0074] If , then execute the response action , and the specific measures include the following operations:

[0075] Remotely lock the battery compartment door,

[0076] Activate the high-pitched alarm,

[0077] And push the abnormal event data to the management center for encrypted storage through the blockchain.

[0078] As a preferred solution of the battery anti-theft system for battery swapping cabinets based on artificial intelligence according to the present invention, among them: The method of encryption through the blockchain is:

[0079] Organize the abnormal event data into a transaction , including the timestamp , threat level , user operation identifier , abnormal behavior description :

[0080] ,

[0081] Among them, represents the abnormal event data transaction, represents the timestamp when the event occurred, represents the threat level, represents the unique identifier of the user, represents the description information of the abnormal behavior,

[0082] will broadcast to the blockchain network to generate a new block :

[0083] ,

[0084] Among them, represents the new block, represents the hash function, represents the current transaction data, represents the hash value of the previous block;

[0085] According to the abnormal event data received by the management center, the security department is linked to perform corresponding operations, and the system monitors in real time the changes, and append stronger emergency responses when the threat level increases.

[0086] The beneficial effects of the present invention are as follows: In the present invention, the user identity is verified through the face recognition module, the operation legitimacy is dynamically evaluated by combining the behavior analysis technology, and a unique user operation identifier is generated, which eliminates unauthorized use from the source. When an abnormality is recognized, the system immediately restricts the user's permissions or triggers multi-factor authentication to further strengthen the security protection. The sensor module collects vibration, displacement, and ambient light data during the battery placement and removal process in real time and transmits it to the central control unit through the Internet of Things module. The central control unit uses an artificial intelligence model to jointly analyze the sensor data and the surrounding environment characteristics obtained by the camera to identify abnormal behaviors such as illegal tool operations, multiple people gathering, and unauthorized operations. The system extracts the dynamic characteristics of multi-modal time-series data through a bidirectional recurrent neural network and combines classification algorithms to accurately evaluate the behavior legitimacy. If an abnormal behavior is confirmed, an emergency response is triggered according to the threat level, including remotely locking the battery compartment door, activating the alarm, and pushing the event data to the management center. At the same time, all battery swapping operation records and abnormal events are encrypted and stored through blockchain technology to ensure the data cannot be tampered with and provide a basis for subsequent tracking;

[0087] Furthermore, the management center can work with the security department to respond to threats in real time and dynamically adjust protection measures.

[0088] The present invention overcomes the problems of traditional mechanical locks and electronic locks being easily damaged and the risk of counterfeiting in RFID technology. At the same time, it avoids the problem of inspection relying on manpower through active monitoring and real-time response, and reduces maintenance costs in high-frequency usage scenarios. It also uses the deep combination of artificial intelligence and blockchain technology to achieve efficient, accurate and dynamic battery anti-theft management. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0090] Figure 1 It is a schematic diagram of the framework of the battery anti-theft system for battery swap cabinets based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0091] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0092] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0093] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0094] Example 1, reference Figure 1 This embodiment provides an artificial intelligence-based battery anti-theft system for a battery swap cabinet, including:

[0095] Facial recognition module, with built-in camera, is used to verify the user's identity and dynamically evaluate the legitimacy of operations in combination with behavioral analysis technology to generate a user operation identifier;

[0096] The sensor module is used to collect vibration, displacement and ambient light data during battery placement in real time;

[0097] An Internet of Things module for transmitting data collected by sensors to a central control unit,

[0098] A central control unit for receiving sensor data, performing multi-dimensional analysis using an artificial intelligence model, identifying abnormal behaviors by combining surrounding environmental characteristics, and triggering an emergency response;

[0099] A management center for receiving abnormal event data, coordinating with the security department for response, and storing all battery replacement operation records and abnormal events through blockchain technology;

[0100] This embodiment also discloses a battery replacement method for an electric cabinet battery anti-theft system based on artificial intelligence, including:

[0101] Step S1: Obtain user facial data through a facial recognition module embedded in the battery replacement cabinet, perform real-time facial feature extraction and comparison on the user facial data using a convolutional neural network (CNN) to verify the user's identity. After successful verification, authorization is granted, and at the same time, a unique user operation identifier is generated;

[0102] In step S1, when facial recognition detects a mismatch in identity, the user's permissions are immediately restricted or additional multi-factor verification is required;

[0103] The step of verifying the user's identity by performing real-time facial feature extraction and comparison on the user facial data using a convolutional neural network (CNN) is as follows:

[0104] Obtain the user's real-time facial image through a built-in camera. Let the image be:

[0105] ,

[0106] Among them, represents the original image collected, represents the real number field, represents the image height, represents the image width, represents the number of image channels;

[0107] Perform grayscale processing on to generate a grayscale image , and perform normalization. The normalization formula is:

[0108] ,

[0109] Among them, represents the grayscale image, represents the pixel mean of the grayscale image, represents the pixel standard deviation of the grayscale image, represents the normalized grayscale image;

[0110] The normalized grayscale image is input into a convolutional neural network (CNN). The used CNN consists of three convolutional layers and two fully connected layers. The output feature map of each convolutional layer is as follows:

[0111] ,

[0112] where represents the feature map after the -th convolutional layer, represents the convolutional kernel of the -th layer, represents the bias of the -th layer, represents the convolution operation, represents the activation function;

[0113] The convolutional output feature is dimension-reduced through a max pooling operation. Let the pooling window be . The pooling operation formula is:

[0114] ,

[0115] where represents the feature after pooling, represents the feature map of the last convolutional layer, represents the pooling window size,

[0116] The pooled feature is input into the fully connected layer for classification processing to generate a feature vector :

[0117] ,

[0118] where represents the weight matrix of the fully connected layer, represents the bias of the fully connected layer, represents the finally extracted feature vector;

[0119] The extracted feature vector is calculated with the feature vector of the registered users in the database for cosine similarity. The calculation formula is:

[0120] ,

[0121] where represents the similarity between the extracted feature and the reference feature, represents the reference user feature vector, and are the norms of the feature vectors respectively;

[0122] If the similarity exceeds the set threshold , the verification passes and a unique user identifier is generated :

[0123] ,

[0124] Among them, represents the unique operation identifier of the user, represents the hash generation function;

[0125] Specifically, this step performs real-time facial feature extraction through a convolutional neural network. In specific operations, the convolutional layer is used to extract local features of the image, and the fully connected layer performs dimensionality reduction and classification processing on the global features. In the feature comparison stage, the cosine similarity is used to determine the identity legitimacy to ensure the verification accuracy; and a unique identifier is generated through feature hashing, taking into account both security and efficiency.

[0126] Step S2, an internal integrated sensor module is installed in the battery swapping cabinet, including a vibration sensor, a displacement sensor, and a light sensor, to collect data during the battery taking and placing process, including vibration, displacement, and ambient light data, and transmit the collected taking and placing process data to the central control unit through the Internet of Things module;

[0127] Step S3, obtain the surrounding environment features through a camera, and use a recurrent neural network (RNN) to jointly analyze the taking and placing process data and the surrounding environment features to identify abnormal behaviors, including illegal tool operations, gathering of multiple people, and unauthorized operations;

[0128] The surrounding environment features include the density of the crowd and the usage of tools;

[0129] All battery swapping operation and abnormal event data are encrypted and stored through blockchain technology;

[0130] The steps of using a recurrent neural network (RNN) to jointly analyze the taking and placing process data and the surrounding environment features to identify abnormal behaviors are as follows

[0131] Obtain the real-time data of the taking and placing process through the sensor module, including vibration data , displacement data , ambient light data , and the surrounding environment features collected by the camera, including the density of the crowd and the usage of tools , and combine all the data into a joint time series for input :

[0132] ,

[0133] Among them, Denotes the vibration data value at time step , Denotes the displacement data value at time step , Denotes the ambient light data value at time step , Denotes the crowd density feature at time step , Denotes the tool usage feature at time step , Denotes the total length of the time series,

[0134] Use a bidirectional recurrent neural network Bi-RNN to extract features from the sequence data. Let the hidden state of the forward RNN be , and the hidden state of the backward RNN be . The combined output is:

[0135] ,

[0136] where Denotes the hidden state of the forward RNN at time step , Denotes the hidden state of the backward RNN at time step , Denotes the combined hidden state, containing forward and backward feature information,

[0137] The update formula for the single-step hidden state is:

[0138] ,

[0139] where Denotes the hidden state of the forward RNN at time step , Denotes the combined input feature at time step , Denotes the weight matrix of the forward hidden state, Denotes the weight matrix of the forward input feature, Denotes the forward bias term, Denotes the hidden state of the backward RNN at time step , Denotes the weight matrix of the backward hidden state, Denotes the weight matrix of the backward input feature, Denotes the backward bias term, Denotes the tanh activation function;

[0140] Average aggregate the hidden states for all time steps to obtain the global feature representation :

[0141] ,

[0142] Among them, represents the global representation vector of the combined feature, represents the total length of the time series, represents the time step of the combined hidden state;

[0143] Input the global feature representation into the fully connected layer and use the Softmax function for classification:

[0144] ,

[0145] Among them, represents the probability distribution of the abnormal behavior classification result, represents the weight matrix of the fully connected layer, represents the bias term of the fully connected layer;

[0146] According to the classification result , select the category with the highest probability as the prediction result. If the probability exceeds the preset threshold, it is determined as abnormal behavior;

[0147] Specifically, analyze the multi-modal time series data through a bidirectional recurrent neural network to effectively capture the dynamic relationship between the picking and placing process and the environmental features; the single-step hidden state update formula synthesizes the forward and backward dependencies of the time series, generates a comprehensive representation vector H through global aggregation, and combines the Softmax classification function for anomaly detection, which has high robustness.

[0148] Step S4, after confirming the existence of abnormal behavior, trigger the emergency response mechanism and at the same time classify the abnormal behavior. Different threat levels trigger different loudness alarms;

[0149] The emergency response mechanism includes remotely locking the battery compartment door and activating a high-pitched alarm, and at the same time pushing the abnormal event data to the management center in real time;

[0150] After confirming the existence of abnormal behavior, the steps of triggering the emergency response mechanism and at the same time classifying the abnormal behavior are,

[0151] According to the probability value of the classification result in step S3 and the threat level corresponding to the category, define the threat value of the abnormal behavior :

[0152] ,

[0153] Among them, represents the threat value of the abnormal behavior, Indicates the probability value of the th type of behavior in the classification result, and represents the predefined threat weight of the

[0154] th type of behavior. Set the grading threshold of the threat value , and map to different threat levels :

[0155] If , then ,

[0156] If , then ,

[0157] …

[0158] If , then ,

[0159] Among them, represents the threat level, and represents the threshold sequence for threat grading;

[0160] According to the threat level , select the corresponding emergency response measures. Let the emergency response matrix be , where the th row represents the set of emergency actions for the threat level :

[0161] ,

[0162] If , then execute the response action , and the specific measures include the following operations:

[0163] Remotely lock the battery compartment door,

[0164] Activate the high - pitched alarm,

[0165] And push the abnormal event data to the management center for encrypted storage through the blockchain.

[0166] The method of encryption through the blockchain is:

[0167] Organize the abnormal event data into a transaction , including the timestamp , threat level , user operation identifier , and abnormal behavior description :

[0168] ,

[0169] Among them, represents the abnormal event data transaction, represents the timestamp when the event occurred, represents the threat level, represents the unique identifier of the user, represents the description information of the abnormal behavior,

[0170] will be broadcast to the blockchain network to generate a new block :

[0171] ,

[0172] Among them, represents the new block, represents the hash function, represents the current transaction data, represents the hash value of the previous block;

[0173] According to the abnormal event data received by the management center, the security department is linked to perform corresponding operations, and the system monitors in real time changes, and stronger emergency responses are added when the threat level is increased;

[0174] Specifically, the abnormal behavior is quantitatively evaluated through calculating the threat value and the grading mechanism, and the emergency response strategy is triggered according to the threat level; the blockchain technology is combined to encrypt and store the abnormal event data to ensure data integrity; the real-time linked response mechanism further enhances the dynamic protection ability of the system.

[0175] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based battery anti-theft system for battery swap cabinets, characterized in that: include, Facial recognition module, with built-in camera, is used to verify the user's identity and dynamically evaluate the legitimacy of operations in combination with behavioral analysis technology to generate a user operation identifier; The sensor module is used to collect vibration, displacement and ambient light data during battery placement in real time; The IoT module is used to transmit the data collected by the sensors to the central control unit. The central control unit receives sensor data and uses artificial intelligence models to perform multi-dimensional analysis, identifying abnormal behavior by combining surrounding environmental characteristics and triggering emergency response; The management center is used to receive abnormal event data, link up with the security department to respond, and store all battery swap operation records and abnormal events through blockchain technology; The identifying of abnormal behavior by combining surrounding environment features includes: Get real-time data of the pick-and-place process, including vibration data, through the sensor module , displacement data , ambient light data , the surrounding environment characteristics collected by the camera, including the density of people and tool usage , combining all data into a joint time series input : ， in, Indicates that at time step The vibration data value, Indicates that at time step The displacement data value, Indicates that at time step The ambient light data value, Indicates that at time step The crowd density characteristics, Indicates that at time step Tool usage characteristics, represents the total length of the time series, Bidirectional recurrent neural network Bi-RNN is used to extract features from sequence data. The hidden state of the forward RNN is , the hidden state of the backward RNN is , the combined output is: ， in, Represents the forward RNN at time step The hidden state of Represents the backward RNN at time step The hidden state of Represents the joint hidden state, which contains the forward and backward feature information. The update formula for the single-step hidden state is: ， in, Represents the forward RNN at time step The hidden state of Indicates that at time step The combined input features of represents the weight matrix of the forward hidden state, represents the weight matrix of the forward input features, represents the forward bias term, Represents the backward RNN at time step The hidden state of represents the weight matrix of the backward hidden state, represents the weight matrix of the backward input features, represents the backward bias term, represents the tanh activation function; The hidden state for all time steps Perform average aggregation to obtain global feature representation : ， in, The global representation vector representing the joint features, represents the total length of the time series, Represents the time step The joint hidden state of The global feature representation Enter the fully connected layer and use the Softmax function for classification: ， in, represents the probability distribution of abnormal behavior classification results, represents the weight matrix of the fully connected layer, represents the bias term of the fully connected layer; According to the classification results , the highest probability category is selected as the prediction result, and if the probability exceeds the preset threshold, it is determined to be an abnormal behavior.

2. The battery anti-theft system for battery swap cabinet based on artificial intelligence as claimed in claim 1 is characterized in that: The battery swapping method of the battery anti-theft system for battery swapping cabinet based on artificial intelligence is: Step S1, obtain user facial data through the facial recognition module embedded in the battery swap cabinet, use the convolutional neural network CNN to perform real-time facial feature extraction and comparison based on the user facial data, verify the user identity, and authorize after successful verification, and generate a unique user operation identifier; Step S2: a sensor module is integrated inside the battery swap cabinet, including a vibration sensor, a displacement sensor and a light sensor, to collect battery access process data, including vibration, displacement and ambient light data, and the collected access process data is transmitted to the central control unit through the Internet of Things module; Step S3, using the camera to obtain the surrounding environment characteristics, and using the recurrent neural network (RNN) to jointly analyze the pick-and-place process data and the surrounding environment characteristics to identify abnormal behaviors, including illegal tool operation, multi-person gathering, and unauthorized operation; Step S4, after confirming the existence of abnormal behavior, trigger the emergency response mechanism, and classify the abnormal behavior at the same time, and different threat levels trigger different loudness alarms.

3. The battery anti-theft system for battery swap cabinet based on artificial intelligence as claimed in claim 2 is characterized by: In step S1, when facial recognition detects an identity mismatch, user permissions are immediately restricted or additional multi-factor verification is required.

4. The battery anti-theft system for battery swap cabinet based on artificial intelligence as claimed in claim 3 is characterized by: The steps of extracting and comparing facial features in real time using a convolutional neural network (CNN) based on user facial data and verifying the user's identity are as follows: Get the user's real-time facial image through the built-in camera, assuming the image is: ， in, represents the original image collected. represents the field of real numbers, Indicates the image height, Indicates the image width, Indicates the number of image channels; right Perform grayscale processing to generate a grayscale image , and normalize it. The normalization formula is: ， in, represents a grayscale image, represents the pixel mean of the grayscale image, represents the pixel standard deviation of the grayscale image, Represents the normalized grayscale image; The normalized grayscale image Input convolutional neural network CNN, the CNN used consists of three convolutional layers and two fully connected layers, where the output feature map of each convolution layer for: ， in, Indicates The feature map after layer convolution, Indicates The convolution kernel of the layer, Indicates The bias of the layer, represents the convolution operation, represents the activation function; The convolution output features are reduced in dimension through the maximum pooling operation, and the pooling window is set to , the pooling operation formula is: ， in, represents the features after pooling, represents the last layer of convolutional feature map, represents the pooling window size, The pooling feature Input to the fully connected layer for classification processing to generate feature vector : ， in, represents the weight matrix of the fully connected layer, represents the bias of the fully connected layer, represents the final extracted feature vector; Extracted feature vectors The feature vector of the registered users in the database The cosine similarity calculation is performed using the following formula: ， in, represents the similarity between the extracted features and the reference features, represents the reference user feature vector, and are the modulus lengths of the eigenvectors respectively; If the similarity Exceeding the set threshold , the verification is successful and a unique user identifier is generated : ， in, Indicates the user's unique operation identifier. Represents a hash generation function.

5. The battery anti-theft system for battery swap cabinet based on artificial intelligence as claimed in claim 4 is characterized by: The surrounding environment characteristics include human density and tool usage.

6. The battery anti-theft system for battery swap cabinet based on artificial intelligence as claimed in claim 5 is characterized by: All battery replacement operations and abnormal event data are encrypted and stored using blockchain technology.

7. The artificial intelligence-based battery anti-theft system for battery swap cabinets according to claim 6 is characterized in that: The emergency response mechanism includes remotely locking the battery compartment door and activating a high-pitched alarm, while pushing abnormal event data to the management center in real time.

8. The battery anti-theft system for battery swap cabinet based on artificial intelligence as claimed in claim 7 is characterized by: After confirming the existence of abnormal behavior, the emergency response mechanism is triggered, and the steps of grading the abnormal behavior are: According to the classification result in step S3 The probability value and threat level corresponding to the category define the threat value of abnormal behavior : ， in, Indicates the threat value of abnormal behavior, Indicates the classification result The probability value of class behavior, Indicates Predefined threat weights for class behaviors, Set the threat level threshold ,Will Mapping to different threat levels : like ,but , like ,but , … like ,but , in, Indicates the threat level, A sequence of thresholds representing threat classification; Based on threat level , select the corresponding emergency response measures, and set the emergency response matrix as , among which The row indicates the threat level is The emergency action set: ， like , then execute the response action Specific measures include the following operations: Remotely lock the battery door, Activate the high-pitched alarm, And push abnormal event data to the management center and store it encrypted through blockchain.

9. The artificial intelligence-based battery anti-theft system for battery swap cabinets according to claim 8 is characterized in that: The encryption method through blockchain is: Organize abnormal event data into transactions , including timestamp , Threat Level , User Operation Identifier , Description of abnormal behavior : ， in, Indicates abnormal event data transaction, The timestamp indicating when the event occurred. Indicates the threat level, A unique identifier for the user. Descriptive information indicating abnormal behavior, Will Broadcast to the blockchain network and generate new blocks : ， in, Represents a new block, represents a hash function, Indicates the current transaction data. Indicates the hash value of the previous block.

Images