An edge computing privacy protection method based on differential privacy

Abstract

The application discloses an edge computing privacy protection method based on differential privacy and belongs to the technical field of privacy protection. The method realizes the cooperative optimization of privacy and efficiency through four core processes: firstly, task data, computing power data, wireless channel condition data and privacy protection demand data are acquired; secondly, the data are processed and a location and use mode double-privacy protection model is established, differential privacy technology and simulated annealing algorithm are fused, a probability density function and a privacy measurement function are designed, and user task offloading frequency is analyzed; thirdly, a confused offloading proportion, an optimal confused interval upper and lower bound and a task offloading scheme after virtual mapping are generated; and finally, a task allocation proportion is determined and an offloading operation is executed, and a strategy is dynamically adjusted. The method realizes comprehensive coverage of multi-dimensional privacy, effectively reduces malicious inference risk, adapts to the characteristics of limited edge computing resources, and maintains the dynamic balance of privacy protection and task efficiency.

Description

Technical Field

[0001] This invention relates to the field of privacy protection technology, and in particular to an edge computing privacy protection method based on differential privacy. Background Technology

[0002] Edge computing, a distributed computing paradigm closely integrated with terminal devices, has been widely applied in various terminal-intensive scenarios due to its advantages of low latency and high bandwidth utilization, becoming a key hub connecting the cloud and terminal devices. This technology moves core capabilities such as data processing and storage down from the cloud to network edge nodes, effectively reducing data transmission distance and network congestion, and significantly improving task response speed, making it particularly suitable for scenarios with high real-time requirements. Meanwhile, differential privacy technology, as a mathematically provable privacy protection method, introduces controllable noise into data to prevent the leakage of individual privacy information while ensuring data availability, and has gradually become one of the core technologies in the field of data privacy protection. Simulated annealing algorithms, with their global optimization capabilities, demonstrate significant advantages in scenarios such as parameter adjustment and scheme optimization, helping the system escape local optima and find a better equilibrium state. Currently, edge computing, differential privacy, and optimization algorithms have all been maturely applied in their respective fields, and research and practice of related technologies are continuously advancing, laying a solid foundation for the integrated application of multiple technologies.

[0003] With the increasing prevalence of edge computing applications, the need for synergistic optimization of user privacy protection and task processing efficiency is becoming increasingly prominent. However, existing technologies still face numerous unresolved issues. First, privacy protection dimensions are relatively singular. Most existing solutions focus only on one type of privacy—location privacy or data privacy—failing to simultaneously cover multiple dimensions such as user usage pattern privacy, making it difficult to comprehensively defend against privacy inference attacks from malicious nodes. Second, the application of differential privacy technology in edge computing scenarios lacks targeted optimization and fails to fully leverage the resource-constrained characteristics of edge nodes, leading to unreasonable privacy budget allocation and an imbalance between insufficient privacy protection strength and a significant drop in task processing efficiency. Third, existing solutions generally lack effective integration with global optimization algorithms, making it impossible to dynamically adjust privacy protection strategies and task offloading schemes, thus hindering the maintenance of a balance between privacy protection and task efficiency in dynamically changing edge computing environments. Finally, some solutions lack comprehensive data acquisition, failing to systematically collect key data such as task characteristics, device capabilities, channel conditions, and user privacy requirements. This results in insufficient data support for subsequent model building and solution generation, affecting the overall adaptability and effectiveness of the technical solutions. The existence of these problems severely restricts the large-scale application of edge computing technology in privacy-sensitive scenarios, and there is an urgent need for a technical solution that can comprehensively cover multi-dimensional privacy protection needs, adapt to the characteristics of edge computing resources, and achieve dynamic optimization. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a privacy protection method for edge computing based on differential privacy.

[0005] The objective of this invention is achieved through the following technical solution:

[0006] A privacy-preserving edge computing method based on differential privacy is provided, which includes the following steps:

[0007] S1. Obtain relevant data, including user-generated task data, computing power data of local devices and edge servers, wireless channel condition data, and user privacy protection requirements data;

[0008] S2. Process the data obtained in S1, establish a location privacy protection model and a usage pattern privacy protection model, integrate differential privacy technology and simulated annealing algorithm, design the corresponding probability density function and privacy metric function, and analyze the user task unloading frequency.

[0009] S3. Generate the obfuscation and unloading ratio, the upper and lower bounds of the optimal obfuscation interval, and the task unloading scheme after virtual mapping based on the data processing results;

[0010] S4. Parse the generated results, determine the task allocation ratio between local computing and edge server computing, perform task unloading operations, dynamically adjust the task allocation ratio and virtual mapping strategy, and achieve a balance between privacy protection and task efficiency.

[0011] Furthermore, step S1 includes the following sub-steps:

[0012] S1.1. Obtain user-generated task data, including total number of tasks and task type;

[0013] S1.2. Based on the acquired task data, acquire the computing power data of the local device and the edge server that support task processing, including the number of CPU clock cycles per second executed by the local device and the computing power parameters of the edge server;

[0014] S1.3. Based on the task transmission requirements, obtain wireless channel condition data, including channel bandwidth, channel gain, and background noise related parameters;

[0015] S1.4. To address privacy protection needs in task processing, obtain parameters on the degree of importance users place on protecting location privacy and usage pattern privacy.

[0016] Furthermore, the establishment of the location privacy protection model in step S2 includes the following sub-steps:

[0017] S2.1. Based on differential privacy technology and acquired task data, design a probability density function for the obfuscation offloading ratio. This function is limited to a set interval and the total probability integral is 1.

[0018] S2.2. Based on the designed probability density function, a privacy metric function is constructed using the cross-entropy correlation method to calculate the degree of privacy leakage P. The calculation formula is:

[0019] ;

[0020] Where Q(r) is the probability distribution of the actual uninstallation ratio without differential privacy protection, and Pdf(r*|r) is the probability distribution of the uninstallation ratio after obfuscation under differential privacy mechanism.

[0021] S2.3. By integrating the simulated annealing algorithm with the constructed privacy metric function, an optimal confusion interval selection model is established, and the upper and lower bounds of the confusion interval are dynamically adjusted.

[0022] Furthermore, step S2, establishing the usage pattern privacy protection model, includes the following sub-steps:

[0023] S2.4. Based on the acquired user task unloading frequency data, design a probability density function for virtual mapping, which is limited to a set interval;

[0024] S2.5. Using the designed probability density function, the privacy attributes of tasks are quantified by the ratio of task unloading probabilities, and tasks that affect user profile construction are identified.

[0025] S2.6. Based on the quantified task privacy attributes, conduct targeted analysis on the user's task uninstallation frequency and filter out tasks with low uninstallation frequency changes.

[0026] Furthermore, step S3 includes the following sub-steps:

[0027] S3.1. Calculate the obfuscation offloading ratio based on the probability density function and privacy metric function in the location privacy protection model;

[0028] S3.2. Based on the calculated obfuscation unloading ratio, the optimal obfuscation interval selection model is used for iterative calculation to output the optimal upper and lower bounds of the obfuscation interval;

[0029] S3.3. Combining the upper and lower bounds of the optimal confusion interval and the user task uninstallation frequency analysis results of the usage pattern privacy protection model, perform virtual mapping processing on the selected tasks to generate a list of virtual mapped tasks and corresponding task uninstallation schemes.

[0030] Furthermore, step S4 includes the following sub-steps:

[0031] S4.1. Determine the task allocation ratio between local computing and edge server computing based on the generated obfuscation unloading ratio and the upper and lower bounds of the optimal obfuscation interval;

[0032] S4.2. According to the determined task allocation ratio and the virtual mapped task list, perform the unloading operation, and perform randomization processing on tasks with low unloading frequency.

[0033] S4.3. Monitor energy consumption and latency data during task processing after the unloading operation, and dynamically adjust task allocation ratio and virtual mapping strategy based on the monitoring data.

[0034] Furthermore, step S2.3 includes:

[0035] Based on the energy consumption factor, latency factor, privacy protection factor, and the corresponding total energy consumption, total latency, and privacy leakage degree P of the task processing, a target optimization function is constructed.

[0036] Based on the constructed objective optimization function, the initial temperature, the termination temperature and the cooling strategy are set, and new confusion interval upper and lower bound solutions are generated in the solution space.

[0037] Calculate the difference in objective function between the newly generated upper and lower bounds of the confusion interval and the current solution;

[0038] Based on the calculated difference in the objective function, determine whether to accept the new solution according to the preset criteria;

[0039] If the termination condition is not met, the temperature is lowered according to the cooling strategy, and the steps of generating a new solution, calculating the difference, and judging acceptance are repeated until the temperature drops to the termination temperature, and the optimal upper and lower bounds of the confusion interval are output.

[0040] Furthermore, step S2.6 includes:

[0041] The original user task uninstallation data is preprocessed to remove invalid data and convert it to a set format.

[0042] The preprocessed user task unloading data is loaded into the storage frame, the data is read and processed, and key-value pairs containing task identifier, user identifier and unloading frequency are output.

[0043] Based on the key-value pairs in the output, the data is summarized and sorted to filter out tasks with lower unloading frequency.

[0044] Furthermore, step S3.3 includes:

[0045] For tasks whose user task unloading frequency changes significantly, virtual mapping is performed directly.

[0046] For tasks with low user task unloading frequency, randomized response judgment is performed based on the probability density function of virtual mapping to determine whether to perform virtual mapping processing.

[0047] The virtual mapping results of tasks with large frequency changes and tasks with low frequency changes are integrated to generate a list of virtual mapped tasks and corresponding task unloading schemes.

[0048] Furthermore, step S4.3 includes:

[0049] Continuously collect task processing energy consumption data of local devices during task processing. Local computing energy consumption is calculated based on the inherent parameters of local devices, the amount of tasks processed locally, and the number of CPU cycles required for local devices to process each bit of task.

[0050] Synchronously collect task processing latency data from edge servers;

[0051] Based on the collected energy consumption and latency data, the probability density function parameters and privacy metric function parameters in the location privacy protection model and the usage pattern privacy protection model are adjusted.

[0052] Based on the adjusted model parameters, the obfuscation offloading ratio and virtual mapping strategy for subsequent tasks are optimized to maintain a balance between privacy protection and task efficiency.

[0053] The beneficial effects of this invention are:

[0054] (1) By comprehensively acquiring data, constructing a dual privacy protection model, generating optimization schemes and dynamically adjusting them, we can achieve comprehensive coverage of location and usage pattern privacy and achieve a dynamic balance between privacy protection and task efficiency.

[0055] (2) By relying on the controllable noise addition of differential privacy technology and the global optimization capability of simulated annealing algorithm, the pain point of privacy protection under limited edge computing resources is solved, and the adaptability of task offloading strategy is improved.

[0056] (3) The virtual mapping and randomization processing mechanism work together to fuzzy the task to unload key features to reduce the risk of malicious inference, while ensuring the stable and efficient operation of the edge computing system. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the specific steps of a privacy protection method for edge computing based on differential privacy.

[0058] Figure 2 A flowchart illustrating the main steps of an edge computing privacy protection method based on differential privacy, provided for an embodiment;

[0059] Figure 3 This is a schematic diagram illustrating the edge computing differential privacy offloading location privacy protection process. Detailed Implementation

[0060] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] See Figure 1 This paper presents a privacy protection method for edge computing based on differential privacy, which includes the following steps:

[0063] S1. Obtain relevant data, including user-generated task data, computing power data of local devices and edge servers, wireless channel condition data, and user privacy protection requirements data;

[0064] S2. Process the data obtained in S1, establish a location privacy protection model and a usage pattern privacy protection model, integrate differential privacy technology and simulated annealing algorithm, design the corresponding probability density function and privacy metric function, and analyze the user task unloading frequency.

[0065] S3. Generate the obfuscation and unloading ratio, the upper and lower bounds of the optimal obfuscation interval, and the task unloading scheme after virtual mapping based on the data processing results;

[0066] S4. Parse the generated results, determine the task allocation ratio between local computing and edge server computing, perform task unloading operations, dynamically adjust the task allocation ratio and virtual mapping strategy, and achieve a balance between privacy protection and task efficiency.

[0067] In some embodiments, step S1 includes the following sub-steps:

[0068] S1.1. Obtain user-generated task data, including total number of tasks and task type;

[0069] S1.2. Based on the acquired task data, acquire the computing power data of the local device and the edge server that support task processing, including the number of CPU clock cycles per second executed by the local device and the computing power parameters of the edge server;

[0070] S1.3. Based on the task transmission requirements, obtain wireless channel condition data, including channel bandwidth, channel gain, and background noise related parameters;

[0071] S1.4. To address privacy protection needs in task processing, obtain parameters on the degree of importance users place on protecting location privacy and usage pattern privacy.

[0072] In some embodiments, establishing the location privacy protection model in step S2 includes the following sub-steps:

[0073] S2.1. Based on differential privacy technology and acquired task data, design a probability density function for the obfuscation offloading ratio. This function is limited to a set interval and the total probability integral is 1.

[0074] S2.2. Based on the designed probability density function, a privacy metric function is constructed using the cross-entropy correlation method to calculate the degree of privacy leakage P. The calculation formula is:

[0075] ;

[0076] Where Q(r) is the probability distribution of the actual uninstallation ratio without differential privacy protection, and Pdf(r*|r) is the probability distribution of the uninstallation ratio after obfuscation under differential privacy mechanism.

[0077] S2.3. By integrating the simulated annealing algorithm with the constructed privacy metric function, an optimal confusion interval selection model is established, and the upper and lower bounds of the confusion interval are dynamically adjusted.

[0078] In some embodiments, establishing the usage pattern privacy protection model in step S2 includes the following sub-steps:

[0079] S2.4. Based on the acquired user task unloading frequency data, design a probability density function for virtual mapping, which is limited to a set interval;

[0080] S2.5. Using the designed probability density function, the privacy attributes of tasks are quantified by the ratio of task unloading probabilities, and tasks that affect user profile construction are identified.

[0081] S2.6. Based on the quantified task privacy attributes, conduct targeted analysis on the user's task uninstallation frequency and filter out tasks with low uninstallation frequency changes.

[0082] In some embodiments, step S3 includes the following sub-steps:

[0083] S3.1. Calculate the obfuscation offloading ratio based on the probability density function and privacy metric function in the location privacy protection model;

[0084] S3.2. Based on the calculated obfuscation unloading ratio, the optimal obfuscation interval selection model is used for iterative calculation to output the optimal upper and lower bounds of the obfuscation interval;

[0085] S3.3. Combining the upper and lower bounds of the optimal confusion interval and the user task uninstallation frequency analysis results of the usage pattern privacy protection model, perform virtual mapping processing on the selected tasks to generate a list of virtual mapped tasks and corresponding task uninstallation schemes.

[0086] In some embodiments, step S4 includes the following sub-steps:

[0087] S4.1. Determine the task allocation ratio between local computing and edge server computing based on the generated obfuscation unloading ratio and the upper and lower bounds of the optimal obfuscation interval;

[0088] S4.2. According to the determined task allocation ratio and the virtual mapped task list, perform the unloading operation, and perform randomization processing on tasks with low unloading frequency.

[0089] S4.3. Monitor energy consumption and latency data during task processing after the unloading operation, and dynamically adjust task allocation ratio and virtual mapping strategy based on the monitoring data.

[0090] In some embodiments, step S2.3 includes:

[0091] Based on the energy consumption factor, latency factor, privacy protection factor, and the corresponding total energy consumption, total latency, and privacy leakage degree P of the task processing, a target optimization function is constructed.

[0092] Based on the constructed objective optimization function, the initial temperature, the termination temperature and the cooling strategy are set, and new confusion interval upper and lower bound solutions are generated in the solution space.

[0093] Calculate the difference in objective function between the newly generated upper and lower bounds of the confusion interval and the current solution;

[0094] Based on the calculated difference in the objective function, determine whether to accept the new solution according to the preset criteria;

[0095] If the termination condition is not met, the temperature is lowered according to the cooling strategy, and the steps of generating a new solution, calculating the difference, and judging acceptance are repeated until the temperature drops to the termination temperature, and the optimal upper and lower bounds of the confusion interval are output.

[0096] In some embodiments, step S2.6 includes:

[0097] The original user task uninstallation data is preprocessed to remove invalid data and convert it to a set format.

[0098] The preprocessed user task unloading data is loaded into the storage frame, the data is read and processed, and key-value pairs containing task identifier, user identifier and unloading frequency are output.

[0099] Based on the key-value pairs in the output, the data is summarized and sorted to filter out tasks with lower unloading frequency.

[0100] In some embodiments, step S3.3 includes:

[0101] For tasks whose user task unloading frequency changes significantly, virtual mapping is performed directly.

[0102] For tasks with low user task unloading frequency, randomized response judgment is performed based on the probability density function of virtual mapping to determine whether to perform virtual mapping processing.

[0103] The virtual mapping results of tasks with large frequency changes and tasks with low frequency changes are integrated to generate a list of virtual mapped tasks and corresponding task unloading schemes.

[0104] In some embodiments, step S4.3 includes:

[0105] Continuously collect task processing energy consumption data of local devices during task processing. Local computing energy consumption is calculated based on the inherent parameters of local devices, the amount of tasks processed locally, and the number of CPU cycles required for local devices to process each bit of task.

[0106] Synchronously collect task processing latency data from edge servers;

[0107] Based on the collected energy consumption and latency data, the probability density function parameters and privacy metric function parameters in the location privacy protection model and the usage pattern privacy protection model are adjusted.

[0108] Based on the adjusted model parameters, the obfuscation offloading ratio and virtual mapping strategy for subsequent tasks are optimized to maintain a balance between privacy protection and task efficiency.

[0109] Example 2

[0110] This embodiment provides a privacy protection method for edge computing based on differential privacy. This implementation constructs a multi-dimensional privacy protection model, integrating differential privacy with a global optimization algorithm, to achieve a synergy between privacy security and task processing efficiency in edge computing resource-constrained scenarios. Figure 2 As shown, the specific implementation steps are as follows:

[0111] S1. Obtain relevant data:

[0112] S1.1. Obtain user-generated task data, including the total number of tasks and task types. In edge computing environments, user terminal devices often have to utilize edge servers to remotely process some computationally intensive tasks due to limitations in computing resources and battery life. During operation, user terminals continuously generate various computing tasks. The total number of these tasks directly reflects the scale of the user's computing needs within a specific time period. Task types differ based on characteristics such as computational complexity and data correlation. Different types of tasks have different requirements for computing resources and data transmission, which form the basis for subsequent task offloading decisions. Assume the total size of tasks generated by the user terminal is... Mobile users choose to process tasks locally. The amount of tasks unloaded to the edge server is The proportion of local computing tasks is: The proportion of computation offloaded to edge servers is: This is collectively referred to as the uninstallation ratio.

[0113] S1.2. Based on the acquired task data, obtain the computing power data of the local device and edge server supporting task processing, including the number of CPU clock cycles per second executed by the local device and the computing power parameters of the edge server. The computing power of the local device directly determines its efficiency in independently processing tasks. The number of CPU clock cycles per second executed by the local device is used as... This indicates that its computing potential can be intuitively quantified; as a node bearing task offloading, the computing power parameter of the edge server is expressed using... This indicates the parallel processing capability for large-scale unloading tasks. Obtaining these parameters based on task data ensures that computing power data matches actual task processing needs, avoiding unreasonable task allocation due to a disconnect between resource assessment and actual task requirements.

[0114] S1.3. Based on the task transmission requirements, obtain wireless channel condition data, including channel bandwidth, channel gain, and background noise parameters. The process of offloading the task from the local device to the edge server relies on the wireless channel to complete data transmission; the wireless channel status directly affects the transmission rate and stability. According to Shannon's theorem, the data transmission rate formula for using frequency division multiple access (FDMA) technology is: ,in, Represents background noise power. This represents the wireless channel gain, and the channel gain is inversely proportional to the distance. For mobile terminal transmission power, Indicates the edge server channel bandwidth. This indicates the proportion of bandwidth allocated to the mobile terminal. Channel bandwidth determines the upper limit of data that can be transmitted per unit time, while background noise can interfere with the transmitted signal, potentially causing data transmission delays or errors. Obtaining these parameters in conjunction with task transmission requirements provides data support for subsequent task transmission latency assessment and transmission strategy adjustments.

[0115] S1.4. To address privacy protection needs in task processing, obtain parameters indicating the user's level of concern regarding location privacy and usage pattern privacy. In the edge computing framework, although edge servers operate with legitimate identities, they may be curious about users' privacy information. They might infer a user's identity or geographical location based on information such as the actual offloading ratio, the real-time offloading frequency of different tasks, and the real-time wireless channel status. Different users have varying degrees of sensitivity to location privacy and usage pattern privacy. By obtaining corresponding protection level parameters, personalized privacy protection needs can be accurately matched, providing a basis for the parameter configuration of the subsequent privacy protection model.

[0116] S2. Process the acquired data:

[0117] S2.1. Based on differential privacy technology and acquired task data, design a probability density function for the obfuscation offloading ratio. This function is limited to a set interval and has a total probability integral of 1. This step is the foundational construction stage of the location privacy protection model, which includes four modules: a data preprocessing module, a probability density function construction module, a privacy measurement module, and a simulated annealing optimization module. These modules are sequentially connected: the output of the data preprocessing module serves as the input to the probability density function construction module; the output of the probability density function construction module is passed to both the privacy measurement module and the simulated annealing optimization module; and the output of the privacy measurement module is fed back to the simulated annealing optimization module, forming a closed-loop optimization chain. Differential privacy technology is a technique for protecting individual privacy during data analysis and mining. Its function is to introduce random noise into the data, preventing attackers from inferring specific individual information from the analysis results. Differential privacy controls the degree of privacy protection by defining a measure of privacy loss. In practical applications, noise is added to query results or data processing to hide the impact of individual data. Combining the acquired task data, and addressing the location privacy protection requirements during task offloading in edge computing scenarios, a probability density function for the obfuscation offloading ratio is designed. Its definition is: when hour, ;(in, ,and , Otherwise, it is 0. This function aims to obfuscate the actual uninstallation rate of users, limiting it to a set interval with a total probability integral of 1. This conforms to the basic characteristics of a probability distribution, laying a theoretical foundation for the effectiveness of subsequent privacy protection mechanisms. Key parameters of the probability density function construction module include the initial upper and lower bounds of the obfuscation interval. , The parameter values ​​are all in the range of [0,1]. The initial settings need to be determined based on the statistical distribution characteristics of the task data.

[0118] S2.2. Based on the designed probability density function, a privacy metric function is constructed using the cross-entropy correlation method to calculate the degree of privacy leakage. The calculation formula is:

[0119] ;

[0120] in, This represents the probability distribution of the actual uninstallation rate when there is no differential privacy protection. This represents the probability distribution of the obfuscated uninstallation rate under differential privacy mechanisms. When using differential privacy mechanisms to protect user location privacy, the probability that the obfuscated uninstallation rate matches the actual uninstallation rate is a key factor. The higher the probability of this match, the greater the risk of privacy leakage; conversely, the lower the probability of the match, the stronger the privacy protection. To quantify the difference between the actual uninstallation rate and the obfuscated rate, a privacy metric function is constructed using the cross-entropy correlation method. Cross-entropy can effectively measure the difference between two probability distributions, and the degree of privacy leakage calculated using the above formula is obtained. This can accurately quantify the level of privacy security under the current privacy protection mechanism, providing a quantitative reference for the optimization and adjustment of subsequent privacy protection strategies.

[0121] S2.3. Integrate the simulated annealing algorithm with the constructed privacy metric function to establish an optimal confusion interval selection model and dynamically adjust the upper and lower bounds of the confusion interval. This step is the optimization stage of the location privacy protection model. The optimal confusion interval selection model, as a sub-module of the location privacy protection model, is connected in parallel with the privacy metric module. It receives the output data from the probability density function construction module and feeds back the optimization results to the probability density function construction module to update the parameters. The simulated annealing algorithm is a general probability-based optimization algorithm. In the solid annealing process, the solid is heated to a high temperature and then slowly cooled. The internal particles gradually change from a disordered state to an ordered state, eventually reaching a stable state with the lowest energy at room temperature. The simulated annealing algorithm draws on this physical phenomenon. By controlling the temperature parameter and the acceptance probability, it performs a random search in the solution space, thus having the opportunity to escape local optima and find the global optimum. The training steps for the optimal confusion interval selection model are as follows: First, data preparation: input probability density function parameters, initial data on task processing energy consumption, initial data on latency, and initial calculation results of privacy leakage level; Second, parameter initialization: set key parameters such as initial temperature, termination temperature, cooling rate, and iteration number threshold; Third, iterative optimization: generate new solutions based on simulated annealing algorithm, calculate the objective function difference, and determine whether to accept the new solution; Fourth, convergence verification: stop training and output optimal parameters when the temperature drops to the termination temperature or the iteration number reaches the threshold.

[0122] Based on energy consumption factor, latency factor, privacy protection factor, and corresponding total energy consumption, total latency, and degree of privacy leakage in task processing. Construct the objective optimization function:

[0123] ;

[0124] in, Indicates the energy consumption factor. Indicates the delay factor. Indicates privacy protection factor, This indicates the total energy consumption for task processing. This indicates the total processing time for the task. The three factors represent the degree of privacy leakage and are the model's weight parameters, each ranging from [0,1], and satisfying the following conditions: It can be dynamically adjusted based on user privacy protection needs and task efficiency requirements. Furthermore, for... and Perform the expectation calculation to obtain , Based on this, an optimization model is constructed.

[0125] A larger confusion interval range indicates a lower risk of user privacy leakage, demonstrating a more significant effect of the privacy protection mechanism in protecting location information. However, a larger confusion interval range also leads to a decrease in the utility of task offloading decisions. This trade-off between privacy protection strength and decision utility reflects the typical characteristics of differential privacy mechanisms in edge computing scenarios. By introducing the above three factors and combining the total energy consumption, total latency, and degree of privacy leakage in task processing to construct an objective optimization function, a synergistic optimization goal of privacy protection and task efficiency can be achieved. The input data of this optimization model is a probability density function. Task processing energy consumption function Delay function and the extent of privacy breaches The output data is the optimal objective function value and the corresponding obfuscation interval parameters. The integration point between the algorithm and business logic is: to quantify the balance between privacy protection and task efficiency through the objective function, so that the optimized obfuscation interval can meet the user's privacy protection needs and ensure the efficiency of task unloading.

[0126] Based on the constructed objective optimization function, initial temperature, termination temperature, and cooling strategy are set to generate new upper and lower bound solutions for the confusion interval in the solution space. Initial temperature (in, The initial temperature coefficient has a value range of [1.5, 3.0]. (Based on the privacy protection and offloading utility under the current upper and lower bounds, and preliminarily calculated from the objective function), termination temperature. The initial temperature setting must ensure that the algorithm has a large search range in the initial stage, allowing it to fully explore the solution space; the termination temperature is one of the key conditions for terminating the algorithm's iterations, ensuring that the algorithm outputs a stable optimal solution after convergence; the cooling strategy adopts... (in, The cooling rate (with a value ranging from [0.8, 0.95]) gradually decays, allowing the algorithm to progressively focus on the optimal solution region during the search process, reducing invalid searches. Under the current temperature conditions, based on the transformation law... ( (The conversion coefficients, with values ​​ranging from [0.1, 0.3]), adjusting the upper and lower bounds to generate a new solution: , This provides a foundation for subsequent solution optimization. This stage is the execution phase of the simulated annealing algorithm. The integration of the algorithm and business logic is reflected in: controlling the search accuracy through dynamic adjustment of temperature parameters to ensure that the upper and lower bounds of the confusion interval within the solution space are found that both meet privacy protection requirements and adapt to task processing efficiency, thus avoiding parameter optimization imbalance caused by insufficient search.

[0127] Calculate the difference between the objective function of the newly generated confusion interval upper and lower bound solutions and the current solution. The difference in the objective function can intuitively reflect the optimization effect of the new solution relative to the current solution. By accurately calculating this difference, it can be determined whether the new solution is more in line with the goal of co-optimizing privacy protection and task efficiency.

[0128] Based on the calculated difference in the objective function, a pre-defined criterion is used to determine whether to accept the new solution. According to the Metropolis criterion of the simulated annealing algorithm, if the new solution is superior, then… If the new solution is good, then accept it directly; if the new solution is bad, then accept it directly. Then, based on probability The probability of accepting a new solution decreases as the temperature decreases. Applying this criterion effectively prevents the algorithm from getting trapped in local optima, ensuring that it can explore the global optimum.

[0129] If the termination condition is not met, the temperature is lowered according to the cooling strategy, and the steps of generating a new solution, calculating the difference, and judging acceptance are repeated until the temperature drops to the termination temperature. The optimal upper and lower bounds of the obfuscation interval are then output. Through multiple iterations and optimizations, the upper and lower bounds of the obfuscation interval are gradually adjusted. The final output optimal upper and lower bounds of the obfuscation interval can achieve a balance between the strength of privacy protection and the utility of task offloading, providing key parameters for the subsequent calculation of the obfuscation offloading ratio.

[0130] S2.4. Based on the acquired user task unloading frequency data, design a probability density function for virtual mapping, which is limited to a set interval. This step is the foundational construction stage of the usage pattern privacy protection model, which includes four modules: a data preprocessing module, a virtual mapping function construction module, a privacy attribute quantification module, and a frequency analysis and filtering module. These modules are sequentially connected: the output of the data preprocessing module is passed to the virtual mapping function construction module and the frequency analysis and filtering module; the output of the virtual mapping function construction module serves as the input to the privacy attribute quantification module; and the output of the privacy attribute quantification module is fed back to the frequency analysis and filtering module for targeted filtering. To enhance the protection of user usage pattern privacy, a virtual mapping mechanism is introduced. By virtually mapping different tasks, usage pattern information leakage is prevented. Based on the acquired user task unloading frequency data, a probability density function for virtual mapping is designed. Its definition is: when hour, ; ( The parameter for the virtual mapping interval is [0.05, 0.2], determined based on the dispersion of the task unloading frequency; otherwise, it is 0. This function describes the correspondence between the actual task unloading probability and the unloading probability after virtual mapping, limiting it within a set interval to ensure that the unloading probability after virtual mapping is within a reasonable range, avoiding abnormal task unloading decisions due to virtual mapping, and ensuring the stability and effectiveness of the task unloading process. The input data of the virtual mapping function construction module is the statistical data of user task unloading frequency, and the output data is the virtual mapping probability density function and its corresponding parameters. The integration point between the algorithm and business logic is: by designing the function to fuzz the distribution of the actual task unloading probability, malicious edge servers cannot infer user usage patterns through unloading frequency, while ensuring that the unloading probability after virtual mapping does not affect normal task allocation.

[0131] S2.5. Using the designed probability density function, the privacy attributes of tasks are quantified through the ratio of task unloading probabilities to identify tasks that affect user profile construction. This step is the quantization stage of the pattern privacy protection model. The training steps of the privacy attribute quantization module are as follows: First, data input: load the virtual mapping probability density function parameters and user task unloading frequency data; Second, parameter initialization: set the quantization threshold. (Values ​​range from [-0.5, 0.5], adjusted according to privacy protection requirements); the third step is quantization calculation, calculating the quantized values ​​of privacy attributes for each task based on the probability density function; the fourth step is threshold calibration, adjusting the quantization threshold using validation set data to ensure recognition accuracy meets requirements. As the number of tasks unloaded by users to the edge server increases, the user data collected by the edge server becomes richer. This data accumulation enables the edge server to build more detailed user profiles, thereby increasing the possibility of inferring user identity information from this data. Using the designed virtual mapping probability density function, the privacy attributes of tasks are quantified through the ratio of task unloading probabilities, that is, the quantified significance is used to represent the amount of privacy inherent in a certain task. The quantization formula is:

[0132] ,in, Unload the task for the user The probability, This represents the percentage of users who uninstalled the software. If the quantification result... This means that uninstalling this task will have a positive impact on the malicious edge server's user profiling; if This indicates that uninstalling this task would negatively impact the construction of user profiles on malicious edge servers. This quantification method can accurately identify tasks that affect user profile construction, providing a basis for the implementation of subsequent targeted privacy protection measures. The output data of this module is the quantified privacy attribute value and impact level of each task, providing a basis for judgment for the frequency analysis and screening module, and realizing the deep integration of algorithms and privacy protection business logic.

[0133] S2.6. Based on quantified task privacy attributes, a targeted analysis of user task uninstallation frequency is performed to filter tasks with low uninstallation frequency changes. The frequency analysis filtering module is the output module of the usage pattern privacy protection model. Its input data includes preprocessed user task uninstallation data and impact level data output by the privacy attribute quantification module. The output data is a list of tasks with low uninstallation frequency changes. The processing logic of this module is as follows: first, the raw data is cleaned and transformed, and then the results of privacy attribute quantification are combined for summarization and sorting to ensure the targeting and accuracy of the filtering. The specific processing steps are as follows: the raw user task uninstallation data is preprocessed to remove invalid data and convert it to a set format. The raw user task uninstallation data may contain missing, erroneous, or other invalid data, which will affect the accuracy of subsequent analysis results. Therefore, data cleaning is required to remove invalid or incomplete data records. At the same time, in order to adapt to the needs of subsequent data processing, the cleaned data needs to be converted into a suitable processing format.

[0134] The preprocessed user task unloading data is loaded into a storage framework, read and processed, and output key-value pairs containing task identifiers, user identifiers, and unloading frequencies. The storage framework stores and manages the preprocessed user task unloading data, ensuring efficient data retrieval and processing. During data processing, key information such as task identifiers, user identifiers, and unloading frequencies is extracted and output as key-value pairs. The key is the task identifier, and the value is a data structure containing the user identifier and the unloading frequency, facilitating subsequent aggregation and analysis of unloading frequency data for the same task.

[0135] Based on the output key-value pair data, the data is aggregated and sorted to filter out tasks with low unloading frequency fluctuations. The output key-value pair data is aggregated by task identifier, and the unloading frequency of each task is calculated over different time periods. The magnitude of the unloading frequency fluctuations is then calculated, and the tasks with low unloading frequency fluctuations are selected. Because these tasks have relatively stable unloading frequencies, they are more easily used by malicious edge servers to build user profiles; therefore, targeted privacy protection measures are necessary for them.

[0136] S3. Generate relevant solutions based on the data processing results:

[0137] S3.1. Calculate the obfuscation offloading ratio based on the probability density function and privacy metric function in the location privacy protection model. This step is the application output stage of the location privacy protection model. The input data for the model are the task data, computing power data, wireless channel condition data, and user privacy protection requirement parameters obtained in S1. The output data is the obfuscation offloading ratio. The probability density function in the location privacy protection model is: This provides a basic distributional basis for calculating the obfuscation offloading ratio, while the privacy metric function provides a quantitative evaluation standard for the reasonableness of the obfuscation offloading ratio. Combining these two functions, based on the relevant data obtained in S1 and the processing results in S2, the obfuscation offloading ratio is calculated. Differential privacy achieves individual privacy protection by adding controllable noise to the data. One of its mechanisms, the Laplace distribution, is particularly suitable for the protection of numerical data. The Laplace distribution is chosen because its mathematical properties are highly compatible with the requirements of differential privacy; this distribution can provide a controllable noise level that satisfies the requirement of indistinguishability between outputs of adjacent datasets. Privacy Budget (Values ​​range from [0.1, 1.0]) plays a crucial balancing role between privacy protection strength and data availability. It quantifies the indistinguishability of the mechanism's outputs from two adjacent datasets that differ by at most one element, and is one of the parameters of the location privacy protection model. When calculating the obfuscation offloading ratio, Laplace noise is introduced to obfuscate the true offloading ratio, preventing malicious edge servers from accurately inferring user location information from the offloading ratio, thereby achieving the purpose of location privacy protection. The integration of the algorithm and business logic here is reflected in: dynamically adjusting the privacy budget according to the user's privacy protection requirements. This ensures that the privacy protection strength of the obfuscated uninstallation ratio is precisely matched with user needs, while guaranteeing the basic effectiveness of task uninstallation.

[0138] S3.2. Based on the calculated obfuscation unloading ratio, the optimal obfuscation interval selection model is used for iterative calculation to output the optimal upper and lower bounds of the obfuscation interval. The obfuscation unloading ratio calculated in S3.1 is input into the optimal obfuscation interval selection model established in S2.3. The model combines the constructed objective optimization function: By simulating the iterative optimization process of the annealing algorithm, the upper and lower bounds of the confusion interval are adjusted and optimized. During the iteration process, the model adjusts and optimizes the boundaries based on the privacy budget. The setting dynamically adjusts the range of the confusion interval when When the value is small, the mechanism requires similar inputs to produce highly similar outputs, significantly improving privacy protection but reducing data utility; conversely, when the value is large... While improving data availability, this may sacrifice some privacy protection. To ensure that privacy protection requirements are met, the impact on task processing efficiency is minimized, and the optimal upper and lower bounds of the confusion interval are ultimately output.

[0139] S3.3. Combining the upper and lower bounds of the optimal confusion interval and the user task unloading frequency analysis results from the usage pattern privacy protection model, virtual mapping is performed on the selected tasks to generate a virtual-mapped task list and corresponding task unloading schemes. This step is the collaborative application stage of the two privacy protection models. The input data are the upper and lower bounds of the optimal confusion interval output by the location privacy protection model, the task list with low unloading frequency variation output by the usage pattern privacy protection model, and the virtual mapping probability density function parameters. The output data are the virtual-mapped task list and task unloading schemes. For tasks with large variations in user task unloading frequency, virtual mapping is performed directly. Tasks with large variations in user task unloading frequency exhibit strong randomness in their unloading behavior. Malicious edge servers find it difficult to construct stable user profiles through the unloading frequency of such tasks. However, to further improve the comprehensiveness of usage pattern privacy protection, virtual mapping is performed directly on them, increasing the uncertainty of unloading behavior by adjusting the task's unloading probability.

[0140] For tasks with low user task unloading frequency, a probability density function based on virtual mapping is used. A randomized response mechanism is used to determine whether to perform virtual mapping processing. Because these types of tasks have a relatively stable offloading frequency, they are more susceptible to malicious exploitation. Therefore, a randomized response mechanism is employed, randomly generating response results based on the probability density function of virtual mapping to decide whether to perform virtual mapping processing. This approach effectively protects privacy while avoiding a decrease in task offloading efficiency due to excessive virtual mapping, ensuring a balance between privacy protection and operational efficiency.

[0141] The virtual mapping processing results of tasks with large and small frequency variations are integrated to generate a virtual-mapped task list and corresponding task unloading schemes. The virtual mapping processing results of the two types of tasks are summarized and integrated to clarify the unloading status and probability of each task after virtual mapping, forming a virtual-mapped task list. Simultaneously, based on the upper and lower bounds of the optimal obfuscation interval and the obfuscation unloading ratio, corresponding task unloading schemes are formulated, clarifying which tasks are unloaded to edge servers, which tasks are processed locally, and the specific unloading ratios, providing clear guidance for subsequent task execution.

[0142] S4. Parse the results and perform adjustments:

[0143] S4.1. Based on the generated obfuscation offloading ratio and the upper and lower bounds of the optimal obfuscation interval, determine the task allocation ratio between local computing and edge server computing. The obfuscation offloading ratio clarifies the overall range of task offloading, while the upper and lower bounds of the optimal obfuscation interval further constrain this ratio. Combining these two parameters, and referring to the computing power data of the local device and edge server obtained in S1, determine the task allocation ratio between local computing and edge server computing. In an edge computing architecture, edge servers can provide service support for a large-scale mobile terminal user group. By migrating computationally intensive tasks to near-end edge servers, users can significantly optimize their energy consumption efficiency and reduce system response latency. For local devices with strong computing power, the proportion of tasks performed locally can be appropriately increased; for edge servers with strong computing power, the proportion of tasks offloaded to edge servers can be appropriately increased, ensuring that the task allocation ratio matches the computing power of both parties and improving overall task processing efficiency.

[0144] S4.2. Based on the determined task allocation ratio and the virtually mapped task list, perform the unloading operation, and perform randomization processing on tasks with low unloading frequency changes. According to the task allocation ratio, the tasks to be unloaded are transmitted to the edge server according to the virtually mapped task list, and the unloading operation is performed. In edge computing environments, user terminal devices often have to utilize edge servers to remotely process some computationally intensive tasks due to limitations in computing resources and battery life. This task unloading strategy can effectively save local resources. During the unloading process, tasks with low unloading frequency changes are additionally randomized, and their unloading behavior is further obscured by introducing random noise, reducing the risk of malicious analysis. Simultaneously, during task transmission, the wireless channel condition data and transmission rate formula obtained in S1 are used: The transmission parameters are dynamically adjusted to ensure the stability and efficiency of data transmission.

[0145] S4.3. Monitor energy consumption and latency data during task processing after the unloading operation, and dynamically adjust task allocation ratios and virtual mapping strategies based on the monitoring data. This step is the dynamic optimization stage of model parameters. It adjusts key parameters of the two privacy protection models through real-time monitoring data feedback to achieve a long-term stable balance between privacy protection and task efficiency. Continuously collect task processing energy consumption data from the local device during task processing. Local energy consumption is calculated using the formula: Calculate, where, Indicates the inherent parameters of the local device. This indicates the amount of work being processed locally. This indicates the number of CPU cycles required for the local device to process each bit of task. This represents the total number of cycles required for local computation; synchronously collect task processing latency data from edge servers, and the total processing latency of edge servers is: ,in, The time allotted for edge server task processing For data transmission latency, the return time of the calculation results is ignored when calculating the total latency because the amount of data generated by the task's calculation results is much smaller than the amount of data that the user unloads to the edge server. Local devices consume energy during task processing. By continuously collecting energy consumption data and combining it with the above formula, the local computing energy consumption can be accurately calculated. Simultaneously, latency data of the edge server processing the unloading task is collected. Energy consumption and latency data together constitute the evaluation index of task processing efficiency, serving as input for adjusting model parameters.

[0146] Based on the collected energy consumption and latency data, the probability density function parameters and privacy metric function parameters in the location privacy protection model and usage pattern privacy protection model were adjusted. Specifically, if the collected energy consumption data is too high, it indicates excessive local computational pressure, requiring an increase in the obfuscation offloading ratio. Correspondingly, the upper and lower bounds of the obfuscation interval in the location privacy protection model were adjusted. , At the same time, increase privacy budget To balance privacy protection and offloading efficiency; if the latency data is too high, it indicates that the edge server is overloaded, and the obfuscation offloading ratio needs to be reduced, the upper and lower bounds of the obfuscation interval adjusted, and the privacy budget reduced. If the privacy breach assessment result is higher than the security threshold, the privacy budget needs to be reduced. Adjust the parameters of the virtual mapping function. and quantization threshold To enhance privacy protection, if the collected energy consumption data is too high or the latency data is too large, it indicates that the current task allocation ratio and virtual mapping strategy are unreasonable, and the location privacy protection model needs to be improved. Privacy protection model for usage patterns The parameters of the privacy metric function are also adjusted. For example, if the edge server's processing latency is too high, the obfuscation offloading ratio can be appropriately reduced to decrease the amount of tasks offloaded to the edge server; if the local device's power consumption is too high, the obfuscation offloading ratio can be appropriately increased to increase the amount of tasks offloaded. Parameter adjustments optimize task processing efficiency. Here, the integration of the algorithm and business logic is manifested in establishing a dynamic mapping relationship between monitoring data and model parameters, enabling the model to adapt to changes in the task processing environment and maintain a long-term balance between privacy protection and task efficiency.

[0147] Based on the adjusted model parameters, the obfuscation offloading ratio and virtual mapping strategy for subsequent tasks are optimized to maintain a balance between privacy protection and task efficiency. By adjusting the model parameters, the obfuscation offloading ratio for subsequent tasks is recalculated, the virtual mapping strategy is optimized, and the task allocation ratio is adjusted to keep energy consumption and latency within a reasonable range during task processing, while ensuring that the strength of privacy protection meets user needs. The simulated annealing algorithm has strong global search capabilities and can avoid getting trapped in local optima by accepting inferior solutions. This algorithm is used to optimize the adjusted parameters to ensure that subsequent task processing can continuously maintain a dynamic balance between privacy protection and task efficiency.

[0148] This technical solution achieves comprehensive multi-dimensional privacy protection by constructing location privacy protection models and usage pattern privacy protection models, integrating differential privacy and simulated annealing algorithms, effectively solving the problem of single privacy protection dimensions in traditional solutions. By introducing controllable noise through differential privacy technology, combined with the probability density function of the obfuscation offloading ratio and the probability density function of virtual mapping, the actual user offloading ratio and task offloading frequency are blurred, significantly reducing the risk of malicious edge servers inferring user location and usage pattern information. The simulated annealing algorithm is integrated to construct an optimal obfuscation interval selection model, dynamically adjusting privacy protection parameters through an objective optimization function. This ensures privacy security while minimizing the impact on task processing efficiency, achieving a precise balance between privacy protection and task efficiency. Utilizing quantitative tools such as local computing energy consumption formulas and wireless channel transmission rate formulas, the resource consumption and transmission efficiency of task processing are accurately evaluated, providing a scientific basis for optimizing task allocation ratios and avoiding problems such as unreasonable privacy budget allocation and impaired model performance that may occur when differential privacy technology is directly applied to edge computing. The entire solution is complete and logically coherent. Through a dynamic adjustment mechanism, it continuously optimizes task processing and privacy protection strategies, ensuring that the system can maintain a stable and good operating status and privacy protection level during long-term operation. At the same time, through the synergistic effect of multi-dimensional privacy protection models, it comprehensively covers the protection needs of location privacy and usage pattern privacy, and is adapted to the characteristics of edge computing resource-constrained scenarios.

[0149] Example 3

[0150] In specific implementation, such as Figure 3 As shown, users generate various computing tasks in an edge computing environment. These tasks need to be reasonably allocated between local devices and edge servers to optimize processing efficiency. However, during task unloading, the user's real location information may be inferred by the edge server based on the unloading ratio. To address this issue, this embodiment integrates differential privacy technology with optimization algorithms to construct a complete privacy protection process, as detailed below:

[0151] First, perform the data acquisition step to collect basic data such as the total number of user-generated tasks and task types, and determine the total size of tasks generated by the user. At the same time, acquire computing resource data such as the number of CPU clock cycles per second executed by the local device and the computing power parameters of the edge server, as well as wireless channel condition data such as channel bandwidth, channel gain, and background noise. In addition, it is also necessary to collect parameters on the degree of importance users attach to the protection of location privacy, so as to provide data support for the subsequent construction of privacy protection models.

[0152] Based on the acquired data, a location privacy protection model is established. User-generated tasks need to be allocated a ratio of local computation to edge server computation through offloading decisions. The proportion of tasks requiring local computation is determined by the ratio of the task volume to the total task volume, and the remaining portion should ideally be offloaded to the edge server. However, if the actual offloading ratio is directly used for transmission, the edge server may use information such as wireless channel status to infer the user's true location. Therefore, a differential privacy offloading strategy needs to be introduced to obfuscate the actual offloading ratio.

[0153] A probability density function for obfuscating the uninstallation ratio is designed, constrained within a set interval with a total probability integral of 1. By adding Laplace noise to the actual uninstallation ratio, an obfuscated uninstallation ratio is generated. A privacy metric function is constructed using the cross-entropy correlation method to calculate the degree of privacy leakage, quantifying the difference between the actual uninstallation ratio and the obfuscated ratio, ensuring that obfuscation effectively hides the user's actual uninstallation behavior. An optimal obfuscation interval selection model is established by incorporating simulated annealing algorithm. Initial temperature, termination temperature, and cooling strategy are set, generating new upper and lower bound solutions for the obfuscation interval in the solution space. The difference in the objective function is calculated, and a new solution is accepted according to a preset criterion. This iteration is repeated until the temperature drops to the termination temperature, outputting the optimal upper and lower bounds of the obfuscation interval, achieving a balance between privacy protection strength and task uninstallation utility.

[0154] After model construction is completed, the obfuscation offloading ratio and the upper and lower bounds of the optimal obfuscation interval are generated. Combined with the user task offloading frequency analysis results, a task offloading scheme after virtual mapping is formulated. During the actual offloading process, the task is transmitted to the edge server via wireless channel according to the obfuscated offloading ratio. At this time, the edge server can only obtain the obfuscated offloading ratio and cannot accurately infer the user's real location from this ratio, thus achieving effective protection of location privacy.

[0155] During task processing, the energy consumption data of local devices and the latency data of edge servers are continuously monitored. Local computing energy consumption is calculated based on the inherent parameters of the device, the amount of local processing tasks, and the number of CPU cycles required per bit of task. The total latency of the edge servers includes data transmission latency and task processing latency. Based on the monitoring data, the probability density function parameters and privacy metric function parameters in the location privacy protection model are dynamically adjusted to optimize the obfuscation offloading ratio and virtual mapping strategy for subsequent tasks, ensuring a dynamic balance between privacy protection and task efficiency throughout the task processing process.

[0156] Through the above implementation process, while achieving efficient task offloading and reducing local device power consumption and task response latency, it effectively prevents edge servers from inferring the user's real location through the offloading ratio. This achieves synergistic optimization of location privacy protection and task processing efficiency, adapts to the characteristics of edge computing resource-constrained scenarios, and provides a feasible implementation path for privacy protection in edge computing environments.

[0157] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A privacy protection method for edge computing based on differential privacy, characterized in that, Includes the following steps: S1. Obtain relevant data, including user-generated task data, computing power data of local devices and edge servers, wireless channel condition data, and user privacy protection requirements data; S2. Process the data obtained in S1, establish a location privacy protection model and a usage pattern privacy protection model, integrate differential privacy technology and simulated annealing algorithm, design the corresponding probability density function and privacy metric function, and analyze the user task unloading frequency. S3. Generate the obfuscation and unloading ratio, the upper and lower bounds of the optimal obfuscation interval, and the task unloading scheme after virtual mapping based on the data processing results; S4. Parse the generated results, determine the task allocation ratio between local computing and edge server computing, perform task unloading operations, dynamically adjust the task allocation ratio and virtual mapping strategy, and achieve a balance between privacy protection and task efficiency. The establishment of the location privacy protection model in step S2 includes the following sub-steps: S2.

1. Based on differential privacy technology and acquired task data, design a probability density function for the obfuscation offloading ratio. This function is limited to a set interval and the total probability integral is 1. S2.

2. Based on the designed probability density function, a privacy metric function is constructed using the cross-entropy correlation method to calculate the degree of privacy leakage P. The calculation formula is: Where Q(r) is the probability distribution of the actual uninstallation ratio without differential privacy protection, and Pdf(r*|r) is the probability distribution of the uninstallation ratio after obfuscation under differential privacy mechanism. S2.

3. By integrating the simulated annealing algorithm with the constructed privacy metric function, an optimal confusion interval selection model is established, and the upper and lower bounds of the confusion interval are dynamically adjusted. Step S2, which establishes the usage pattern privacy protection model, includes the following sub-steps: S2.

4. Based on the acquired user task unloading frequency data, design a probability density function for virtual mapping, which is limited to a set interval; S2.

5. Using the designed probability density function, the privacy attributes of tasks are quantified by the ratio of task unloading probabilities, and tasks that affect user profile construction are identified. S2.

6. Based on the quantified task privacy attributes, conduct targeted analysis on the user's task uninstallation frequency and filter out tasks whose uninstallation frequency fluctuation is less than a preset threshold.

2. The method according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1.

1. Obtain user-generated task data, including total number of tasks and task type; S1.

2. Based on the acquired task data, acquire the computing power data of the local device and the edge server that support task processing, including the number of CPU clock cycles per second executed by the local device and the computing power parameters of the edge server; S1.

3. Based on the task transmission requirements, obtain wireless channel condition data, including channel bandwidth, channel gain, and background noise related parameters; S1.

4. To address privacy protection needs in task processing, obtain parameters on the degree of importance users place on protecting location privacy and usage pattern privacy.

3. The method according to claim 1, characterized in that, Step S3 includes the following sub-steps: S3.

1. Calculate the obfuscation offloading ratio based on the probability density function and privacy metric function in the location privacy protection model; S3.

2. Based on the calculated obfuscation unloading ratio, the optimal obfuscation interval selection model is used for iterative calculation to output the optimal upper and lower bounds of the obfuscation interval; S3.

3. Combining the upper and lower bounds of the optimal confusion interval and the user task uninstallation frequency analysis results of the usage pattern privacy protection model, perform virtual mapping processing on the selected tasks to generate a list of virtual mapped tasks and corresponding task uninstallation schemes.

4. The method according to claim 1, characterized in that, Step S4 includes the following sub-steps: S4.

1. Determine the task allocation ratio between local computing and edge server computing based on the generated obfuscation unloading ratio and the upper and lower bounds of the optimal obfuscation interval; S4.

2. According to the determined task allocation ratio and the virtual mapped task list, perform the unloading operation, and perform randomization processing on tasks whose unloading frequency fluctuation is less than the preset threshold. S4.

3. Monitor energy consumption and latency data during task processing after the unloading operation, and dynamically adjust task allocation ratio and virtual mapping strategy based on the monitoring data.

5. The method according to claim 3, characterized in that, Step S2.3 includes: Based on the energy consumption factor, latency factor, privacy protection factor, and the corresponding total energy consumption, total latency, and privacy leakage degree P of the task processing, a target optimization function is constructed. Based on the constructed objective optimization function, the initial temperature, the termination temperature and the cooling strategy are set, and new confusion interval upper and lower bound solutions are generated in the solution space. Calculate the difference in objective function between the newly generated upper and lower bounds of the confusion interval and the current solution; Based on the calculated difference in the objective function, determine whether to accept the new solution according to the preset criteria; If the termination condition is not met, the temperature is lowered according to the cooling strategy, and the steps of generating a new solution, calculating the difference, and judging acceptance are repeated until the temperature drops to the termination temperature, and the optimal upper and lower bounds of the confusion interval are output.

6. The method according to claim 1, characterized in that, Step S2.6 includes: The original user task uninstallation data is preprocessed to remove invalid data and convert it to a set format. The preprocessed user task unloading data is loaded into the storage frame, the data is read and processed, and key-value pairs containing task identifier, user identifier and unloading frequency are output. Based on the output key-value pair data, the data is summarized and sorted to filter out tasks whose uninstallation frequency fluctuations are less than a preset threshold.

7. The method according to claim 3, characterized in that, Step S3.3 includes: For user tasks whose unloading frequency fluctuations exceed a preset threshold, virtual mapping processing is executed directly. For user tasks whose unloading frequency fluctuation is less than a preset threshold, randomized response judgment is performed based on the probability density function of virtual mapping to determine whether to perform virtual mapping processing. The virtual mapping results of tasks with large frequency changes and tasks with low frequency changes are integrated to generate a list of virtual mapped tasks and corresponding task unloading schemes.

8. The method according to claim 4, characterized in that, Step S4.3 includes: Continuously collect task processing energy consumption data of local devices during task processing. Local computing energy consumption is calculated based on the inherent parameters of local devices, the amount of tasks processed locally, and the number of CPU cycles required for local devices to process each bit of task. Synchronously collect task processing latency data from edge servers; Based on the collected energy consumption and latency data, the probability density function parameters and privacy metric function parameters in the location privacy protection model and the usage pattern privacy protection model are adjusted. Based on the adjusted model parameters, the obfuscation offloading ratio and virtual mapping strategy for subsequent tasks are optimized to maintain a balance between privacy protection and task efficiency.

Images