Sensitive data privacy protection computing method in cloud environment

Abstract

The invention discloses a sensitive data privacy protection calculation method in a cloud environment, and the method comprises the steps: collecting the flow data of an electric power Internet of Things, and carrying out the structural processing of the collected flow data; under a privacy computing technical framework, performing multi-dimensional privacy feature extraction on the flow data after the structured processing, and identifying sensitive information including a user address, a name and a contact number; wherein a user address in the traffic data is identified by adopting an address self-adaptive identification algorithm; and marking and grading the identified sensitive information, and carrying out monitoring protection according to a grading result. According to the method, a brand-new address tree adaptive identification technology is provided in combination with traffic data characteristics of the power Internet of Things, sensitive data in a cloud computing environment can be rapidly identified, a privacy computing technology is deeply fused, a multi-level protection system is constructed in the full life cycle of the data, and the security and privacy of the sensitive data are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power Internet of Things, and in particular to a sensitive data privacy protection calculation method in a cloud environment. Background Art

[0002] With the continuous advancement of digitalization, cloud computing, and intelligentization in the power Internet of Things (IoT), and the widespread use of intelligent terminals, including smart meters, temperature and humidity sensors, and power distribution equipment, an increasing amount of data collected by these terminals is being transmitted in cloud computing environments. This data contains sensitive information such as user addresses, names, electricity usage, and devices. While the widespread development of the IoT has brought significant convenience to people's lives, the accompanying data privacy issues in cloud computing have also become a major concern.

[0003] These sensitive data need to be properly protected and managed to prevent data leakage and destruction. Necessary security measures and technical means need to be taken in the generation, transmission, use, storage, and destruction of these sensitive data to ensure the security and privacy of sensitive data. Summary of the Invention

[0004] The purpose of this invention is to provide a sensitive data privacy protection calculation method in a cloud environment. Combined with the traffic data characteristics of the power Internet of Things, a new address tree adaptive recognition technology is proposed, which can quickly identify sensitive data in a cloud computing environment and deeply integrate privacy computing technology to build a multi-level protection system throughout the data life cycle to ensure the security and privacy of sensitive data.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0006] The present invention discloses a sensitive data privacy protection calculation method in a cloud environment, the method comprising the following steps:

[0007] S1, collects flow data of the power Internet of Things and performs structured processing on the collected flow data;

[0008] S2, under the privacy computing technology framework, extracts multi-dimensional privacy features from the structured traffic data to identify sensitive information such as user address, name, and contact number. An adaptive address recognition algorithm is used to identify user addresses in the traffic data. The recognition process includes the following sub-steps:

[0009] S21, target data segmentation; specifically, segmenting the text information obtained in the traffic data to obtain address segmentation segments at multiple levels of administrative districts;

[0010] S22, initialize the standard address database; specifically, import the standard address database, which includes detailed address information of provinces, cities, districts and streets across the country and corresponding address hash data;

[0011] S23, global hash matching of the address tree; specifically, obtaining the hash data corresponding to the target address to be identified; comparing the hash data of the target address with the hash data of the address in the standard address library one by one. If equal data is found, it is determined that address-sensitive data exists in this traffic data and the address information is output, and the process proceeds to step S3; otherwise, the process proceeds to step S24;

[0012] S24, address tree node depth-breadth weight matching; hash the target address segment to be identified, and perform breadth and depth searches on the hash data of multiple address segments of the target address to be identified in descending order of administrative district level based on the standardized address database. If each address segment can match equal data, it is determined that address-sensitive data exists in this traffic data and the address information is output, and the process proceeds to step S3; otherwise, the process proceeds to step S25;

[0013] S25: For different administrative district levels, corresponding matching weights are introduced. Combined with the hash data of multiple address word segments of the target address to be identified, the hierarchical sensitivity coefficient of the depth and breadth of the address tree node is calculated. If the hierarchical sensitivity coefficient of the depth and breadth of the address tree node exceeds a preset threshold, it is determined that this piece of traffic data of the power Internet of Things contains address-sensitive data and the address information is output. The process then proceeds to step S3. Otherwise, it is determined that this piece of traffic data of the power Internet of Things does not contain address-sensitive data, and the process ends.

[0014] S3, marks and classifies the identified sensitive information, and monitors and protects it based on the classification results.

[0015] Furthermore, in step S1, a bypass mode is adopted to copy the current network traffic of the cloud Internet in real time from the power Internet of Things, the north-south network traffic of the unified entrance and exit of the external network, and the data communication traffic between the inside and outside of the network, that is, the network traffic that vertically crosses the network boundary.

[0016] Furthermore, in step S1, the process of performing structured processing on the collected traffic data includes the following steps:

[0017] Clean the collected traffic data to remove invalid, expired, duplicate and erroneous data;

[0018] Convert the cleaned traffic data into a unified format;

[0019] Add common parameters including storage time and collection time to the traffic data after format conversion.

[0020] Furthermore, in step S2, the name recognition process includes the following steps:

[0021] Match the surnames against hundreds of surnames, then filter them based on single-character and double-character Chinese names to preliminarily identify all possible Chinese names in the sentence.

[0022] Match the initially recognized Chinese names with the words in the dictionary and output all possible combinations;

[0023] The process of identifying a contact number includes the following steps:

[0024] Convert the input text to a unified format, removing punctuation and spaces;

[0025] Use regular expressions to find strings that match the mobile phone number format;

[0026] The secure multi-party computing protocol in privacy computing is introduced to perform distributed collaborative verification of mobile phone number matching rules.

[0027] Furthermore, in step S2, the text information obtained in the flow data of the power Internet of Things is segmented to obtain address segmentation segments of multiple levels of administrative districts. If some of the address segmentation segments are empty, they are filled with null values ​​to obtain:

[0028] S = {S1, S2, S3, S4, S5};

[0029] Where S is the target address information to be identified, S1 represents the province data to be identified, S2 represents the city data to be identified, S3 represents the district data to be identified, S4 represents the street data to be identified, and S5 represents the detailed address data to be identified;

[0030] Import the standard address library D:

[0031]

[0032] Where m represents the total number of all complete addresses in the address database; Indicates the name of the i-th level administrative district in the j-th address, D1 represents the province data of the standard address, D2 represents the city data of the standard address, D3 represents the district data of the standard address, D4 represents the street data of the standard address, and D5 represents the detailed address data of the standard address;

[0033] Hash the address data in the standard address library D to obtain the corresponding address hash data:

[0034]

[0035] Where Hd It is the hash information in the address information of the standard address library. HD1 represents the hash data of a province, HD2 represents the hash data of a city, HD3 represents the hash data of a district, HD4 represents the hash data of a street, and HD5 represents the hash data of a detailed address.

[0036] Furthermore, in step S23, the process of global hash matching of the address tree includes the following steps:

[0037] S231, hash the target address to be identified in the power Internet of Things traffic to obtain the following three types of data: hash data of the province, city, district and street detailed address S h1 =Hash*(S1+S2+S3+S4+S5), Hash data S of the district street address h2 =Hash*(S3+S4+S5), Hash data S of street address h3 =Hash*(S4+S5);

[0038] S232, S h1 and D h1 Compare and if they are equal, it is found that this traffic data contains address-sensitive data and the address information is a province S1, a city S2, a district S3, a street S4, and a detailed address S5; otherwise, go to step S233;

[0039] S233, S h2 and D h2 Compare and if they are equal, it is found that this traffic data contains address-sensitive data and the address information is a certain district S3, a certain street S4, and a certain detailed address S5. Otherwise, go to step S234;

[0040] S234, S h3 and D h3 Compare them. If they are equal, it is found that this traffic data contains address-sensitive data and the address information is a certain street S4 and a certain detailed address S5;

[0041] Where D h1 This is the hash data obtained by splicing the detailed address information of a province, city, district and street in the power Internet of Things through traffic; D h2 The hash data after splicing the detailed address information of a certain district street obtained through traffic in the power Internet of Things; D h3 This is hash data obtained by splicing the detailed address information of a street obtained through traffic in the power Internet of Things.

[0042] Step S24 further includes:

[0043] S241, hash the target address segment to be identified, using the hash value H of S1 s1 Perform breadth search until the search is found s1 Same standard address save data H d1 , go to step S242, if the same data cannot be found, go to step S25;

[0044] S242, using the hash value H of S2 s2 Perform a deep search until you find the one that matches H s2 Same standard address city data H d2 , go to step S243, if the same data cannot be found, go to step S25;

[0045] S243, using the hash value H of area S3 s3 Perform a deep search until you find the same s3 Same standard address area data H d3 , go to step S244, if the same data cannot be found, go to step S25;

[0046] S244, using the hash value H of street S4 s4 Perform a deep search until you find the same s4 Same standard address street data H d4 , go to step S245, if the same data cannot be found, go to step S25;

[0047] S245, using the hash value H of the detailed address S5 s5 Perform a deep search until you find the one that matches H s5 Same standard address detailed address data H d5 , go to step S246, if the same data cannot be found, go to step S25;

[0048] S246, Comprehensive H d1 、H d2 、H d3 、H d4 and H d5 , and obtain the identification result of the target address to be identified.

[0049] Step S25 further includes:

[0050] The sensitive address information to be identified in the traffic of the power Internet of Things is segmented to obtain multiple administrative district level address data including provincial, municipal, district, street and detailed address information. The address data at each level is hashed to obtain a unique key value H. S ;

[0051] Through the known standard address database, we build a full address information of multiple levels, including full provincial address information, full city address information, full district address information, full street address information and full detailed address information. We perform Hash calculation on the full address information to obtain the corresponding full Key value, and form a national address Hash address tree. Each node of the tree is H d ;

[0052] Perform a weighted comparison of the hash values ​​of the address-sensitive information to be identified in the traffic of the power Internet of Things and the address tree of the national address Hash, and obtain the hierarchical sensitivity coefficient through hierarchical dependency and weighted average.

[0053]

[0054] Where, is the hierarchical sensitivity coefficient, which is used to represent the comprehensive sensitivity score of address data; is the dynamic weight of the i-th level administrative district, which is determined by the preset weight Dynamic adjustment of the security risk coefficient corresponding to the corresponding administrative area; i Indicates the hash matching similarity of the i-th level administrative district address; Indicates progressive hierarchical dependency, where k is the upper level number and s κ is the matching degree of the k-th level address; λ represents the level attenuation compensation factor, which is negatively correlated with the data traffic complexity; σ() represents the dynamic threshold function.

[0055] Furthermore, the preset threshold adopts an adaptive sensitive threshold, which is the boundary value of the sensitivity of the address to be identified as sensitive information found in the traffic of the power Internet of Things. Through the joint modeling of security risks, traffic complexity, and user policies, the sensitive data judgment level is adaptively adjusted.

[0056] Furthermore, the calculation formula of the adaptive sensitivity threshold AST is:

[0057]

[0058] in, It is the security risk factor, which is related to the attack frequency and vulnerability exposure level; The complexity of data flow is related to the proportion of unstructured text and the depth of address segmentation level; is the user policy weight, which is related to the user-defined sensitivity level; β, γ, and δ are weight factors, and β+γ+δ=1;

[0059] like is less than the adaptive sensitivity threshold, excluding the addresses to be identified found in the traffic of the power Internet of Things as sensitive information; if If the value is greater than or equal to the adaptive sensitivity threshold, the address to be identified found in the traffic of the power Internet of Things is determined to be sensitive information.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] First, the sensitive data privacy protection calculation method in the cloud environment of the present invention uses address tree adaptive recognition technology to accurately and quickly identify the address information in the sensitive data in the power Internet of Things traffic, as well as name matching and mobile phone number recognition during the data life cycle of transmission, use, storage, and destruction.

[0062] Second, the sensitive data privacy protection calculation method in the cloud environment of the present invention, in the sensitive address identification scenario of the power Internet of Things, the "hierarchical sensitivity coefficient" algorithm improves the judgment accuracy through "dynamic adjustment and progressive level", thereby realizing "the higher the risk, the stricter it is, and the more chaotic the data, the more detailed it is" adaptive protection in a complex network environment.

[0063] Third, the sensitive data privacy protection calculation method in the cloud environment of the present invention adopts strong encryption technology and secure transmission protocol during the transmission process to protect the confidentiality, integrity and availability of sensitive data, and introduces homomorphic encryption technology to realize the secret transmission and calculation of sensitive data; a multiple authentication mechanism is adopted during use to ensure that only authorized users can access sensitive data, and at the same time, a privacy computing sandbox is built based on the trusted execution environment (TEE) to support the secure collaborative analysis of multi-party data; during the storage process, an asymmetric RSA encryption algorithm is used to process the data, and differential privacy technology is used to add noise disturbance to the aggregated data to ensure that the real content cannot be obtained even if the data is illegally accessed during the storage process. At the same time, a strict access control mechanism is established, and only authorized personnel can access and operate sensitive data. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flow diagram of the sensitive data privacy protection calculation method in a cloud environment of the present invention;

[0065] Figure 2 This is the flow chart for flow data processing;

[0066] Figure 3 This is the flow chart of the address adaptive recognition algorithm;

[0067] Figure 4 It is a complete address tree map at the provincial, municipal, district, street and detailed address levels. DETAILED DESCRIPTION

[0068] The embodiments of the present invention are described in further detail below with reference to the accompanying drawings.

[0069] The present invention discloses a sensitive data privacy protection calculation method in a cloud environment, the method comprising the following steps:

[0070] S1, collects flow data of the power Internet of Things and performs structured processing on the collected flow data;

[0071] S2, under the privacy computing technology framework, extracts multi-dimensional privacy features from the structured traffic data to identify sensitive information such as user address, name, and contact number. An adaptive address recognition algorithm is used to identify user addresses in the traffic data. The recognition process includes the following sub-steps:

[0072] S21, target data segmentation; specifically, segmenting the text information obtained in the traffic data to obtain address segmentation segments at multiple levels of administrative districts;

[0073] S22, initialize the standard address database; specifically, import the standard address database, which includes detailed address information of provinces, cities, districts and streets across the country and corresponding address hash data;

[0074] S23, global hash matching of the address tree; specifically, obtaining the hash data corresponding to the target address to be identified; comparing the hash data of the target address with the hash data of the address in the standard address library one by one. If equal data is found, it is determined that address-sensitive data exists in this traffic data and the address information is output, and the process proceeds to step S3; otherwise, the process proceeds to step S24;

[0075] S24, address tree node depth-breadth weight matching; hash the target address segment to be identified, and perform breadth and depth searches on the hash data of multiple address segments of the target address to be identified in descending order of administrative district level based on the standardized address database. If each address segment can match equal data, it is determined that address-sensitive data exists in this traffic data and the address information is output, and the process proceeds to step S3; otherwise, the process proceeds to step S25;

[0076] S25: For different administrative district levels, corresponding matching weights are introduced. Combined with the hash data of multiple address word segments of the target address to be identified, the hierarchical sensitivity coefficient of the depth and breadth of the address tree node is calculated. If the hierarchical sensitivity coefficient of the depth and breadth of the address tree node exceeds a preset threshold, it is determined that this piece of traffic data of the power Internet of Things contains address-sensitive data and the address information is output. The process then proceeds to step S3. Otherwise, it is determined that this piece of traffic data of the power Internet of Things does not contain address-sensitive data, and the process ends.

[0077] S3, marks and classifies the identified sensitive information, and monitors and protects it based on the classification results.

[0078] like Figure 1 As shown, the sensitive data privacy protection calculation method in the cloud environment of the present invention includes: identification of sensitive data, sensitive data monitoring and alarm, encrypted transmission of sensitive data, and access authorization of sensitive data, and deeply integrates privacy computing technology to build a multi-level protection system in the entire data life cycle. Among them, the focus of data privacy protection of the power Internet of Things is to use the network traffic of the power Internet of Things terminal accessed from the network as the entry point, and to perform sensitive data identification, sensitive data monitoring, and alarm notification after sensitive data is discovered from the network traffic collection; as well as the sensitive data of the power Internet of Things management center and various applications are transmitted through strong encryption, and authorization operations are required if sensitive data is accessed, so as to ensure the confidentiality and security of sensitive data during the generation, transmission, use, and storage process, ensure the safe use of sensitive data of the power Internet of Things, and promote the sustainable development of the power Internet of Things.

[0079] 1. Sensitive data identification:

[0080] In the cloud computing environment of the power Internet of Things, there are both external application systems that call business systems through API interfaces, such as electricity bill inquiries, payment record inquiries, etc.; there are also calls for information sharing between internal systems, such as cross-regional electricity consumption information statistics, and real-time analysis of power terminals. This information contains sensitive data of the power Internet of Things, which needs to be protected to ensure the normal operation of the power system and prevent data from being tampered with or destroyed, thereby maintaining the security and stability of the power system. It is necessary to set up a traffic collection point for the Internet access entrance on the cloud, and set up a network traffic collection device by copying the mirror traffic of the switch port. This device performs real-time traffic replication to obtain network traffic information for outbound visits to the cloud environment, such as Figure 2 .

[0081] 1.1 Collection of cloud traffic:

[0082] In the Power Internet of Things (PoT), our network traffic collection system operates in bypass mode by default. The key benefit of this bypass mode is its ability to replicate current network traffic in real time without intercepting or interrupting it. This non-intrusive approach ensures smooth business processes and provides reliable support for traffic data collection and analysis. This allows us to effectively monitor and manage the operational status of the Power Internet of Things (PoT) without disrupting normal business operations.

[0083] A traffic replication and collection strategy is adopted, and then an asynchronous transmission mechanism is used to efficiently transmit traffic information to the data structure analysis device deployed in a cluster at the back end to support subsequent data structure analysis and sensitive data identification.

[0084] 1.2 Structural analysis of collected data:

[0085] When collecting network traffic data, we need to select the appropriate collector based on the data type. Cloud network traffic involves a variety of data formats, including HTML, XML, MIME, JSON, ASCII, and mixed formats. The collector must be compatible with these formats to ensure accurate data parsing. Network device log data is also supported.

[0086] After collection, traffic data undergoes a series of processing steps. The first is data cleansing, which aims to eliminate invalid, outdated, duplicate, and erroneous data to ensure data consistency and accuracy. Next is data conversion. To address the varying formats of standard parameters in traffic data, we standardize these parameters into a unified and recognizable format.

[0087] Finally, to enable better data traceability and multi-dimensional query, data enrichment is performed by adding necessary public parameters such as storage time and collection time to the flow data. These steps together ensure the efficient collection, processing, and application of network flow data in the power Internet of Things.

[0088] 1.3 Sensitive Data Identification:

[0089] Under the privacy computing technology framework, multi-dimensional privacy feature extraction is performed on the collected and analyzed data to identify sensitive information such as user address, name, contact number, etc. The specific process is as follows:

[0090] 1.3.1 Name Recognition: A specific algorithm is used to process the sampled data, which consists of two key steps: preprocessing and disambiguation.

[0091] The preprocessing phase aims to identify all possible Chinese names in a sentence. This process first matches the surname against the Hundred Family Surnames (Baijiaxing) database, then filters common single- and two-character Chinese names. Due to the complexity of the Chinese language, ambiguity may still exist after preprocessing.

[0092] To address this issue, we perform a disambiguation step: matching the preprocessed Chinese names with dictionary words and outputting all possible combinations. Finally, we select the optimal combination based on the size of the gaps between words and the number of words extracted, resulting in accurate name recognition. Incorporating secure query (Private Information Retrieval, PIR) technology from privacy-preserving computing, we implement "query privacy protection" during the name matching process, ensuring that external systems cannot infer user identity information through query operations.

[0093] 1.3.2 Contact Number Identification: The process of identifying mobile phone numbers usually relies on specific rules and algorithms. The first step is data preprocessing, which converts the input text into a format suitable for analysis, such as removing punctuation and spaces. Regular expression matching: Regular expressions are used to find strings that match the mobile phone number format. Regular expressions such as 1[345789]\d{9} are used to match and discover mobile phone numbers.

[0094] The secure multi-party computing (MPC) protocol in privacy computing is introduced to perform distributed collaborative verification of mobile phone number matching rules to prevent the original number from being exposed in plain text in computing nodes.

[0095] 1.3.3 Address Identification: Identifying sensitive data in the traffic of the power Internet of Things is a complex task. The target address to be identified is not a standard address style, the format is diverse, and the data fusion is messy. How to quickly and accurately identify it requires a unique adaptive algorithm.

[0096] The present invention adopts an address adaptive recognition algorithm which is divided into four parts: target data segmentation, initialization of standard address library, global Hash matching of address tree, hierarchical dependency and weighted average of address tree node depth and breadth, such as Figure 3 .

[0097] Based on address tree adaptive recognition technology, the following four parts are used to achieve accurate identification of address-sensitive data, and privacy computing technology is integrated to achieve cross-domain collaborative modeling:

[0098] 1.3.3.1 Target Data Segmentation

[0099] Before executing the address tree adaptive algorithm, it is necessary to segment the text information obtained from the traffic of the power Internet of Things. During the segmentation process, data is filtered through the unique attributes of the address, that is, address segmentation segments are obtained from a string of text: a province, a city, a district, a street, a detailed address. If some of the address segmentation segments are empty, they are filled with null values.

[0100] S={S1,S2,S3,S4,S5}

[0101] Among them, S is the identified target address information, S1 represents a province data, S2 represents a city data, S3 represents a district (county) data, S4 represents a street (town) data, and S5 represents a detailed address.

[0102] 1.3.3.2 Initialize the standard address library

[0103] Next, we import a standard address database, which includes detailed address information for provinces, cities, districts, and streets across the country. In order to facilitate fast global queries, we also generate address hash data.

[0104]

[0105] Where m is the total number of all complete addresses in the address database (for example, all possible combinations of "province → city → district → street → detailed address" in the country), The name of the i-th administrative district in the j-th address (for example, i=1 is the province, i=5 is the detailed address); D is the address information of the standard address database, D1 represents the province data of the standard address, D2 represents the city data of the standard address, D3 represents the district data of the standard address, D4 represents the street data of the standard address, and D5 represents the detailed address data of the standard address. s H is the Hash information in the target address information that needs to be identified in the traffic. d It is the hash information in the address information of the standard address library. HD1 represents the hash data of a province, HD2 represents the hash data of a city, HD3 represents the hash data of a district, HD4 represents the hash data of a street, and HD5 represents the hash data of a detailed address.

[0106] By dynamically updating the standard address library features through federated learning technology, all participants (such as power nodes in different regions) share address feature gradients under the privacy computing framework to achieve collaborative optimization of the global address model while ensuring that the original address data does not leave the local domain.

[0107] 1.3.3.3 Address Tree Global Hash Matching

[0108] Hash the target address to obtain three types of data, S h1 :Hash of detailed address of province, city, district and street, S h2 :Hash of the district street address, S h3 : Hash of street address.

[0109] S h1 =Hash*(S1+S2+S3+S4+S5)

[0110] S h2 =Hash*(S3+S4+S5)

[0111] S h3 =Hash*(S4+S5)

[0112] S h1 and D h1 Compare them. If they are equal, it is found that this traffic data contains address-sensitive data and the address information is a province (S1), a city (S2), a district (S3), a street (S4), and a detailed address (S5).

[0113] S h2 and D h2Compare them. If they are equal, it is found that this traffic data contains address-sensitive data and the address information is a certain district (S3), a certain street (S4), and a certain detailed address (S5).

[0114] S h3 and D h3 Compare them. If they are equal, it is found that this traffic data contains address-sensitive data and the address information is a certain province (a certain street (S4) and a certain detailed address (S5);

[0115] If the sensitive data of the power Internet of Things is not recognized in the address database’s Hash data through the global Hash of the address tree, the next step is to perform depth-breadth-weighted matching of the address tree nodes. Using the Blind Hash technology in privacy computing, the S h With D h When doing so, ensure that nodes cannot reversely deduce each other’s original address data. h1 This is the hash data obtained by splicing the detailed address information of a province, city, district and street in the power Internet of Things through traffic; D h2 The hash data after splicing the detailed address information of a certain district street obtained through traffic in the power Internet of Things; D h3 This is hash data obtained by splicing the detailed address information of a street obtained through traffic in the power Internet of Things.

[0116] 1.3.3.4 Address Tree Node Depth, Breadth, and Weight Matching

[0117] First, we construct a full address tree. We divide the known full address database into two steps to construct the tree. First, we perform a hash operation on the provincial, municipal, district, street, and detailed address information in the full address database to obtain a unique key value. Then, we draw the address of each level from the root to the leaf node of the tree. Finally, we construct the full address database information into a multi-level, multi-node address tree. Each node of the tree is H d .

[0118] The sensitive address information to be identified in the traffic of the power Internet of Things is segmented to obtain the provincial, municipal, district, street and detailed address information. Then, the data at each level is hashed to obtain a unique Key value, namely H S .

[0119] Hash the target address segments to be identified, first using the hash value (H s1 ) to conduct a breadth search and find H s1 With H d1 If the same, it will inspire a deep search, using the Hash value (H s2) conducted a deep search and found H s2 With H d2 Similarly, the search is carried out in a sequential manner until a detailed address is searched in depth. This scenario is an ideal state, where the target address completely matches the address in the address library, and the matching coefficient is 100%.

[0120] Traverse the address tree nodes in breadth order to obtain all provincial node weight data, then deeply traverse the city-level nodes, district-level nodes, street-level nodes, and detailed address nodes, and obtain all city-level node weight data, district-level node weight data, street-level node weight data, and detailed address node weight data in turn. The "hierarchical dependency and weighted average" algorithm is used to identify the sensitivity of the address to be identified in the flow of the power Internet of Things, that is, the "hierarchical sensitivity coefficient ”.

[0121] The address sensitive information to be identified found in the traffic of the power Internet of Things (H S ) and the National Address Tree (H d ) performs weighted comparison of Hash, obtains the hierarchical sensitivity coefficient through hierarchical dependency and weighted average; among them, the hierarchical sensitivity coefficient of the depth and breadth weight of the address tree node The calculation formula is:

[0122]

[0123] The formula means as follows:

[0124] (Hierarchical sensitivity coefficient): Indicates the comprehensive sensitivity score of address data, ranging from [0,1]. Adaptive Sensitivity Threshold: determines that the address to be identified found in the traffic of the power Internet of Things is sensitive information; if Adaptive Sensitivity Threshold determines whether the addresses to be identified found in the traffic of the power Internet of Things are non-sensitive information.

[0125] (Dynamic weight): The weight distribution of the i-th level administrative district is determined by the preset weight The security risk factor is dynamically adjusted. When the security risk is high, the weight of higher levels (such as provincial levels) is increased, while the weight of lower levels (such as street levels) is reduced.

[0126] (Hierarchical similarity): The hash matching similarity of the i-th level administrative district address (such as provincial matching s1, municipal matching s2, etc.), calculated by hash value comparison (for example: a complete match is 1, a partial match is 0.8).

[0127] (Progressive hierarchical dependency): k represents the upper level number (k = 1 is provincial level, k = 2 is municipal level, and so on), s k Indicates the matching degree of the k-th level address (s k ∈[0,1]), It means taking the square root of the superior matching degree to reduce over-inhibition.

[0128] λ (layer attenuation compensation factor): negatively correlated with data traffic complexity (λ = 1-complexity / 2). The more chaotic the traffic (higher complexity), the slower the attenuation, the smaller λ, and the greater the contribution of deep addresses.

[0129] σ() (dynamic threshold function): The score is compressed to [0, 1] through the Sigmoid function. The parameters kk (steepness) and θθ (threshold offset) of the dynamic threshold function are dynamically controlled by user policy and security risk.

[0130] For example, when the power monitoring system detects a suspected address fragment, the initial data is: security risk factor = 0.8 (abnormal login behavior detected); data traffic complexity = 0.7 (address fragmentation and containing unstructured text); user policy weight = 0.7 (high sensitivity mode); administrative level: 3 levels (provincial, municipal, street level); preset weight: W = [0.6, 0.3, 0.1], then the corresponding level sensitivity coefficient is:

[0131]

[0132] In the sensitive address identification scenario of the power Internet of Things (IoT), the "hierarchical sensitivity coefficient" algorithm improves identification accuracy through a "dynamic adjustment and progressive" approach. When the system detects a network attack (such as a DDoS attack), the "hierarchical sensitivity coefficient" algorithm automatically increases the matching weight of provincial-level addresses (similar to "focusing on high-risk entry points") while reducing the stringency of deep street-level addresses (allowing for partial fuzzy matches). If traffic data is complex (such as severe address fragmentation), a "hierarchical attenuation compensation" mechanism is used to enhance the matching contribution of deep addresses (to avoid missing hidden sensitive segments). The system also dynamically adjusts the strictness of the identification threshold based on the user-defined security level (low / medium / high). For example, if the user selects "high sensitivity" mode, even if the address segment has only 60% similarity (such as "** Street, Haidian District, Beijing" is partially fuzzy), it will trigger a sensitive flag. This achieves adaptive protection in complex network environments: "The higher the risk, the stricter the matching; the more complex the data, the more detailed the matching."

[0133] The present invention adopts the AST adaptive sensitivity threshold as the boundary value of the sensitivity of the address to be identified as sensitive information found in the traffic of the power Internet of Things, and realizes the adaptation of sensitive data judgment through the joint modeling of security risks, traffic complexity and user policies.

[0134] The adaptive sensitivity threshold formula is as follows:

[0135]

[0136] in, It is the security risk factor, which is related to the attack frequency and vulnerability exposure level; The complexity of data flow is related to the proportion of unstructured text and the depth of address segmentation level; is the user policy weight, which is related to the user-defined sensitivity level; β, γ, and δ are weighting factors, where β + γ + δ = 1. The security risk coefficient (ranging from 0 to 1) represents the environmental threat level and is related to attack frequency and vulnerability exposure. Greater risk indicates a higher value. Data traffic complexity (ranging from 0 to 1) is related to indicators such as the proportion of unstructured text and the depth of address segmentation levels. Higher complexity indicates a higher value. The user policy weight (ranging from 0 to 1) is obtained by mapping the user-defined sensitivity level to a numerical value. For example, the user-defined sensitivity levels of low, medium, and high are mapped to corresponding user policy weights of 0.3, 0.5, and 0.7. Stricter sensitivity modes indicate higher values.

[0137] like Adaptive Sensitivity Threshold: excludes the addresses to be identified found in the traffic of the power Internet of Things as sensitive information; if Adaptive Sensitivity Threshold: Determines whether the addresses to be identified found in the traffic of the power Internet of Things are sensitive information.

[0138] For example, when a DDoS attack is detected and the traffic complexity is low, the user policy is "medium sensitive", then the priority is to protect sensitive data from being leaked. Therefore, the adaptive sensitivity threshold is calculated according to the scenario: AST = 0.6 × 0.1 + 0.2 × 0.8 + 0.2 × 0.5 = 0.32, at which point the hierarchical sensitivity coefficient is As long as the value is greater than or equal to 0.32, the address to be identified in the traffic of the Power Internet of Things is considered sensitive information. If it is less than 0.32, the address to be identified in the traffic of the Power Internet of Things is excluded as sensitive information. In this example, the security risk coefficient = 0.9; the data traffic complexity = 0.2; the user policy weight = 0.5 (medium sensitivity mode); the weight distribution: β = 0.6, γ = 0.2, δ = 0.2 (satisfying β + γ + δ = 1).

[0139] For example, when there is no security threat but the traffic complexity is high, the user policy is "medium sensitive", then the priority is to avoid false alarms. Therefore, the adaptive sensitivity threshold is calculated according to the scenario: AST = 0.6 × 0.9 + 0.2 × 0.2 + 0.2 × 0.5 = 0.68, at this time the hierarchical sensitivity coefficient If the value is greater than or equal to 0.68, the address to be identified in the traffic of the Power Internet of Things is considered sensitive information. If it is less than 0.68, the address to be identified in the traffic of the Power Internet of Things is excluded as sensitive information. In this example, the security risk coefficient = 0.1; the data traffic complexity = 0.8; the user policy weight = 0.5 (medium sensitivity mode); the weight distribution: β = 0.6, γ = 0.2, δ = 0.2 (satisfying β + γ + δ = 1).

[0140] During the weight calculation stage, a privacy computing sandbox can also be built based on the trusted execution environment (TEE) to securely isolate the generation process of the hierarchical sensitivity coefficient R and prevent the leakage of sensitive weight parameters during the calculation process.

[0141] 3Sensitive data monitoring alarm:

[0142] Through timely and accurate sensitive data identification, we can precisely capture sensitive data within network traffic in the cloud environment of the Power Internet of Things (PoI), particularly data involving personal privacy, such as address information, mobile phone numbers, and user names when inquiring about electricity bills. The leakage of this type of information could not only damage a company's reputation and profits but also severely impact its daily operations. Therefore, we must attach great importance to and strictly protect this sensitive information.

[0143] Once such sensitive information is detected within the Power IoT cloud environment, a tagging and classification mechanism is activated to meticulously label and classify the sensitive data. Emergency response procedures are initiated, and alerts are sent to relevant personnel via the Power IoT's dedicated email and SMS notification mechanisms. When an alert is triggered, the leaked data is minimized and desensitized using the privacy-preserving middleware layer. Differential privacy techniques are used to add noise perturbations to the statistical features of the alert information, minimizing the risk of secondary leaks.

[0144] When sensitive data is confirmed to have been leaked through the network, respond and accurately assess the scope, content and severity of the data leak to ensure a comprehensive understanding of the situation.

[0145] Data breaches need to be reported immediately to the security team, legal counsel, and management so they can take prompt action to mitigate potential risks. To contain the impact of the breach, emergency measures will be taken, such as adjusting firewall rules and suspending affected systems, to minimize the scope of sensitive data leakage.

[0146] Furthermore, detailed information about the breach, such as the time of the breach, the affected business systems, and the possible causes, should be collected to accurately assess the losses inflicted on businesses and individuals, including the impact on the company's brand and reputation. Finally, a thorough investigation of this sensitive data leak is necessary to identify and promptly address security vulnerabilities, establishing a closed-loop process to ensure the comprehensive protection of sensitive data within the Power IoT cloud environment.

[0147] 4. Encrypted transmission of sensitive data:

[0148] After discovering sensitive data transmission in the power Internet of Things cloud environment, tracing the source to find the application source requires transforming and optimizing the application's transmission mechanism.

[0149] Sensitive data is encrypted and transmitted primarily using symmetric and non-compositional encryption. Comprehensive security protection begins at the source of the data. This includes end-to-end encrypted transmission between power IoT applications, IoT centers, sensors, collectors, and edge computing centers, ensuring data remains encrypted throughout the entire data transmission chain, thus guaranteeing the secure transmission of sensitive data.

[0150] To address the unique characteristics of the Power IoT cloud environment, a combination of AES and RSA encryption is used. This optimized encryption method ensures encryption security while maintaining the resources and terminal performance of the Power IoT cloud environment.

[0151] AES encryption, a symmetric advanced encryption standard, plays a vital role in the power IoT. It utilizes block ciphers, including ECB, CBC, OFB, CTR, and CFB modes, and supports key lengths of 128, 192, and 256 bits. This encryption method offers high computation speed and minimal memory usage, making it ideally suited for encryption requirements during data transmission in the power IoT.

[0152] RSA encryption, as an asymmetric encryption method, provides an additional layer of security for the Power Internet of Things. It consists of a public-private key pair with a key length of 1024 bits or longer, offering high security. While this method consumes resources for large amounts of data transmission, we primarily utilize RSA encryption to transmit symmetric keys, achieving a perfect combination of security and efficiency, ensuring secure key transmission.

[0153] In the power IoT, we utilize end-to-end encrypted transmission, from cloud-based application calls and terminal data collection to edge computing. Node-to-node encryption is implemented from cloud-based applications to the IoT management center. During data transmission, we use AES-256 encryption by default, with dynamic keys for encryption and decryption. When the receiving end receives the encrypted data, it decrypts the AES-Key using the RSA private key to obtain the symmetric encryption key, which is then used to decrypt the data. This ensures the confidentiality of terminal data during collection, transmission, use, and storage, effectively preventing the leakage of sensitive data.

[0154] Homomorphic encryption technology is introduced in end-to-end encrypted transmission, enabling direct cloud-based aggregate analysis of encrypted data (such as electricity consumption statistics), achieving the privacy computing feature of "data available but invisible." For cross-institutional data collaboration scenarios, a secure multi-party computing (MPC) protocol is used to implement multi-party key collaborative management, ensuring that no single node can obtain complete key information during encryption and decryption operations.

[0155] 5. Sensitive data access authorization:

[0156] In the power Internet of Things cloud environment, it is necessary to promptly identify sensitive data in network traffic and ensure the security of the transmission process by encrypting the sensitive data. At the same time, to ensure the security and integrity of sensitive data, access rights need to be designed.

[0157] First, a strong password policy needs to be implemented, including long password lengths, multiple character types, etc., to increase the difficulty of cracking. Secondly, a role-based access control (RBAC) model is adopted to assign different access rights according to the user's role and responsibilities to ensure that only appropriate users can access sensitive data. In addition, dynamic permission management is introduced to adjust user access rights in real time according to business needs and data sensitivity. Build a trusted access control layer based on privacy computing: Implement dynamic permission decisions in the TEE environment, combined with attribute-based encryption (ABE) technology to achieve ciphertext execution of "fine-grained data access policies." Introduce privacy computing authorization credentials (such as zk-SNARKs) to support users to complete permission verification without exposing their identity information, meeting the principle of minimizing privacy exposure.

[0158] At the same time, we will implement data auditing and monitoring mechanisms to record all user access to sensitive data and set alerts for abnormal behavior to promptly identify and address potential security risks. Finally, we will strengthen employee security training to raise awareness of sensitive data protection and ensure the security of sensitive data in the Power IoT cloud environment.

[0159] In summary, the present invention constructs a data privacy protection system for the power Internet of Things based on a cloud computing environment, through real-time collection and analysis of north-south traffic, precise identification of privacy data using an address tree adaptive algorithm combined with federated learning to optimize the cross-domain feature library, encrypted transmission introducing homomorphic encryption to achieve secret state computing, dynamic authorization based on TEE and attribute-based encryption and monitoring alarm fusion differential privacy, to achieve "available invisible" protection of data throughout its life cycle, timely discover sensitive data in the data flow of the cloud environment, and provide real-time feedback for data privacy protection; ensure that sensitive data is not illegally intercepted or tampered with during transmission; ensure that only authorized personnel can access relevant sensitive data in the cloud environment, which greatly improves the security of sensitive data; improves the security of sensitive data, reduces the risk of sensitive data leakage, and provides a solid guarantee for the privacy information of the power Internet of Things.

[0160] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0161] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A sensitive data privacy protection computing method in a cloud environment, characterized by: The method comprises the following steps: S1, collects flow data of the power Internet of Things and performs structured processing on the collected flow data; S2, under the privacy computing technology framework, extracts multi-dimensional privacy features from the structured traffic data to identify sensitive information such as user address, name, and contact number. An adaptive address recognition algorithm is used to identify user addresses in the traffic data. The recognition process includes the following sub-steps: S21, target data segmentation; specifically, segmenting the text information obtained in the traffic data to obtain address segmentation segments at multiple levels of administrative districts; S22, initialize the standard address database; specifically, import the standard address database, which includes detailed address information of provinces, cities, districts and streets across the country and corresponding address hash data; S23, global hash matching of the address tree; specifically, obtaining the hash data corresponding to the target address to be identified; comparing the hash data of the target address with the hash data of the address in the standard address library one by one. If equal data is found, it is determined that address-sensitive data exists in this traffic data and the address information is output, and the process proceeds to step S3; otherwise, the process proceeds to step S24; S24, address tree node depth-breadth weight matching; hash the target address segment to be identified, and perform breadth and depth searches on the hash data of multiple address segments of the target address to be identified in descending order of administrative district level based on the standardized address database. If each address segment can match equal data, it is determined that address-sensitive data exists in this traffic data and the address information is output, and the process proceeds to step S3; otherwise, the process proceeds to step S25; S25: For different administrative district levels, corresponding matching weights are introduced. Combined with the hash data of multiple address word segments of the target address to be identified, the hierarchical sensitivity coefficient of the depth and breadth of the address tree node is calculated. If the hierarchical sensitivity coefficient of the depth and breadth of the address tree node exceeds a preset threshold, it is determined that this piece of traffic data of the power Internet of Things contains address-sensitive data and the address information is output. The process then proceeds to step S3. Otherwise, it is determined that this piece of traffic data of the power Internet of Things does not contain address-sensitive data, and the process ends. S3, marks and classifies the identified sensitive information, and monitors and protects it based on the classification results.

2. The sensitive data privacy protection computing method in a cloud environment according to claim 1 is characterized in that: In step S1, the bypass mode is adopted to copy the current network traffic of the cloud Internet in real time from the power Internet of Things, the north-south network traffic of the unified entrance and exit of the external network, and the data communication traffic between the inside and outside of the network, that is, the network traffic that vertically crosses the network boundary.

3. The sensitive data privacy protection computing method in a cloud environment according to claim 1 is characterized in that: In step S1, the process of structuring the collected traffic data includes the following steps: Clean the collected traffic data to remove invalid, expired, duplicate and erroneous data; Convert the cleaned traffic data into a unified format; Add common parameters including storage time and collection time to the traffic data after format conversion.

4. The sensitive data privacy protection computing method in a cloud environment according to claim 1 is characterized in that: In step S2, the name recognition process includes the following steps: Match the surnames against hundreds of surnames, then filter them based on single-character and double-character Chinese names to preliminarily identify all possible Chinese names in the sentence. Match the initially recognized Chinese names with the words in the dictionary and output all possible combinations; The process of identifying a contact number includes the following steps: Convert the input text to a unified format, removing punctuation and spaces; Use regular expressions to find strings that match the mobile phone number format; The secure multi-party computing protocol in privacy computing is introduced to perform distributed collaborative verification of mobile phone number matching rules.

5. The sensitive data privacy protection computing method in a cloud environment according to claim 1 is characterized in that: In step S2, the text information obtained from the flow data of the power Internet of Things is segmented to obtain address segmentation segments at multiple levels of administrative districts. If some of the address segmentation segments are empty, they are filled with null values ​​to obtain: S = {S1, S2, S3, S4, S5}; Where S is the target address information to be identified, S1 represents the province data to be identified, S2 represents the city data to be identified, S3 represents the district data to be identified, S4 represents the street data to be identified, and S5 represents the detailed address data to be identified; Import the standard address library D: Where m represents the total number of all complete addresses in the address database; Indicates the name of the i-th level administrative district in the j-th address, D1 represents the province data of the standard address, D2 represents the city data of the standard address, D3 represents the district data of the standard address, D4 represents the street data of the standard address, and D5 represents the detailed address data of the standard address; Hash the address data in the standard address library D to obtain the corresponding address hash data: Where H d It is the hash information in the address information of the standard address library. HD1 represents the hash data of a province, HD2 represents the hash data of a city, HD3 represents the hash data of a district, HD4 represents the hash data of a street, and HD5 represents the hash data of a detailed address.

6. The sensitive data privacy protection computing method in a cloud environment according to claim 5, characterized in that: In step S23, the process of global hash matching of the address tree includes the following steps: S231, hash the target address to be identified in the power Internet of Things traffic to obtain the following three types of data: hash data of the province, city, district and street detailed address S h1 =Hash*(S1+S2+S3+S4+S5), Hash data S of the district street address h2 =Hash*(S3+S4+S5), Hash data S of street address h3 =Hash*(S4+S5); S232, S h1 and D h1 Compare and if they are equal, it is found that this traffic data contains address-sensitive data and the address information is a province S1, a city S2, a district S3, a street S4, and a detailed address S5; otherwise, go to step S233; S233, S h2 and D h2 Compare and if they are equal, it is found that this traffic data contains address-sensitive data and the address information is a certain district S3, a certain street S4, and a certain detailed address S5. Otherwise, go to step S234; S234, S h3 and D h3 Compare them. If they are equal, it is found that this traffic data contains address-sensitive data and the address information is a certain street S4 and a certain detailed address S5; Where D h1 This is the hash data obtained by splicing the detailed address information of a province, city, district and street in the power Internet of Things through traffic; D h2 The hash data after splicing the detailed address information of a certain district street obtained through traffic in the power Internet of Things; D h3 This is hash data obtained by splicing the detailed address information of a street obtained through traffic in the power Internet of Things.

7. The sensitive data privacy protection computing method in a cloud environment according to claim 6, characterized in that: Step S24 further includes: S241, hash the target address segment to be identified, using the hash value H of S1 s1 Perform breadth search until the search is found with H s1 Same standard address save data H d1 , go to step S242, if the same data cannot be found, go to step S25; S242, using the hash value H of S2 s2 Perform a deep search until you find the one that matches H s2 Same standard address city data H d2 , go to step S243, if the same data cannot be found, go to step S25; S243, using the hash value H of area S3 s3 Perform a deep search until you find the same s3 Same standard address area data H d3 , go to step S244, if the same data cannot be found, go to step S25; S244, using the hash value H of street S4 s4 Perform a deep search until you find the one that matches H s4 Same standard address street data H d4 , go to step S245, if the same data cannot be found, go to step S25; S245, using the hash value H of the detailed address S5 s5 Perform a deep search until you find the one that matches H s5 Same standard address detailed address data H d5 , go to step S246, if the same data cannot be found, go to step S25; S246, Comprehensive H d1 、H d2 、H d3 、H d4 and H d5 , and obtain the identification result of the target address to be identified.

8. The sensitive data privacy protection computing method in a cloud environment according to claim 1 is characterized in that: Step S25 further includes: The sensitive address information to be identified in the traffic of the power Internet of Things is segmented to obtain multiple administrative district level address data including provincial, municipal, district, street and detailed address information. The address data at each level is hashed to obtain a unique key value H. S ; Through the known standard address database, we build a full address information of multiple levels, including full provincial address information, full city address information, full district address information, full street address information and full detailed address information. We perform Hash calculation on the full address information to obtain the corresponding full Key value, and form a national address Hash address tree. Each node of the tree is H d ; Perform a weighted comparison of the hash values ​​of the address-sensitive information to be identified in the traffic of the power Internet of Things and the address tree of the national address Hash, and obtain the hierarchical sensitivity coefficient through hierarchical dependency and weighted average. Where, is the hierarchical sensitivity coefficient, which is used to represent the comprehensive sensitivity score of address data; is the dynamic weight of the i-th level administrative district, which is determined by the preset weight Dynamic adjustment of the security risk coefficient corresponding to the corresponding administrative area; Indicates the hash matching similarity of the i-th level administrative district address; Indicates progressive hierarchical dependency, where k is the upper level number and s k is the matching degree of the k-th level address; λ represents the level attenuation compensation factor, which is negatively correlated with the data traffic complexity; σ() represents the dynamic threshold function.

9. The sensitive data privacy protection computing method in a cloud environment according to claim 1, characterized in that: The preset threshold adopts an adaptive sensitive threshold, which is the boundary value of the sensitivity of the address to be identified as sensitive information found in the traffic of the power Internet of Things. Through the joint modeling of security risks, traffic complexity, and user policies, the sensitive data judgment level is adaptively adjusted.

10. The sensitive data privacy protection computing method in a cloud environment according to claim 1, characterized in that: The calculation formula of the adaptive sensitivity threshold AST is: in, It is the security risk factor, which is related to the attack frequency and vulnerability exposure level; The complexity of data flow is related to the proportion of unstructured text and the depth of address segmentation level; is the user policy weight, which is related to the user-defined sensitivity level; β, γ, and δ are weight factors, and β+γ+δ=1; like is less than the adaptive sensitivity threshold, excluding the addresses to be identified found in the traffic of the power Internet of Things as sensitive information; if If the value is greater than or equal to the adaptive sensitivity threshold, the address to be identified found in the traffic of the power Internet of Things is determined to be sensitive information.

Images