Vehicle-mounted user credit prediction method and system
Through the enhanced PCA-LSTM model, the creditworthiness of on-board users is predicted in the Internet of Vehicles Group Intelligence Sensing System, the problems of participant selection and data security are solved, efficient and accurate user data analysis and privacy protection are achieved, and real-time processing needs are met.
Patent Information
- Application Number
- CN202410979106.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-22
- Publication Date
- 2024-11-26
AI Technical Summary
Improper selection of participants in the Internet of Vehicles Group Intelligence Sensing System leads to redundancy of sensing data and data information security issues, affecting system performance and user privacy.
Using the vehicle-mounted user credit prediction method based on the enhanced PCA-LSTM model, combined with principal component analysis and long-term memory network, distributed processing is carried out through edge nodes to predict user credit, and generate pseudonyms during data processing to achieve anonymity and ability. Retrospective.
It improves the accuracy and system performance of user data analysis, reduces computing complexity, protects user privacy, and meets the real-time processing needs in large-scale data environments.
Smart Images

Figure CN119026726A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of participant selection in a group intelligence perception system of an Internet of Vehicles, and in particular, relates to a method and system for predicting the credit rating of an in-vehicle user based on an enhanced PCA-LSTM model. Background Art
[0002] At present, with the rapid development of wireless communication and sensor technology, as well as the widespread popularity of various wireless terminal devices, the Internet of Things has ushered in a new development opportunity, making it possible to carry out larger-scale, more complex, more thorough and more comprehensive sensing through ubiquitous mobile devices. Although smartphones still occupy an important position in crowd sensing, their limited resources and unpredictable movement patterns limit their application potential.
[0003] Crowd perception is one of the core functions of the Internet of Vehicles (IoV). Compared with ordinary mobile devices, vehicles have greater mobility, a wider sensing range, and can be equipped with precision sensors and high-performance data processing, storage and wireless communication equipment, and are not subject to strict energy constraints, making them modern products that integrate multiple advanced technologies. These technological innovations not only improve the performance of traditional vehicles, but also give vehicles intelligent functions such as automatic driving, intelligent navigation, and voice control, transforming vehicles from a single means of transportation to a modern consumer electronic product. In addition, the vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication modes provided by the Internet of Vehicles have greatly expanded the capacity and form of crowd perception.
[0004] A typical Internet of Vehicles crowd-sensing system consists of three main parts: participants, perception platform and requester, such as Figure 1 As shown in the figure. Participants refer to vehicles that provide perception data to the crowd perception system; the perception platform is usually composed of multiple perception servers located in a data center; the requester refers to the entity that sends a request to the perception platform to obtain the required perception data. These three parts work together to form the core framework of the Internet of Vehicles crowd perception system.
[0005] In addition, based on the division of labor and cooperation among the above entities, the system sensing task can be completed through operations such as collection, uploading, calculation, and storage of sensing data. Therefore, improper selection of sensing users leads to redundancy of sensing data, resulting in poor quality of sensing tasks. At the same time, due to malicious users uploading false data, the performance of the sensing task will inevitably decline. In addition, there is a risk of user and data privacy leakage. In crowd sensing applications, by using vehicles to collect large-scale sensing data, the available on-board resources and larger mobility patterns are increased, but it will bring many new challenges to the selection of vehicle users. Given the limited budget and the necessity of high-quality task completion, selecting ideal participants is crucial to improving the quality of crowd sensing. Therefore, how to accurately screen high-quality users from many participants to optimize the performance of the Internet of Vehicles system, improve data quality, enhance user satisfaction, and protect user privacy has become a new research hotspot.
[0006] Through the above analysis, the problems and defects of the prior art are as follows:
[0007] (1) Participant selection problem: The selection of participants has an important impact on the quality of the system’s perception task. Improper selection of onboard users will lead to redundant sensing data, which not only increases the system’s storage and processing burden, but also reduces the quality of the sensing task. Redundant data will mask useful information, causing the system to be unable to accurately complete the sensing task.
[0008] (2) Data information security and user privacy security: In the Internet of Vehicles crowd-intelligence perception system, data information and security are key challenges. First, malicious users upload false data, which will directly affect the performance of the perception task and make the system unable to accurately perceive and analyze the environment. Second, the risk of user privacy leakage is also an issue that cannot be ignored. Summary of the invention
[0009] In view of the problems existing in the prior art, the present invention provides a method and system for predicting the credit of in-vehicle users based on an enhanced PCA-LSTM model to solve the problem of participant selection in the Internet of Vehicles group intelligence perception system.
[0010] The present invention is implemented as follows: a vehicle-mounted user credit prediction method based on an enhanced PCA-LSTM model, the method adopts an enhanced PCA-LSTM model, combines principal component analysis PCA dimensionality reduction and long short-term memory network LSTM time series processing capabilities, analyzes user data, and uses distributed processing of edge nodes to predict the user's credit; the method generates a pseudonym in the data processing process to achieve user anonymity and traceability.
[0011] Furthermore, the implementation steps of the method are:
[0012] Step 1: System settings, which is used by the key generation center to publish the system parameter list;
[0013] Step 2: User registration, which is used for the in-vehicle user to register their identity. The Road Traffic Administration will generate a pseudonym for them to protect the user's identity privacy and to trace the real identity of the illegal vehicle;
[0014] Step 3: key generation, which is used by the key generation center to generate the user's private key;
[0015] Step 4: Task distribution. The perception platform sends the perception task to the edge node. The edge node distributes the perception task to the vehicle user. The vehicle user who is willing to participate in the perception task will respond to the request of the edge node. The edge node sends the feedback of the vehicle user to the perception platform. The perception platform sends the vehicle user data to the edge node.
[0016] Step 5, credit prediction, after receiving the in-vehicle user data from the perception platform, the edge node conducts in-depth analysis of the user data and predicts the user's creditworthiness.
[0017] Further, the step 1 comprises:
[0018] The key generation center PKG selects the security parameter k, and the order of the additive cyclic group G1 and the multiplicative cyclic group G2 is a prime number q>2 k , the bilinear mapping satisfied by these two cyclic groups is G1×G1→G2, the generator of group G1 is P; secondly, PKG randomly selects the master key Calculate the public key P pub =s·P, then select a secure hash function H1, H3, H4: {0, 1} * →G1, H2: PKG public system parameter list
[0019] Further, the step 2 includes:
[0020] Car users need to register with the Road Traffic Administration (RTA) for their ID i For in-vehicle users, RTA generates a pseudonym ID for them i , RTA selects H1:{0,1} * →G1, calculate ID' i =H1(ID i ). When tracing illegal vehicles, RTA can show or expose the real identity of the user;
[0021] The step three includes: the vehicle user sends an ID′ i Give PKG, PKG calculates the user's private key S ID =s·ID′i .
[0022] Further, the step 4 includes:
[0023] The perception platform uses a private key to encrypt the perception task task i ={j, ID′ R , type, cr}, and send it to the edge node; the perception task contains the task number j, the pseudonym ID of the data demander ′ R , data type requirements type and the credit threshold cr for vehicle users and other information; the edge node uses the public key of the perception platform to decrypt and obtain the perception task, and then distributes it to nearby vehicle users; vehicle users who are willing to participate in the perception task will respond to the request of the edge node;
[0024] The edge node uses its own private key to encrypt the vehicle user's feedback and sends it to the perception platform; then, the perception platform uses the edge node's public key to encrypt the vehicle user data and sends it to the edge node; finally, the edge node decrypts it to obtain the vehicle user data.
[0025] Further, the step five includes:
[0026] After receiving the intelligent vehicle user data from the perception platform, the edge node uses an enhanced PCA-LSTM model, combined with principal component analysis PCA dimensionality reduction and long short-term memory network LSTM time series processing capabilities, to conduct in-depth analysis of the user data and predict the user's credit; the model's method structure is divided from bottom to top: input layer, normalization method, enhanced PCA method, convolution layer, ReLU activation function, pooling layer, Dropout layer, LSTM layer, fully connected layer and output layer.
[0027] Further, the enhanced PCA-LSTM model includes the following steps:
[0028] Step S1: Input layer: The edge node reads the in-vehicle user data, which covers the basic information, behavior records, and historical credit status of the in-vehicle user;
[0029] Step S2: Normalization method: The maximum and minimum normalization method is used to preprocess the data and scale its value range to [0, 1];
[0030] Step S3: Enhanced PCA method: The normalized data is subjected to dimensionality reduction processing using the principal component analysis (PCA) method. Through the PCA method, the original 17-dimensional features are mapped to 11 dimensions, and a new orthogonal feature set is constructed, where the 11 dimensions are determined based on the proportion of PCA explained variance of 0.95.
[0031] Step S4: Convolution layer: extract features from input data through convolution operation. This layer sets 64 output channels, the convolution kernel size is 3x3, the step size is 1, and the 'same' padding method is used to ensure that the output has the same spatial dimension as the input;
[0032] Step S5: ReLU activation function: In order to enhance the nonlinear expression ability of the model, the Rectified Linear Unit activation function is introduced, and the expression is: f(x) = max(0, x). Where x is the input value and f(x) is the output value;
[0033] Step S6: Pooling layer: Downsampling operation is performed through the pooling layer to reduce the spatial dimension of the data and retain key features. Maximum pooling is used to select the largest value from each pooling window as the output;
[0034] Step S7: Dropout layer: By setting the drop probability to 0.3, the output of some neurons is randomly dropped to prevent the model from overfitting during training;
[0035] Step S8: LSTM layer: The LSTM layer is the core part of the model. It uses a special gate structure, namely input gate, forget gate and output gate, to capture long-term dependencies in sequence data.
[0036] Step S9: Fully connected layer: The features are mapped to the output space through the fully connected layer to obtain the credit prediction value of the vehicle user. The output calculation formula of this layer is: y = W x + b. Where W is the weight vector and b is the bias term;
[0037] Step S10: Output layer: Finally, the predicted credit value of the vehicle user is output; if the credit of the vehicle user meets the corresponding task requirements, the edge node will send the review result to the vehicle user; otherwise, the result will be marked as failed review; after the vehicle user who passes the review completes the task, the data result will be sent back to the perception platform; finally, the perception platform will reward those vehicle users who submit valuable data.
[0038] Further, the state update process of the LSTM in step S8 includes the following steps:
[0039] Step S811: First, the LSTM unit processes the information of the previous memory state through the forget gate to determine which information needs to be forgotten from the memory state. The formula is: t =σ(W f ·[h t-1 ,x t ]+b f ). Among them, f t is the value of the forget gate, σ is the sigmoid activation function, h t-1is the hidden state at the previous moment, x t is the input at the current moment, W f is the weight vector of the forget gate, b f is the bias term of the forget gate;
[0040] Step S812: The LSTM unit decides which information is stored in the memory; on the one hand, the input gate decides which information is updated; on the other hand, the candidate vector is updated by the tanh layer, the formula is: t =σ(W i ·[h t-1 , x t ]+b i ), Among them, i t is the value of the input gate, W i is the weight vector of the input gate, b i is the bias term of the input gate, is the candidate vector, tanh is the tanh activation function, W C is the weight vector of the candidate vector, b C is the candidate vector bias term;
[0041] Step S813: The LSTM unit combines the above two parts to update the memory state. The formula is: Among them, C t is the current unit state, f t is the value of the forget gate, C t-1 is the unit state at the previous moment, i t is the value of the input gate, is a candidate vector;
[0042] Step S814: Finally, the LSTM unit uses the output gate to control the memory state that needs to be output. The formula is: t =σ(W o ·[h t-1 , x t ]+b o ), h t =o t tanh(C t ). Among them, o t is the value of the output gate, W o is the weight vector of the output gate, b o is the bias term of the output gate, h t is the hidden state at the current moment, C t The current unit status.
[0043] Another object of the present invention is to provide a vehicle-mounted user credit prediction system based on an enhanced PCA-LSTM model, comprising:
[0044] System setting module: used for the parameter list of the key generation center public system;
[0045] User registration module: used for in-vehicle users to register their identities. The Road Traffic Administration generates pseudonyms for them to protect user privacy and to track down the true identities of illegal vehicles.
[0046] Key generation module: used by the key generation center to generate user private keys;
[0047] Task distribution module: used for the perception platform to send perception tasks to edge nodes, which in turn distribute perception tasks to vehicle users. Vehicle users who are willing to participate in perception tasks will respond to requests from edge nodes, and edge nodes will send feedback from vehicle users to the perception platform. The perception platform will send vehicle user data to edge nodes.
[0048] Credit prediction module: After receiving the in-vehicle user data from the perception platform, the edge node conducts in-depth analysis of the user data and predicts the user's creditworthiness;
[0049] Enhanced PCA-LSTM analysis module: used to analyze user data and predict user credit by combining PCA dimensionality reduction and LSTM time series processing capabilities;
[0050] User anonymity and traceability module: Generate pseudonyms during data processing to achieve user anonymity and traceability.
[0051] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of a vehicle user credit prediction method based on an enhanced PCA-LSTM model.
[0052] The problems existing in the prior art solved by the present invention are:
[0053] Data dimension is too high: In the prediction of in-vehicle user credit, the data is usually of high dimension and complexity, which makes traditional prediction methods difficult to handle and has high computational complexity.
[0054] Ignoring time series relationships: Many existing methods fail to effectively capture and utilize long-term dependencies in time series data, resulting in low prediction accuracy.
[0055] Insufficient privacy protection: Existing systems often lack effective privacy protection measures during user data processing and analysis, which can easily lead to user identity leakage and data abuse.
[0056] Insufficient distributed processing capabilities: Traditional centralized processing methods are inefficient in large-scale data environments and cannot meet real-time processing requirements.
[0057] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:
[0058] First, the present invention addresses the problems existing in the prior art, such as high data dimension, neglect of time series relationship, insufficient privacy protection and insufficient distributed processing capability, and proposes a method and system for predicting the credit rating of vehicle users based on an enhanced PCA-LSTM model. The significant technical advances of the present invention are:
[0059] Enhanced PCA-LSTM model: Combining PCA dimensionality reduction and LSTM time series processing capabilities, it effectively reduces data dimensions, reduces computational complexity, and improves prediction accuracy. LSTM can capture long-term dependencies in data, making prediction results more accurate and reliable.
[0060] User privacy protection: By generating pseudonyms during data processing, user anonymity and traceability are achieved, protecting user identity privacy. Even during data analysis and prediction, the user's real identity information will not be leaked.
[0061] Distributed edge computing: Use edge nodes for distributed processing to effectively share the computing load and improve the processing efficiency and response speed of the system. Edge nodes can quickly process and analyze user data, predict user credit in real time, and meet the real-time processing needs in large-scale data environments.
[0062] System integrity: System settings, user registration, key generation, task distribution, credit prediction and data anonymity modules work closely together to form a complete credit prediction system. Each module has clear functions and reasonable division of labor to ensure the stability and reliability of the system.
[0063] Second, from the product perspective, the technical solution to be protected by the present invention has the following technical effects and advantages:
[0064] Efficient and accurate: Combining PCA dimensionality reduction and LSTM time series processing capabilities, it achieves in-depth analysis and accurate prediction of user data, improving the prediction accuracy and efficiency of the system.
[0065] Privacy protection: By generating pseudonyms to achieve user anonymity and traceability, user identity privacy is effectively protected and the security and credibility of the system are enhanced.
[0066] Distributed processing: Using edge nodes for distributed processing effectively shares the computing load, improves the processing efficiency and response speed of the system, and meets the real-time processing requirements in large-scale data environments.
[0067] System stability and reliability: System settings, user registration, key generation, task distribution, credit prediction, and data anonymity modules work closely together to ensure system stability and reliability.
[0068] Third, (1) Expected benefits and commercial value:
[0069] The technical solution of the present invention can be widely used in the Internet of Vehicles group intelligence perception system to help the system achieve efficient selection and credit evaluation of participants. With the continuous development of Internet of Vehicles technology, this technical solution will have broad market prospects and huge commercial value. By improving prediction accuracy, enhancing privacy protection and improving system processing efficiency, the present invention can bring significant benefits and convenience to related enterprises and users.
[0070] (2) Filling the technological gap:
[0071] At present, there is no technical solution similar to the present invention in the field of vehicle user credit prediction at home and abroad. The proposal of the present invention fills the technical gap in this field and provides a new solution to the problem of participant selection in the group intelligence perception system of the Internet of Vehicles.
[0072] (3) Overcoming technical bias:
[0073] Traditional methods for predicting the creditworthiness of in-vehicle users often only focus on short-term data dependencies or simply perform data dimensionality reduction, ignoring the long-term dependencies and complexity in the data. This invention successfully overcomes this technical bias by combining PCA dimensionality reduction and LSTM time series processing capabilities, achieving in-depth analysis and accurate prediction of user data.
[0074] Fourth, in the field of vehicle user credit prediction, traditional methods are often unable to effectively process complex user data, especially when evaluating credit while ensuring user privacy. At the same time, due to the diversity and high dimensionality of user data, traditional algorithms are difficult to capture the deep features and time series dependencies of the data, resulting in low prediction accuracy. In addition, existing technologies often have difficulty in achieving a balance between user anonymity and data traceability.
[0075] In order to solve the above problems, the present invention proposes a method for predicting the credit rating of vehicle users based on an enhanced PCA-LSTM model. The method combines the dimensionality reduction of principal component analysis (PCA) and the time series processing capability of long short-term memory (LSTM) network, and conducts in-depth analysis of vehicle user data through distributed processing of edge nodes. In the data processing process, the present invention adopts the technology of generating pseudonyms, which not only protects the privacy of user identity, but also ensures that the true identity of the illegal vehicle can be traced when necessary.
[0076] The implementation of the method includes the steps of system setting, user registration, key generation, task distribution and credit prediction. First, the key generation center publishes the system parameter list and generates a private key for the user. Then, the in-vehicle user obtains a pseudonym through the registration process and participates in the perception task. The perception platform distributes the task to the edge node, and the edge node distributes the task to the in-vehicle user. After receiving the in-vehicle user data, the edge node uses the enhanced PCA-LSTM model to predict the credit.
[0077] By adopting the enhanced PCA-LSTM model, the present invention successfully solves the difficulties faced by traditional methods in processing complex user data and improves the accuracy of credit prediction. At the same time, the application of pseudonym technology ensures the traceability of data while ensuring user privacy, providing strong support for road traffic management. In addition, the distributed processing architecture of the method improves the scalability and robustness of the system, enabling it to adapt to the processing needs of large-scale user data. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] Figure 1 It is a flow chart of a method for predicting the credit rating of an in-vehicle user based on an enhanced PCA-LSTM model provided by an embodiment of the present invention;
[0079] Figure 2 Schematic diagram of a typical Internet of Vehicles group intelligence perception system provided by an embodiment of the present invention;
[0080] Figure 3 It is a schematic diagram of the model calculation graph structure provided by an embodiment of the present invention.
[0081] Figure 4 4 is a model performance comparison diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0082] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0083] Example 1: Vehicle User Credit Prediction Method Based on Enhanced PCA-LSTM Model
[0084] The present invention provides a method for predicting the credit of an in-vehicle user based on an enhanced PCA-LSTM model. The method is special in that the enhanced PCA-LSTM model is used. The model combines PCA dimension reduction and LSTM time series processing capabilities to conduct in-depth analysis of user data, predict the credit of the user, reduce computational overhead, and improve the accuracy of credit prediction. Figure 1The figure shows a flow chart of a method for predicting the creditworthiness of an in-vehicle user based on an enhanced PCA-LSTM model provided by an embodiment of the present invention, including: system setting, user registration, key generation, task distribution and credit prediction.
[0085] Specifically, the system is set up to use the parameter list of the key generation center to disclose the system, specifically including: the key generation center (PKG) selects the security parameter k, the order of the additive cyclic group G1 and the multiplicative cyclic group G2 is a prime number q>2 k , the bilinear mapping satisfied by these two cyclic groups is G1×G1→G2, the generator of group G1 is P; secondly, PKG randomly selects the master key Calculate P pub =s·P, then select a secure hash function H1, H3, H4: {0, 1} * →G1, H2: PKG public system parameter list
[0086] User registration is used for in-vehicle users to register their identities. The Road Traffic Authority generates a pseudonym for them to protect the user's identity privacy. At the same time, the real identity of the illegal vehicle can be traced. Specifically, in-vehicle users need to register with the Road Traffic Authority (RTA). i For in-vehicle users, RTA generates a pseudonym ID for them i , RTA selects H1:{0,1} * →G1, calculate ID' i =H1(ID i ). When tracking down illegal vehicles, RTA can show or expose the real identity of the user.
[0087] Key generation, used by the key generation center to generate user private keys, specifically including: the vehicle user sends ID′ i Give PKG, PKG calculates the user's private key S ID =s·ID′ i .
[0088] Task distribution, which is used by the perception platform to send the perception task to the edge node. The edge node distributes the perception task to the vehicle user. The vehicle user who is willing to participate in the perception task will respond to the request of the edge node. The edge node sends the feedback of the vehicle user to the perception platform. The perception platform sends the vehicle user data to the edge node. Specifically, the perception platform uses the private key to encrypt the perception task task. i ={j, ID′ R , type, cr}, and sends it to the edge node. The perception task contains the task number j, the pseudonym ID of the data demander ′ R, data type requirements type, and the credit threshold cr for vehicle users. The edge node uses the public key of the perception platform to decrypt the perception task and then distributes it to nearby vehicle users. Vehicle users who are willing to participate in the perception task will respond to the edge node's request. The edge node then uses its own private key to encrypt the vehicle user's feedback and sends it to the perception platform. After that, the perception platform uses the public key of the edge node to encrypt the vehicle user data and then sends it to the edge node. Finally, the edge node decrypts and obtains the vehicle user data.
[0089] Credit prediction: After receiving the vehicle user data from the perception platform, the edge node conducts in-depth analysis of the user data and predicts the user's credit. Specifically, after receiving the intelligent vehicle user data from the perception platform, the edge node uses an enhanced PCA-LSTM model, which combines PCA dimensionality reduction and LSTM time series processing capabilities to conduct in-depth analysis of the user data and predict the user's credit. The method structure of the model is divided from bottom to top into: input layer, normalization method, enhanced PCA method, convolution layer, ReLU activation function, pooling layer, Dropout layer, LSTM layer, fully connected layer and output layer.
[0090] In an embodiment of the present invention, the vehicle-mounted user credit prediction method adopts an enhanced PCA-LSTM model, which combines PCA dimensionality reduction and LSTM time series processing capabilities to conduct in-depth analysis of user data, predict the user's credit, reduce computing overhead, and improve the accuracy of credit prediction.
[0091] Before describing the method of the present invention, the definition of the normalization method is described:
[0092] Definition of normalization method:
[0093] The present invention adopts the maximum and minimum normalization method, which linearly rescales the data from a range to [0,1]. The eigenvalues of X are based on their minimum and maximum values X min and X max Normalize. It calculates the value s of feature X i Convert to a predefined range [nX min , nX max ] value s′ i Calculation method: Among them, X min is the minimum value of the column where the data is located, X max is the maximum value of the column where the data is located, and n represents a parameter for adjusting the original normalized range.
[0094] like Figure 2As shown, the embodiment of the present invention also provides the model calculation graph structure from bottom to top, as shown below:
[0095] Step S1: Input Layer: The edge node reads the vehicle user data. This data mainly covers the basic information, behavior records and historical credit status of the vehicle user.
[0096] Step S2: Normalization method: To ensure the consistency and comparability of the data, the MinMaxScaler method is used to preprocess the data and scale its numerical range to [0,1].
[0097] Step S3: Enhanced PCA method (PCA-Enhanced): The normalized data is subjected to dimensionality reduction using the principal component analysis (PCA) method. Through the PCA method, the original 17-dimensional features are mapped to 11 dimensions, and a new orthogonal feature set is constructed. The 11 dimensions are determined based on the proportion of PCA explained variance of 0.95. This step helps to reduce the data dimension while retaining the main information.
[0098] Step S4: Convolutional Layer: Extract features from input data through convolution operation. This layer sets 64 output channels, the convolution kernel size is 3x3, the step size is 1, and the 'same' padding method is used to ensure that the output has the same spatial dimension as the input.
[0099] Step S5: ReLU activation function (ReLUActivation): In order to enhance the nonlinear expression ability of the model, the ReLU (Rectified Linear Unit) activation function is introduced. The expression of this function is: f(x) = max(0, x).
[0100] Step S6: Pooling Layer: Downsampling is performed through the pooling layer to reduce the spatial dimension of the data and retain key features. The solution of the present invention adopts MaxPooling to select the maximum value from each pooling window as the output.
[0101] Step S7: Dropout layer: By setting the drop probability to 0.3, the output of some neurons is randomly dropped to prevent the model from overfitting during training.
[0102] Step S8: LSTM layer (LSTMLayer): The LSTM layer is the core part of the model, which uses a special gate structure (input gate, forget gate and output gate) to capture long-term dependencies in sequence data.
[0103] Step S9: Fully Connected Layer: The fully connected layer maps the features to the output space to obtain the credit prediction value of the vehicle user. The output calculation formula of this layer is: y = W x + b.
[0104] Step S10: Output Layer: Finally, the predicted credit value of the vehicle user is output. If the vehicle user's credit meets the corresponding task requirements, the edge node will send the audit result to the vehicle user; otherwise, the result will be marked as audit failure. After the vehicle user who passes the audit completes the task, the data result will be sent back to the perception platform. Finally, the perception platform will reward the vehicle users who submit valuable data.
[0105] Preferably, the state update process of the LSTM in step S8 includes the following steps:
[0106] Step S811: First, the LSTM unit processes the information of the previous memory state through the forget gate to determine which information needs to be forgotten from the memory state. The formula is: t =σ(W f ·[h t-1 , x t ]+b f );
[0107] Step S812: Next, the LSTM unit decides which information to store in memory. On the one hand, the input gate decides which information to update. On the other hand, the candidate vector is updated by the tanh layer, and the formula is: t =σ(W i ·[h t-1 ,x t ]+b i ),
[0108] Step S813: The LSTM unit combines the above two parts to update the memory state. The formula is:
[0109] Step S814: Finally, the LSTM unit uses the output gate to control the memory state that needs to be output. The formula is: t =σ(W o ·[h t-1 , x t ]+b o ), h t =o t tanh(C t ).
[0110] Application Example 1: User Credit Prediction in Intelligent Vehicle System
[0111] In intelligent vehicle systems, user credit prediction is crucial to ensuring road safety, optimizing traffic flow, and providing personalized services. By applying the enhanced PCA-LSTM model of the present invention to intelligent vehicle systems, the present invention can achieve the following functions:
[0112] Real-time credit evaluation: The vehicle-mounted equipment collects the user's driving behavior data in real time, such as speed, acceleration, steering angle, etc., and transmits this data to the edge node. The edge node uses the enhanced PCA-LSTM model to process and analyze this data and predict the user's credit in real time. This real-time evaluation method enables the system to quickly respond to changes in the user's driving behavior and take corresponding measures.
[0113] Personalized service recommendations: Based on the user's credit prediction results, the system can provide personalized service recommendations to users. For example, for users with high credit, the system can recommend higher-level safety driving assistance functions, such as automatic emergency braking, adaptive cruise control, etc. For users with lower credit, the system provides more driving behavior guidance or restricts the use of certain functions.
[0114] Violation warning: Through real-time credit assessment, the system can promptly detect potential violations of users, such as speeding, running red lights, etc. Once these behaviors are detected, the system can immediately issue a warning to the user to remind them to pay attention to traffic safety. At the same time, the system can also send this information to the traffic management department for further supervision and processing.
[0115] Distributed processing and collaborative work: In smart vehicle systems, edge nodes play an important role. They are not only responsible for processing and analyzing the data uploaded by vehicle devices, but also work in collaboration with other edge nodes and central servers. Through the distributed processing architecture, the system can operate efficiently in a large-scale data environment and respond quickly to user needs.
[0116] Data anonymity and privacy protection: The system always maintains the anonymity of users throughout the application process. The user's real identity information will not be leaked at any stage. At the same time, the system ensures the security of user data through encryption technology and access control strategies to prevent unauthorized access and leakage.
[0117] Through the demonstration of this application embodiment, the present invention can see the great potential of the enhanced PCA-LSTM model of the present invention in the vehicle-mounted user credit prediction system. It can not only improve the real-time performance and accuracy of the system, but also provide users with more personalized and safe services. At the same time, the model also has wide applicability and scalability, and can be applied to credit prediction problems in other fields.
[0118] The embodiments of the present invention have achieved some positive effects during the development or use process, and indeed have great advantages over the prior art. The following content is described in conjunction with data, charts, etc. of the test process.
[0119] In order to comprehensively and deeply evaluate the prediction performance of the present invention, the present invention carefully selected the currently popular stacking model (Stacking) and the efficient XGBoost model (XGBoost) for comparison. The present invention adopts MSE and MAE as the criteria for evaluating prediction accuracy, and performs detailed data analysis and visualization. The prediction results of the three schemes are compared in Table 1. The specific data are: the MSE value of the present invention is only 0.027, which is much lower than the 2.948 of the stacking model and the 0.0524 of the XGBoost model. Similarly, in terms of MAE, the scheme of the present invention has also achieved impressive results, with a MAE value of 0.104, which is slightly higher than the 0.01 of the XGBoost model, but much lower than the 0.14 of the stacking model. Figure 4 As shown, the comparison of the prediction results of the three models clearly shows their respective performances. It can be intuitively seen from the figure that the present invention is significantly better than the other two models in terms of prediction accuracy.
[0120] Table 1 Performance comparison of three models
[0121] Model Stacking XGBoost The present invention MSE 0.5 0.054 0.027 MAE 0.14 0.1 0.104
[0122] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware portion can be implemented using dedicated logic; the software portion can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. A person of ordinary skill in the art will appreciate that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.
[0123] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.
Claims
1. A vehicle user credit prediction method based on an enhanced PCA-LSTM model, characterized in that: The method adopts an enhanced PCA-LSTM model, combines principal component analysis PCA dimensionality reduction and long short-term memory network LSTM time series processing capabilities, analyzes user data, and uses distributed processing of edge nodes to predict the user's credit; the method generates pseudonyms during data processing to achieve user anonymity and traceability.
2. The vehicle user credit prediction method based on the enhanced PCA-LSTM model according to claim 1, characterized in that: The implementation steps of the method are: Step 1: System settings, which is used by the key generation center to publish the system parameter list; Step 2: User registration, which is used for the in-vehicle user to register their identity. The Road Traffic Administration will generate a pseudonym for them to protect the user's identity privacy and to trace the real identity of the illegal vehicle; Step 3: key generation, which is used by the key generation center to generate the user's private key; Step 4: Task distribution. The perception platform sends the perception task to the edge node. The edge node distributes the perception task to the vehicle user. The vehicle user who is willing to participate in the perception task will respond to the request of the edge node. The edge node sends the feedback of the vehicle user to the perception platform. The perception platform sends the vehicle user data to the edge node. Step 5, credit prediction, after receiving the in-vehicle user data from the perception platform, the edge node conducts in-depth analysis of the user data and predicts the user's creditworthiness.
3. The vehicle user credit prediction method based on the enhanced PCA-LSTM model as claimed in claim 2, characterized in that: The step one comprises: The key generation center PKG selects the security parameter k, and the order of the additive cyclic group G1 and the multiplicative cyclic group G2 is a prime number q>2 k , the bilinear mapping satisfied by these two cyclic groups is The generator of group G1 is P; secondly, PKG randomly selects the master key Calculate P pub =s·P, then select a secure hash function H1, H3, H4: {0, 1} * →G1, PKG public system parameter list 4. The vehicle user credit prediction method based on the enhanced PCA-LSTM model as claimed in claim 2, characterized in that: The second step comprises: Car users need to register with the Road Traffic Administration (RTA) for their ID i For in-car users, RTA generates a pseudonymous ID for them i ', RTA selects H1:{0,1} * →G1, calculate ID i '=H1(ID i ); When tracking down illegal vehicles, RTA can show or expose the real identity of the user; The step three includes: the vehicle user sends an ID i 'Give it to PKG, PKG calculates the user's private key S ID =s·ID i '.
5. The vehicle user credit prediction method based on the enhanced PCA-LSTM model as claimed in claim 2, characterized in that: The fourth step comprises: The perception platform uses a private key to encrypt the perception task task i ={j,ID' R ,type,cr}, and send it to the edge node; the perception task contains the task number j, the pseudonym ID of the data demander' R , data type requirements type and the credit threshold cr for vehicle users and other information; the edge node uses the public key of the perception platform to decrypt the perception task and then distributes it to nearby vehicle users; vehicle users who are willing to participate in the perception task will respond to the request of the edge node; The edge node uses its own private key to encrypt the vehicle user's feedback and sends it to the perception platform; then, the perception platform uses the edge node's public key to encrypt the vehicle user data and sends it to the edge node; finally, the edge node decrypts it to obtain the vehicle user data.
6. The vehicle user credit prediction method based on the enhanced PCA-LSTM model as claimed in claim 2, characterized in that: The step five comprises: After receiving the intelligent vehicle user data from the perception platform, the edge node uses an enhanced PCA-LSTM model, combined with principal component analysis PCA dimensionality reduction and long short-term memory network LSTM time series processing capabilities, to conduct in-depth analysis of the user data and predict the user's credit; the model's method structure is divided from bottom to top: input layer, normalization method, enhanced PCA method, convolution layer, ReLU activation function, pooling layer, Dropout layer, LSTM layer, fully connected layer and output layer.
7. The vehicle user credit prediction method based on the enhanced PCA-LSTM model according to claim 6, characterized in that: The PCA-LSTM model includes the following steps: Step S1: Input layer: The edge node reads the in-vehicle user data, which covers the basic information, behavior records, and historical credit status of the in-vehicle user; Step S2: Normalization method: The maximum and minimum normalization method is used to preprocess the data and scale its value range to [0, 1]; Step S3: Enhanced PCA method: The normalized data is subjected to dimensionality reduction processing using the principal component analysis (PCA) method. Through the PCA method, the original 17-dimensional features are mapped to 11 dimensions, and a new orthogonal feature set is constructed, where the 11 dimensions are determined based on the proportion of PCA explained variance of 0.
95. Step S4: Convolution layer: extract features from input data through convolution operation. This layer sets 64 output channels, the convolution kernel size is 3x3, the step size is 1, and the 'same' padding method is used to ensure that the output has the same spatial dimension as the input; Step S5: ReLU activation function: In order to enhance the nonlinear expression ability of the model, the RectifiedLinearUnit activation function is introduced, and the expression is: f(x) = max(0,x); Step S6: Pooling layer: Downsampling operation is performed through the pooling layer to reduce the spatial dimension of the data and retain key features. Maximum pooling is used to select the largest value from each pooling window as the output; Step S7: Dropout layer: By setting the drop probability to 0.3, the output of some neurons is randomly dropped to prevent the model from overfitting during training; Step S8: LSTM layer: The LSTM layer is the core part of the model. It uses a special gate structure, namely input gate, forget gate and output gate, to capture long-term dependencies in sequence data. Step S9: Fully connected layer: The features are mapped to the output space through the fully connected layer to obtain the credit prediction value of the vehicle user. The output calculation formula of this layer is: y = W x + b; Step S10: Output layer: Finally, the predicted credit value of the vehicle user is output; if the credit of the vehicle user meets the corresponding task requirements, the edge node will send the review result to the vehicle user; otherwise, the result will be marked as failed review; after the vehicle user who passes the review completes the task, the data result will be sent back to the perception platform; finally, the perception platform will reward those vehicle users who submit valuable data.
8. The vehicle user credit prediction method based on the enhanced PCA-LSTM model according to claim 7, characterized in that: The state updating process of the LSTM in step S8 comprises the following steps: Step S811: First, the LSTM unit processes the information of the previous memory state through the forget gate to determine which information needs to be forgotten from the memory state. The formula is: t =σ(W f ·[h t-1 ,x t ]+b f ); Step S812: The LSTM unit decides which information is stored in the memory; on the one hand, the input gate decides which information is updated; on the other hand, the candidate vector is updated by the tanh layer, the formula is: t =σ(W i ·[h t-1 ,x t ]+b i ), Step S813: The LSTM unit combines the above two parts to update the memory state. The formula is: Step S814: Finally, the LSTM unit uses the output gate to control the memory state that needs to be output. The formula is: t =σ(W o ·[h t-1 ,x t ]+b o ), h t =o t tanh(C t ).
9. A vehicle user credit prediction system based on an enhanced PCA-LSTM model according to the method of claim 1, characterized in that: include: System setting module: used for the parameter list of the key generation center public system; User registration module: used for in-vehicle users to register their identities. The Road Traffic Administration generates pseudonyms for them to protect user privacy and to track down the true identities of illegal vehicles. Key generation module: used by the key generation center to generate user private keys; Task distribution module: used for the perception platform to send perception tasks to edge nodes, which in turn distribute perception tasks to vehicle users. Vehicle users who are willing to participate in perception tasks will respond to requests from edge nodes, and edge nodes will send feedback from vehicle users to the perception platform. The perception platform will send vehicle user data to edge nodes. Credit prediction module: After receiving the in-vehicle user data from the perception platform, the edge node conducts in-depth analysis of the user data and predicts the user's creditworthiness; Enhanced PCA-LSTM analysis module: used to analyze user data and predict user credit by combining PCA dimensionality reduction and LSTM time series processing capabilities; Data anonymity and traceability module: Generate pseudonyms during data processing to achieve user anonymity and traceability.
10. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the vehicle user credit prediction method based on the enhanced PCA-LSTM model as claimed in claim 1.
