High-precision truth value discovery method for privacy protection in vehicle crowd sensing

Abstract

The invention discloses a high-precision truth value discovery method for privacy protection in vehicle crowd sensing, and the method comprises the steps: enabling a vehicle to determine a privacy budget, obtaining noise through a smart contract, encrypting the noise and original data, and submitting the encrypted noise and original data to a block chain; the vehicle generates a re-encryption key according to the private key of the vehicle and the public key of the cloud server, and sends the re-encryption key to the block chain; the smart contract performs homomorphic aggregation on the noise ciphertext and the data ciphertext to generate disturbance data encrypted by a vehicle public key; the intelligent contract uses the re-encryption key to convert the vehicle encrypted disturbance data into cloud server encrypted disturbance data, and sends the disturbance data to the cloud server to restore the disturbance data; and the cloud server screens the data sources, fuses the disturbance data based on the weight and the credibility of the data sources to obtain a disturbance truth phase, and carries out noise asymptotic elimination on all the data sources to obtain a truth phase aggregated by all the data sources. According to the method, the accuracy of true value discovery can be remarkably improved on the premise of protecting vehicle data privacy, and hostile attack resistance is achieved.

Description

Technical Field

[0001] This invention belongs to the field of vehicle networking and privacy computing technology, specifically relating to a high-precision truth discovery method for privacy protection in vehicle crowd perception. Background Technology

[0002] Truth-finding technology infers the truth by aggregating data sources with different weights. Its core comprises two iterative modules: a weight update phase and a truth-finding phase. The weight update phase adjusts weights based on how closely the data resembles the currently estimated truth; the truth-finding phase fixes the data source weights, prioritizing aggregation of high-weight data sources. These two modules iterate alternately until convergence. Due to its importance, this technology has been widely applied in fields such as vehicle crowdsourcing, healthcare, and website monitoring. With the rapid development of vehicle crowdsourcing technology, vehicles, as ubiquitous mobile sensors, can collect massive amounts of data for intelligent transportation systems. However, when vehicles submit raw data to often untrusted cloud servers, they face privacy risks, severely impacting their willingness to participate in data sharing. To address this, a perturbation-based privacy-preserving truth-finding technology has been proposed. This technology allows vehicles to generate noise using preset perturbation parameters based on differential privacy techniques to perturb the data, and then the cloud server performs truth-finding based on the perturbated data. Compared to traditional encryption schemes that rely on complex processing of encrypted raw data, this plaintext perturbation-based technology balances data integrity and processing efficiency, thus gaining widespread acceptance. However, existing perturbation-based privacy-preserving truth-finding technologies share a common flaw: while effectively protecting privacy, the added perturbation noise inevitably compromises the accuracy of truth discovery.

[0003] To mitigate the negative impact of noise, existing research mainly employs two strategies. The first strategy attempts to control the perturbation parameters of the data source during the data perturbation phase, reducing aggregation error by limiting the amount of added noise. The second strategy considers the perturbation level of the data source in the weight update module of truth discovery, mitigating its impact by reducing the aggregation weight of high-noise data sources. However, both methods have significant shortcomings. Noise reduction methods require assigning less noise to high-weight data sources, which violates the core principle of privacy-preserving truth discovery technology: data subjects should be able to autonomously choose perturbation parameters according to their own privacy needs to achieve quantitative privacy protection. Forcibly constraining the amount of noise undermines the privacy autonomy of data subjects and weakens their long-term motivation to contribute data. Weight adjustment methods, due to their inherent limitations, are difficult to effectively improve accuracy: on the one hand, all perturbed data must participate in aggregation, and the adjusted weights are always greater than zero, failing to completely eliminate the impact of noise; on the other hand, weight updates need to simultaneously weigh the perturbation level of the data source and data quality (i.e., the difference between the data and the true value). When high noise and high-quality data coexist (such as in time series data), weight adjustment struggles to reconcile this contradiction, resulting in poor accuracy recovery.

[0004] Furthermore, existing research indicates that recruiting trustworthy vehicles can improve the accuracy of truth discovery, and various trust assessment methods have been developed. However, these methods typically do not explicitly distinguish between a vehicle's trustworthiness and the quality of its data, implicitly assuming that trusted vehicles necessarily provide high-quality data. In reality, the data quality of trusted vehicles may fluctuate due to differences in the accuracy of their sensing devices. Therefore, to optimize truth discovery, vehicle swarm intelligence perception systems should prioritize vehicles that possess both high trustworthiness and high-quality data. Summary of the Invention

[0005] This invention provides a high-precision truth discovery method for privacy protection in vehicle crowd sensing, aiming to significantly improve the accuracy of truth discovery while protecting vehicle data privacy, and possessing the ability to resist malicious attacks.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0007] A high-precision truth discovery method for privacy protection in vehicle crowdsourcing perception includes:

[0008] The vehicle determines a differential privacy budget and generates noise through a smart contract. Then, it uses the vehicle's public key to encrypt the noise and the raw data perceived locally, and then submits them to the blockchain.

[0009] The vehicle generates a re-encryption key based on its own private key and the cloud server's public key, and sends it to the blockchain;

[0010] The smart contract homomorphically aggregates the received vehicle noise ciphertext and data ciphertext to generate perturbation data encrypted with the vehicle's public key.

[0011] The smart contract uses a re-encryption key to convert the perturbation data encrypted with the vehicle's public key into perturbation data encrypted with the cloud server's public key, and sends it to the cloud server to restore the perturbation data.

[0012] After obtaining disturbance data from all vehicles, the cloud server filters data sources based on the current global trust level of each vehicle.

[0013] The cloud server uses data source weights and trust levels to merge disturbed data from various data sources to obtain the truth of the disturbance and updates the global trust level of each data source.

[0014] The cloud server uses the disturbed truth as the initial cleaned truth, and works with all data sources to asymptotically eliminate noise from all data sources, thus obtaining the true truth aggregated from all data sources.

[0015] Furthermore, the vehicle determines a differential privacy budget and generates noise via a smart contract, specifically:

[0016] No. vehicle Choose a differential privacy budget independently based on your own privacy needs. Send a transaction to the smart contract and submit it. ;

[0017] Smart contract received Differential privacy budget Then, based on the preset global sensitivity... ,for Calculate the scale parameter of Laplace noise :

[0018]

[0019] The smart contract calls an on-chain verifiable random function to generate uniform random numbers. And calculate the unique corresponding satisfying - Differential privacy Laplace noise value :

[0020]

[0021] Smart contracts will reduce noise Returned to the corresponding [system / entity] via secure channel .

[0022] Furthermore, the vehicle generates a re-encryption key based on its own private key and the cloud server's public key as follows:

[0023]

[0024] In the formula, For vehicles The generated re-encryption key, For vehicles Randomly selected security parameters and vehicles The public and private keys for homomorphic encryption.

[0025] Furthermore, the smart contract uses a re-encryption key to convert the perturbation data encrypted with the vehicle's public key into perturbation data encrypted with the cloud server's public key. Specifically:

[0026] First, the perturbation data encrypted with the vehicle's public key. According to its definition, it can be expressed as:

[0027]

[0028] In the formula, It is the random number used when encrypting the vehicle. It is plaintext perturbation data;

[0029] Then, use the first part of the re-encryption key. right Perform modular exponentiation to extract plaintext related terms from the perturbation data:

[0030]

[0031] Reuse Combined with re-encryption key Part Two ,Will The public key space mapped to the cloud server is ultimately used to generate ciphertext conforming to the Paillier format. This refers to the perturbation data encrypted with the cloud server's public key. :

[0032] .

[0033] Furthermore, the cloud server uses its own private key. Restore perturbation data , specifically:

[0034] First, the cloud server encrypts the perturbation data using its own public key. Perform Paillier decryption:

[0035]

[0036] In the formula, This indicates the Paillier decryption operation. This indicates that the SP uses its own private key. Decrypting re-encrypted ciphertext The value after that, This represents a dedicated auxiliary function defined in the Paillier encryption algorithm. Its core function is to extract linear terms related to the plaintext from the modular exponentiation result of the ciphertext, providing crucial support for decryption.

[0037] Then, Substituting the expression into the above decryption calculation formula and simplifying, we get:

[0038]

[0039] Furthermore, because the vehicle's private key satisfies... , Modulus The multiplicative group is a standard mathematical structure in number theory, used to constrain safety parameters. The range of values ​​ensures the security of the re-encryption process; combined with the derived information in the re-encryption key... and The coprime property of cloud servers can eliminate The impact of this was ultimately used to reconstruct the vehicle's disturbance data. :

[0040] .

[0041] Furthermore, the cloud server filters data sources based on the current global trust level of each vehicle, specifically:

[0042] Based on the overall trust level of each vehicle Filter out the trustworthy data sources;

[0043] For each collective intelligence sensing task, recruit data sources from the above trusted data sources: if the number of trusted data sources is greater than the preset recruitment number N, then randomly select N data sources without repetition; if the number of trusted data sources is less than N, then recruit all trusted data sources.

[0044] Furthermore, the cloud server, based on data source weights and trust levels, merges the perturbation data from various data sources to obtain the truth of the perturbation, specifically including:

[0045] (1) Initialization;

[0046] Weight initialization: for each data source initial weights Set to equal values:

[0047]

[0048] in, This refers to the vehicle number for the current crowd sensing task, i.e., the data source number. Total number of data sources;

[0049] Trust initialization: For each data source Initial level of trust within the current crowd-sensing task Set to global trust :

[0050]

[0051] (2) Iterative truth discovery process; each iteration includes: truth calculation, weight update and trust update;

[0052] (3) Convergence judgment: The iterative truth discovery process continues until the change in the truth estimate is less than a preset threshold. :

[0053]

[0054] In the formula, It is a distance metric function;

[0055] (4) When the iteration converges, each data source Current level of trust Normalization is performed to obtain the final trust level. :

[0056]

[0057] (5) Based on the final trust level of all data sources and its disturbance data The truth of the perturbation in the final aggregate of weighted calculations :

[0058] .

[0059] Furthermore, the iterative truth discovery process includes:

[0060] First, cloud servers use data sources. Previous round weight and perturbation data Calculate the current round The true value of the perturbation :

[0061]

[0062] Then, based on the current round The true value of the perturbation Disturbance data from various data sources Distance, update each data source In the current round Weights:

[0063]

[0064] in, It is a distance metric function;

[0065] Based on weights Update the trust level of each data source. :

[0066]

[0067] in, Trust threshold; The highest weight among all data sources is selected; if there is no reliable data source, the highest weight among all vehicles is selected. To ensure reliable update rates, This is used to control the balance between historical trust levels and current assessments; For distance sensitivity parameters, This is used to adjust the degree of influence of weight differences on trust levels.

[0068] Furthermore, the global trust level updates for each data source include:

[0069] (1) Record the trust level data of the current crowd perception task: in each round After all task iterations are completed, collect all data sources for each iteration in this round. In-task trust level in all tasks ;

[0070] (2) Calculate this round Average trust level: For each data source, if in this round If someone participates in a task, their participation in this round will be calculated. The mean of in-task trust across all tasks, and constrain the mean to... Within the range;

[0071] (3) Update global trust level: based on the current round The average trust level is compared with the malicious threshold, and the global trust level of the data source is adjusted: if it is greater than the malicious threshold, the global trust level of the data source is increased by a preset value, and vice versa.

[0072] Furthermore, the cloud server collaborates with all data sources to asymptotically eliminate noise from all data sources, specifically including:

[0073] (1) Initialization: The cloud server will disturb the truth As the initial, cleansed truth: and obtain various data sources from the blockchain. Noise ciphertext ;

[0074] (2) Each data source One round of noise cancellation is performed accordingly, and each round Perform the following noise cancellation operations:

[0075] When performing blind noise cancellation on a cloud server, it first uses the data source of the current round. The public key is used to encrypt the previously cleaned truth: Subsequently, the cloud server retrieves the data source from the blockchain. Encryption noise And based on the ultimate trust in this data source in the discovery of the truth. Perform noise cancellation in encrypted state:

[0076]

[0077] Leveraging the homomorphic property of the Paillier cryptosystem, the above operation is equivalent to:

[0078]

[0079] Then from the data source Collaborate with the cloud server to perform cleaned truth encryption / decryption operations: The cloud server will encrypt the text... Send to data source The request is for it to decrypt; among which, the data source Use your own private key Decryption reveals the cleaned truth :

[0080]

[0081] Data source The decrypted and cleaned truth Use the next data source Re-encrypt using the public key:

[0082]

[0083] The re-encrypted result The data is returned to the cloud server for use in the next round of noise cancellation.

[0084] (3) Obtaining the final result: When participating in the task After all data sources participated in noise reduction, the final... Data sources Use your own private key Decryption reveals the cleaned truth Afterwards, instead of using public keys from other data sources for encryption, the decrypted result directly reveals the cleaned truth. Send to the cloud server; the cloud server obtains the final round of cleaned truth. This is the true truth of removing weighted noise components from all data sources. :

[0085] .

[0086] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0087] (1) Dual protection of privacy and high accuracy. Unlike existing solutions that require a trade-off between privacy protection and data accuracy, this invention allows data subjects to fully and autonomously choose differential privacy parameters, thus protecting their privacy. At the same time, through a progressive noise reduction mechanism, the impact of noise is gradually reduced in the encrypted state, thereby significantly improving the aggregation accuracy of truth discovery without disclosing individual data. This overcomes the common defect of noise accumulation leading to decreased accuracy in traditional perturbation methods.

[0088] (2) Decentralized Trusted Noise Generation and Secure Ciphertext Transfer. By combining blockchain smart contracts with proxy re-encryption technology, distributed trusted noise generation and secure conversion in ciphertext form are achieved. Blockchain, as the cornerstone of decentralized trust, ensures that the noise generation process is publicly verifiable; the proxy re-encryption module enables the cloud server to process only the converted ciphertext and not to obtain the original data or noise information, thereby enhancing the end-to-end privacy protection strength from the system architecture perspective.

[0089] (3) Fine-grained trust assessment and anti-malicious behavior capability. A weight-based dynamic trust identification mechanism is proposed to clearly distinguish between the historical trust level of a vehicle and the quality of its submitted data, overcoming the limitation of confusion between the two in traditional methods. Through the coordinated update of weights and trust values ​​in multiple iterations, the system can effectively identify and suppress the influence of malicious data sources, improving the robustness and reliability of the system in an open participation environment.

[0090] (4) Progressive and collaborative noise cancellation mechanism. A progressive noise cancellation process based on homomorphic encryption was designed, in which the cloud server and the data source collaboratively complete the removal of noise components without exposing plaintext. This mechanism not only supports efficient and accurate recovery under privacy protection, but also reduces the computational burden on a single entity, which is in line with the distributed and lightweight collaborative characteristics of vehicle swarm perception systems.

[0091] (5) Improved overall system efficiency and scalability. Compared with traditional privacy protection schemes that rely on complex multi-party secure computation or fully homomorphic encryption, this method reduces the overall computation and communication overhead through modular design (such as on-chain homomorphic computation, proxy re-encryption, and iterative trust updates) while ensuring the same level of privacy, making it more suitable for dynamic and large-scale vehicle crowd sensing application scenarios. Attached Figure Description

[0092] Figure 1 This is a schematic diagram of the system model of the method described in this invention;

[0093] Figure 2 This illustrates the impact of the total number of tasks M in the method described in this invention on the mean absolute error of the five schemes; where subgraphs (a) and (b) correspond to the differential privacy budget as follows: and ;

[0094] Figure 3 The influence of the total number of tasks M in the method described in this invention on the root mean square error of the five schemes is shown; where subgraphs (a) and (b) correspond to the differential privacy budget as follows: and ;

[0095] Figure 4 The impact of the total number of tasks M in the method described in this invention on the relative error rate performance of the five schemes is shown; where subgraphs (a) and (b) correspond to the differential privacy budget as follows: and . Detailed Implementation

[0096] The embodiments of the present invention will be described in detail below. These embodiments are based on the technical solutions of the present invention and provide detailed implementation methods and specific operation processes to further explain the technical solutions of the present invention.

[0097] The high-precision truth discovery method for privacy protection in vehicle crowd perception of the present invention, while retaining the perturbation-based privacy-preserving truth discovery framework and allowing data subjects to autonomously select perturbation parameters, innovatively introduces the following mechanism:

[0098] (1) Independently guaranteed data disturbance mechanism: Combining blockchain smart contracts and proxy re-encryption technology, noise that meets differential privacy requirements is generated in a distributed manner, which supports vehicles to secretly obtain noise to complete local disturbance, while ensuring that the cloud server only holds the noise ciphertext.

[0099] (2) Progressive privacy noise reduction mechanism: Two iterative processes are designed to reduce the impact of noise in the encrypted state. The cloud server first removes the retained encrypted noise components from the perturbed data based on homomorphic encryption technology; then, the relevant vehicles are required to decrypt part of the cleaned data. This process iterates step by step until all vehicles participate, achieving efficient noise reduction while ensuring data privacy.

[0100] (3) Weight-based Trust Identification Scheme: A two-layer trust evaluation mechanism is proposed. The intra-task trust level reflects the immediate performance of the data source in a single task, while the global trust level accumulates its historical behavior to form a long-term reputation profile. The two work together: the initial value of the intra-task trust level is derived from the global trust level, ensuring that data sources with good reputations have a higher starting point; after the task ends, the performance of the intra-task trust level updates the global trust level in reverse, forming a positive incentive and dynamic penalty mechanism. This design not only improves the accuracy of truth discovery but also provides a quantitative basis for participant screening and system anti-attack capabilities.

[0101] A high-precision method for privacy-preserving truth discovery in vehicle crowd sensing includes the following steps:

[0102] Step 1: System initialization.

[0103] The system comprises four types of entities: vehicles (DS), cloud servers (SP), the blockchain network, and smart contracts deployed on the chain. Each data source generates its own homomorphic cryptographic key pair. With private key The cloud server also generates a key pair: public key. With private key Smart contracts are deployed on the blockchain and have publicly verifiable code logic, which is used for subsequent noise generation, homomorphic computation, and proxy re-encryption.

[0104] Step 2: Privacy parameter submission and trusted noise generation.

[0105] Each vehicle Select differential privacy budget parameters independently based on your own privacy needs. , Send a transaction to the smart contract and submit it. After receiving the parameters, the smart contract will determine the appropriate parameters based on the preset global sensitivity. , for each Calculate the scale parameters of Laplace noise:

[0106]

[0107] The contract calls an on-chain verifiable random function to generate uniform random numbers. And calculate the unique corresponding satisfying - Differential privacy Laplace noise value:

[0108]

[0109] The noise Returned to the corresponding [system / entity] via secure channel and only Know its plaintext value

[0110] Step 3: Encrypt local data and noise on the blockchain.

[0111] The generated noise plaintext is received locally. Then use its public key The Paillier homomorphic encryption algorithm, respectively, is used to analyze its original data. and noise Encryption yields two separate ciphertexts: the original data ciphertext and the ciphertext. Noise-based ciphertext: Submit both encrypted messages to the blockchain storage.

[0112] Step 4: On-chain homomorphic aggregation generates encrypted perturbation data.

[0113] Smart contracts collected After submitting the encrypted raw data and encrypted noise, a homomorphic addition operation is automatically performed on the chain: leveraging the homomorphic addition property of the Paillier encryption system, the two ciphertexts are multiplied, and the decryption result is the sum of the plaintexts.

[0114]

[0115] This step directly calculates the encrypted perturbation data in the ciphertext state. It is executed publicly on the blockchain without decrypting any original information and without involving any plaintext data leakage.

[0116] Step 5: Agent re-encrypts and transfers data to cloud server.

[0117] Each DS uses its own private key and SP's public key Generate a re-encryption key The specific steps are as follows:

[0118] DS randomly selects security parameters It is not disclosed to any entity, ensuring the security of re-encryption, preventing proxies from deriving the private key, and calculating the re-encryption key, which consists of two parts to ensure security and verifiability:

[0119]

[0120] Part One The "authorization factor" used for proxy ciphertext conversion; Part Two Used to verify the legitimacy of re-encryption during SP decryption, associated with the DS private key parameter. Public key base of SP .

[0121] The smart contract then acts as a trusted agent to perform re-encryption, responsible for re-encrypting the perturbed data using the DS public key on the blockchain. "Transferred to the base of SP" Finally, "SP public key encrypted perturbation data" is generated. ",Will" Plaintext information carried "SP can decrypt ciphertext" ,and It must conform to the Paillier format to ensure that SP can subsequently perform truth discovery without revealing plaintext throughout the process. .

[0122]

[0123] in It is the random number generated during DS encryption.

[0124] Among them, "perturbation data encrypted with the DS public key" "Converted to "SP public key encrypted perturbation data" Specifically, it includes:

[0125] Step 1: Smart Contract Invocation Part 1 right Perform modular exponentiation to extract plaintext-related terms:

[0126]

[0127] Due to Paillier's additive homomorphism, the decryption result of this first step is... This is equivalent to putting plaintext Transform into " ".

[0128] Step 2: Replace the base with the "authorization mapping factor". Because... It's a binding item for "DS Authorization → SP Base", and the agent will " "Mapped to " The essence is to put " Bearing "Transfer to " through authorization factor superior" .

[0129] Step 3: The agent replenishes the random numbers required for SP encryption. To form a Paillier-compliant format Complete from arrive Valid mappings:

[0130]

[0131] Finally, SP performs decryption verification using its own private key. Restore the perturbation data :

[0132] SP Perform Paillier decryption:

[0133]

[0134] Substitute The expression, simplified, yields:

[0135]

[0136] Because DS's private key satisfies SP can eliminate the default DS private key parameter. Final restoration:

[0137]

[0138] Step 6: SP Recruitment Data Source

[0139] SP obtains disturbance data for all vehicles. Afterwards, it is necessary to filter more credible data as much as possible for subsequent truth discovery. In order to resist malicious DS from uploading erroneous data, it is necessary to filter and recruit data sources. The specific steps are as follows.

[0140] 1. Filter non-malicious data sources

[0141] At the start of each round of tasks, based on the global trust level Data sources currently identified as "non-malicious" are considered trustworthy. The criterion is that the global trust level of the data source is greater than the malicious threshold (default 0.3).

[0142] 2. Recruit data sources by task

[0143] For each task, recruit data sources from the above non-malicious data sources: if the number of non-malicious data sources is greater than the preset recruitment number N, then randomly select N data sources without repetition; if the number of non-malicious data sources is less than N, then recruit all non-malicious data sources.

[0144] Step 7: Iterative truth discovery based on weights and trust.

[0145] To mitigate the impact of data bias or noise on system accuracy, this embodiment introduces a parallel trust evaluation mechanism in addition to the traditional weighted iterative update. This mechanism effectively identifies and reduces the influence of malicious data sinks (DSs). The specific steps are as follows.

[0146] 1. Initialization.

[0147] Before the discovery of the truth began, the system examined each data source. weight and in-task trust Perform initialization.

[0148] (1) Weight initialization: The initial weights of all DS are set to equal values, that is:

[0149]

[0150] in The total number of DSs participating in the current task.

[0151] (2) Trust initialization: At the start of each new task, the initial trust level within the data source task is determined. Derived directly from its global trust level, this ensures that data sources with a good history of performance start with a higher trust level in new tasks, thus being more likely to be prioritized, creating a positive incentive.

[0152]

[0153] 2. Iterative truth discovery process.

[0154] Each iteration consists of three steps: truth calculation, weight update, and trust update. Iteration continues until the truth estimate converges to a steady state.

[0155] (1) Truth calculation. SP uses the weights from the previous round. And DS disturbance data Calculate the true value of the disturbance in the current wheel. :

[0156]

[0157] (2) Weight Update. The weights are updated based on the distance between the current perturbation truth and each DS data point. The principle is that "the closer the data is to the current truth, the higher the weight."

[0158]

[0159] in, This is a distance metric function.

[0160] (3) Trust Update. To dynamically reflect the reliability of DS, this embodiment introduces a weighted trust update mechanism. Let the weight threshold be... If the weight of a data storage device (DS) is below this threshold, its data quality or behavior is considered suspicious, and its trust level is set to zero; otherwise, the trust level is updated according to the following rules:

[0161]

[0162] in, The highest weight is selected from the trusted vehicles. If there are no trusted vehicles in the system, the highest weight is selected from all vehicles. Trust update rate Maintaining a balance between historical trust and current assessment; Distance sensitivity parameter This is used to adjust the degree of influence of weight differences on trust levels. It ensures that the greater the difference in behavior between a DS and a trusted vehicle, the faster its trust level decreases.

[0163] 3. Convergence Judgment. The iterative process continues until the following condition is met: the change in the truth estimate is less than a preset threshold. :

[0164]

[0165] When the iteration converges (assuming convergence at the nth iteration) (round), will determine the trust level of all DSs participating in this round of missions. Normalization was performed to gain their ultimate trust. :

[0166]

[0167] The ultimate truth of aggregation perturbation Ultimate trust of all DS The following is calculated by weighting the data with its disturbances:

[0168]

[0169] By introducing a trust mechanism, even if some DSs are malicious, the system can reduce their trust value, thereby reducing their weight in the final aggregation, to maintain overall accuracy.

[0170] Step 8: Global trust level update.

[0171] 1. Record the trust level data for this round of tasks. After each round of task iterations, perform a global trust level assessment. Update: During each round of task execution, the system collects the in-task trust levels of all data sources across all tasks in that round.

[0172] 2. Calculate the average trust level for this round. After a round of tasks, for each vehicle, if it participated in any tasks in this round, calculate the average trust level across all tasks in this round, and constrain the average value to within a certain range. Within the range.

[0173] 3. Update Global Trust Level. Based on a comparison between the average trust level of this round and the malicious threshold, adjust the global trust level of the data source. If it exceeds the malicious threshold, the global trust level will be increased appropriately; otherwise, it will be decreased appropriately. The final result will be further constrained to ensure it does not exceed the boundaries. The updated global trust level will be used for selecting participating vehicles in subsequent tasks and for deriving trust levels within tasks.

[0174] Step 9: Progressive privacy noise cancellation.

[0175] After the weighted and trust-based perturbation truth discovery process converges, the system obtains the perturbation truth. To restore aggregation accuracy to the greatest extent possible while protecting data privacy, the cloud server initiates a noise cancellation mechanism. Assuming the system has... Data sources Each data source participating in this task The noise added during the data perturbation phase is The final trust reached during the convergence of the truth-discovery process is The specific process for progressive privacy noise cancellation is as follows:

[0176] 1. Initialization. SP will disturb the truth. As the initial, cleansed truth: Meanwhile, SP obtains the ciphertext form of noise from various data sources from the blockchain. .

[0177] 2. Iterative noise elimination.

[0178] For each round of elimination The following operations are performed by the SP to perform blinding noise cancellation, first using the data source of the current round. The public key is used to encrypt the previously cleaned truth: Subsequently, the SP retrieved data from the blockchain storage. Encryption noise And based on the ultimate trust in this data source in the discovery of the truth. Perform noise cancellation in encrypted state:

[0179]

[0180] Using the homomorphic property of the Paillier cryptosystem, the above operation is equivalent to:

[0181]

[0182] Then by In collaboration with SP, the cleaned truth is decrypted / encrypted, and SP will then decrypt the ciphertext. Send to , requesting that it be decrypted. Decrypt the ciphertext using your own private key:

[0183]

[0184] To ensure the continuity of subsequent elimination operations, The decrypted and cleaned truth needs to be revealed. Use the next participating data source Re-encrypt using the public key:

[0185]

[0186] The re-encrypted result is returned to SP for the next round of noise cancellation.

[0187] 3. Obtaining the final result.

[0188] When all participants in the task After all data sources participated in noise cancellation, in the final round of noise cancellation, the first... Data sources Use your own private key Decryption reveals the cleaned truth Afterwards, instead of using public keys from other data sources for encryption, the decrypted result directly reveals the cleaned truth. The data is sent to the cloud server, and the SP obtains the final, cleaned truth. Since each round of elimination removes the weighted noise component of the corresponding data source, the final result can be expressed as:

[0189]

[0190] Considering the truth of the disturbance It can be represented as:

[0191]

[0192] Therefore, the truth after elimination is approximately equal to the true truth aggregated from the original data. :

[0193] .

[0194] Figures 2-4The paper presents a performance comparison of the proposed method (TAPP-TD) with four other existing technologies (PTTD-noDP, CRH, PTTD, and Median) under different task numbers, focusing on mean absolute error, root mean square error, and relative error rate. Subfigures (a) and (b) correspond to differential privacy budgets. and This demonstrates the accuracy of the truth-finding method of the present invention.

[0195] The above embodiments are preferred embodiments of this application. Those skilled in the art can make various changes or improvements based on them. Without departing from the overall concept of this application, such changes or improvements should fall within the scope of protection claimed in this application.

Claims

1. A high-precision truth discovery method for privacy protection in vehicle crowd sensing, characterized in that, include: The vehicle determines a differential privacy budget and generates noise through a smart contract. Then, it uses the vehicle's public key to encrypt the noise and the raw data perceived locally, and then submits them to the blockchain. The vehicle generates a re-encryption key based on its own private key and the cloud server's public key, and sends it to the blockchain; The smart contract homomorphically aggregates the received vehicle noise ciphertext and data ciphertext to generate perturbation data encrypted with the vehicle's public key. The smart contract uses a re-encryption key to convert the perturbation data encrypted with the vehicle's public key into perturbation data encrypted with the cloud server's public key, and sends it to the cloud server to restore the perturbation data. After obtaining disturbance data from all vehicles, the cloud server filters data sources based on the current global trust level of each vehicle. The cloud server uses data source weights and trust levels to merge disturbed data from various data sources to obtain the truth of the disturbance and updates the global trust level of each data source. The cloud server uses the disturbed truth as the initial cleaned truth, and works with all data sources to asymptotically eliminate noise from all data sources, thus obtaining the true truth aggregated from all data sources.

2. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 1, characterized in that, The vehicle determines a differential privacy budget and generates noise through a smart contract, specifically: No. vehicle Choose a differential privacy budget independently based on your own privacy needs. Send a transaction to the smart contract and submit it. ; Smart contract received Differential privacy budget Then, based on the preset global sensitivity... ,for Calculate the scale parameter of Laplace noise : ； The smart contract calls an on-chain verifiable random function to generate uniform random numbers. And calculate the unique corresponding satisfying - Differential privacy Laplace noise value : ； Smart contracts will reduce noise Returned to the corresponding [system / entity] via secure channel .

3. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 1, characterized in that, The vehicle generates a re-encryption key based on its own private key and the cloud server's public key as follows: ； In the formula, For vehicles The generated re-encryption key, For vehicles Randomly selected security parameters , Modulus The multiplicative group is a standard mathematical structure in number theory, used to constrain safety parameters. The range of values ​​ensures the security of the re-encryption process; and vehicles The public and private keys for homomorphic encryption.

4. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 3, characterized in that, The smart contract uses a re-encryption key to convert the perturbation data encrypted with the vehicle's public key into perturbation data encrypted with the cloud server's public key. Specifically: First, the perturbation data encrypted with the vehicle's public key. According to its definition, it can be expressed as: ； In the formula, It is the random number used when encrypting the vehicle. It is plaintext perturbation data; Then, use the first part of the re-encryption key. right Perform modular exponentiation to extract plaintext related terms from the perturbation data: ； Reuse Combined with re-encryption key Part Two ,Will The public key space mapped to the cloud server is ultimately used to generate ciphertext conforming to the Paillier format. This refers to the perturbation data encrypted with the cloud server's public key. : 。 5. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 4, characterized in that, Cloud servers use their own private keys Restore perturbation data , specifically: First, the cloud server encrypts the perturbation data using its own public key. Perform Paillier decryption: ； In the formula, This indicates the Paillier decryption operation. This indicates that the SP uses its own private key. Decrypting re-encrypted ciphertext The value after that, This refers to the special auxiliary functions defined in the Paillier encryption algorithm; Then, Substituting the expression into the above decryption calculation formula and simplifying, we get: ； Furthermore, because the vehicle's private key satisfies... Combined with the derived information in the re-encryption key and and The coprime property of cloud servers can eliminate The impact of this was ultimately used to reconstruct the vehicle's disturbance data. : 。 6. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 1, characterized in that, The cloud server filters data sources based on the current global trust level of each vehicle, specifically: Based on the overall trust level of each vehicle Filter out the trustworthy data sources; For each collective intelligence sensing task, recruit data sources from the above trusted data sources: if the number of trusted data sources is greater than the preset recruitment number N, then randomly select N data sources without repetition; If the number of trusted data sources is less than N, then all trusted data sources will be recruited.

7. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 1, characterized in that, The cloud server, based on data source weights and trust levels, merges disturbance data from various data sources to obtain the truth about the disturbance, specifically including: (1) Initialization; Weight initialization: for each data source initial weights Set to equal values: ； in, This refers to the vehicle number for the current crowd sensing task, i.e., the data source number. Total number of data sources; Trust initialization: For each data source Initial level of trust within the current crowd-sensing task Set to global trust : ； (2) Iterative truth discovery process; each iteration includes: truth calculation, weight update and trust update; (3) Convergence judgment: The iterative truth discovery process continues until the change in the truth estimate is less than a preset threshold. : ； In the formula, It is a distance metric function; (4) When the iteration converges, each data source Current level of trust Normalization is performed to obtain the final trust level. : ； (5) Based on the final trust level of all data sources and its disturbance data The truth of the perturbation in the final aggregate of weighted calculations : 。 8. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 7, characterized in that, The iterative truth discovery process includes: First, cloud servers use data sources. Previous round weight and perturbation data Calculate the current round The true value of the perturbation : ； Then, based on the current round The true value of the perturbation Disturbance data from various data sources Distance, update each data source In the current round Weights: ； in, It is a distance metric function; Based on weights Update the trust level of each data source. : ； in, Trust threshold; The highest weight among all data sources is selected; if there is no reliable data source, the highest weight among all vehicles is selected. To ensure reliable update rates, This is used to control the balance between historical trust levels and current assessments; For distance sensitivity parameters, This is used to adjust the degree of influence of weight differences on trust levels.

9. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 1, characterized in that, Global trust level updates for each data source include: (1) Record the trust level data of the current crowd perception task: in each round After all task iterations are completed, collect all data sources for each iteration in this round. In-task trust level in all tasks ; (2) Calculate this round Average trust level: For each data source, if in this round If someone participates in a task, their participation in this round will be calculated. The mean of in-task trust across all tasks, and constrain the mean to... Within the range; (3) Update global trust level: based on the current round The average trust level is compared with the malicious threshold, and the global trust level of the data source is adjusted: if it is greater than the malicious threshold, the global trust level of the data source is increased by a preset value, and vice versa.

10. The high-precision truth discovery method for privacy protection in vehicle crowd sensing according to claim 1, characterized in that, Cloud servers collaborate with all data sources to asymptotically eliminate noise from all data sources, specifically including: (1) Initialization: The cloud server will disturb the truth As the initial, cleansed truth: and obtain various data sources from the blockchain. Noise ciphertext ; (2) Each data source One round of noise cancellation is performed accordingly, and each round Perform the following noise cancellation operations: When performing blind noise cancellation on a cloud server, it first uses the data source of the current round. The public key is used to encrypt the previously cleaned truth: Subsequently, the cloud server retrieves the data source from the blockchain. Encryption noise And based on the ultimate trust in this data source in the discovery of the truth. Perform noise cancellation in encrypted state: ； Leveraging the homomorphic property of the Paillier cryptosystem, the above operation is equivalent to: ； Then from the data source Collaborate with the cloud server to perform cleaned truth encryption / decryption operations: The cloud server will encrypt the text... Send to data source The request is for it to decrypt; among which, the data source Use your own private key Decryption reveals the cleaned truth : ； Data source The decrypted and cleaned truth Use the next data source Re-encrypt using the public key: ； The re-encrypted result The data is returned to the cloud server for use in the next round of noise cancellation. (3) Obtaining the final result: When participating in the task After all data sources participated in noise reduction, the final... Data sources Use your own private key Decryption reveals the cleaned truth Afterwards, instead of using public keys from other data sources for encryption, the decrypted result directly reveals the cleaned truth. The data is sent to the cloud server, which then obtains the final, cleaned truth. This is the true truth of removing weighted noise components from all data sources. : 。

Images