A method and system for dynamically evaluating information sensitivity of a smart grid

Abstract

The application discloses a kind of dynamic evaluation method and system of smart grid information sensitivity, mainly including the following steps: first, the influence of data objectivity information factor I, power business operation environment C and business participant experience P on smart grid information sensitivity is considered, and a multi-dimensional measurement system of smart grid business information sensitivity is constructed;Second, the multi-factor correlation mapping relationship between I, C and P is analyzed, and a dynamic measurement system of information sensitivity under the influence of multi-factor power business scenario is proposed;Finally, a data sensitivity dynamic evaluation method of power business scenario self-learning is proposed, a dynamic feedback mechanism based on deep reinforcement learning is used, and the dynamic optimization of information sensitivity evaluation model based on scene self-adaptation is realized.The beneficial effects of the application are that the evaluation of information sensitivity is more universal and targeted, the dynamic evaluation of information sensitivity based on scene self-adaptation is realized, and the accuracy of smart grid information sensitivity calculation is improved.

Description

Technical Field

[0001] This invention relates to the field of information security technology, and more specifically, to a dynamic assessment method and system for the information sensitivity of smart grids. Background Technology

[0002] With the increasing intelligence and interactivity of smart grid terminal equipment and the growing diversity and complexity of network communication methods, the perception, transmission, sharing, and utilization of sensitive information in smart grid operations, such as remote sensing, remote control, and power consumption data collection, face significant security and reliability challenges. First, in the smart grid environment, massive amounts of IoT devices, relay protection systems, and PMU devices are distributed throughout the grid. These devices perceive, monitor, and publish real-time power business information such as grid equipment status, grid operation status, and market transaction status. During data transmission, sensitive information such as relevant business status information or control commands may be subject to attacks such as eavesdropping or traffic tracing analysis. This could lead to the leakage of personal privacy of power users or, more seriously, threaten the safe operation of the power grid. Second, in the vertical industry application ecosystem of the power sector, various parties, including power generation, transmission, substation, distribution, energy storage, power consumption, dispatching, and operation management, often share their respective energy flows, business flows, and information flows according to their own needs. This results in sensitive data facing privacy and security threats during cross-domain sharing and circulation. To provide effective full lifecycle security protection for sensitive data in sharing and circulation, we need to know the actual sensitivity of power business-related information in advance.

[0003] In fact, accurate and reliable sensitivity assessment of power business information security is a key prerequisite for ensuring the secure circulation and reliable sharing of sensitive power business data across the entire domain. Currently, several studies have proposed methods for measuring the privacy of sensitive data from different perspectives: one uses information entropy to measure privacy sensitivity based on the degree of uncertainty contained in the privacy information; another uses probability statistics to infer the possibility of privacy information leakage using probability distribution information, thereby measuring the sensitivity of the information; and a third uses set pair analysis theory, combining qualitative and quantitative methods to solve the problem of information sensitivity assessment. However, each of these methods has certain shortcomings, and they consider only a single factor. For example, the information entropy-based measurement method essentially measures the degree of uncertainty of information, thus it has a good effect on measuring the sensitivity of a single type of privacy data. However, in the complex scenario of a smart grid, facing multi-source data types, and data from different sources having sensitive and non-sensitive aspects, it cannot accurately assess sensitive data due to its sole consideration of information chaos and change factors. Probability statistics-based methods require a certain degree of prior information such as the probability of event occurrence. However, in most cases, this prior information cannot be obtained in advance, and the accuracy of data inferred subjectively cannot be predicted, thus limiting its application in various scenarios. Methods based on set pair analysis primarily measure the uncertainty of privacy information associated with two datasets, but they require a given dataset as a prerequisite and cannot effectively measure issues such as the real-time nature of data in smart grids.

[0004] In reality, the sensitivity of power business-related information is spatiotemporal. Not only does the timeliness of the data itself directly affect the risk of privacy breaches, but the operational environment also influences the level of security threats the data faces during transmission. Furthermore, the sensitivity of power information is also related to business scenario requirements and the subjective experiences and feelings of information producers / users. Therefore, the sensitivity of information is inevitably influenced by multiple factors. How can we effectively measure the sensitivity of smart grid business information? Current research in this area is neither systematic nor in-depth. On the one hand, there is a lack of suitable descriptive models for the privacy and security threats faced by sensitive data under the combined influence of multiple factors. On the other hand, existing sensitivity measurement methods only focus on the objective determining role of the core carrier of sensitive attributes (i.e., data / information) on that attribute, while neglecting the subjective perception of data sensitivity by the privacy and security threat experiences of various participating entities (e.g., data producers, data owners, data users, data intermediaries, etc.) in data sharing and circulation.

[0005] In view of the above, this application is hereby submitted. Summary of the Invention

[0006] The technical problems to be solved by this invention are as follows: First, how to comprehensively consider the objective privacy attributes of data and the subjective cognitive role of entities related to business scenarios to construct a smart grid information sensitivity assessment index system that is both universal and targeted; second, how to comprehensively consider the impact of the dynamic correlation between multiple factors in heterogeneous and diverse business scenarios on sensitivity assessment, and realize the quantification of the complex mapping relationship between various factors; third, how to intelligently realize the real-time dynamic analysis and assessment of power business information sensitivity. The purpose is to provide a dynamic assessment method and system for smart grid information sensitivity, realize adaptive dynamic assessment of information sensitivity in power business scenarios, and improve the accuracy, universality, and targeting of smart grid information sensitivity assessment.

[0007] This invention is achieved through the following technical solution:

[0008] A dynamic assessment method for the information sensitivity of a smart grid, comprising the following steps:

[0009] S1: Obtain raw power grid data information, which includes historical structured electricity consumption data of the power grid to be evaluated, business data circulation environment information, and privacy data information of historical business participants;

[0010] S2: Based on the original power grid data information, perform calculation and analysis to obtain the quantitative values ​​of objective information factor I, power business operation environment C, and business participant experience P information sensitivity evaluation index;

[0011] S3: Analyze the correlation mapping relationship between objective information factor I, the power business operation environment C, and the experience of business participants P, and use a self-learning method to obtain the correlation mapping function f under the influence of multiple factors. I f C f P And the sensitivity metrics I′, C′, and P′ under the influence of multiple factors;

[0012] S4: Based on the quantified values ​​I, C, and P of the sensitivity assessment index and the quantified values ​​I′, C′, and P′ of the sensitivity under the influence of multiple factors, construct a dynamic assessment model of the data sensitivity of the power grid to be assessed, and use deep reinforcement learning to intelligently optimize the dynamic assessment model to obtain the final assessment model of the data sensitivity of the power grid to be assessed.

[0013] S5: Input the data information generated by the power grid to be evaluated into the data sensitivity assessment model to obtain the data sensitivity of the power grid to be evaluated.

[0014] Traditional methods for analyzing the sensitivity of smart grids often employ information entropy, probability statistics, or set pair analysis. However, these methods can only analyze the sensitivity of smart grids under the influence of a single factor, failing to address the sensitivity under multiple influencing factors. This invention provides a dynamic assessment method for smart grid information sensitivity. It combines multiple factors to establish a relevant model and, based on this model, achieves accurate analysis of smart grid sensitivity under the combined influence of multiple factors. This enables adaptive dynamic assessment of information sensitivity in power business scenarios, improving the accuracy, versatility, and relevance of smart grid information sensitivity assessment.

[0015] Preferably, the historical structured electricity consumption data information includes user location information, user electricity consumption data information, user electricity consumption time, and user service request information. The privacy data information of the historical business participants' experience includes user rating information on the degree to which they are unwilling to disclose various types of privacy data information, user settings changes for various types of privacy data information, and information on the degree of privacy protection of the current business operation environment, data privacy attributes, and the electricity service attributes used.

[0016] Preferably, in step S2, the sub-step of analyzing and processing the original power grid data to obtain objective information factor I includes:

[0017] Historical structured electricity consumption data is digitized to obtain objective information attribute I. a The information attribute I a Numerical mapping of objective information factor attributes;

[0018] The time intervals and time spans of collecting historical structured electricity consumption data are calculated and analyzed to obtain an objective information scale I. s The information scale I s The size of the data representing objective information factors;

[0019] Based on the characteristic that electricity consumption data changes over time, the timeliness of obtaining objective information is crucial. t The information timeliness I t The impact of time variations on objective information factors in data attributes;

[0020] Attribute I of objective information of parameters a Scale of objective information I s and the timeliness of objective information I t The sensitive information of the power grid to be evaluated is comprehensively measured to obtain objective information factor I.

[0021] Preferably, in step S2, the sub-step of analyzing and processing the original power grid data information to obtain the power business operation environment C includes:

[0022] A correlation dependency graph model is constructed using the mutual information correlation analysis method.

[0023] Based on the state of each node in the data flow path in the aforementioned dependency graph model, the service reliability C in the power business scenario is obtained. t The service credibility C t The level of trust in the ability to provide data privacy protection for the power business operating environment under the current services;

[0024] Mutual information is used to identify the relationships between nodes in the dependency graph model, and the dependency relevance C in the power business scenario is obtained. r The dependency relevance C r This refers to the interrelationships between nodes that have dependencies in the power business operation environment.

[0025] Through the dependency correlation C r Quantify the correlation sensitivity C of the power business operating environment s The correlation sensitivity C s The level of privacy and security of sensitive information caused by the interrelationships and dependencies between nodes in the power business operation environment;

[0026] Based on parameter service reliability and correlation sensitivity Calculate and obtain the power business operating environment .

[0027] Preferably, in step S2, the sub-step of analyzing and processing the original power grid data information to obtain the business participant experience P includes:

[0028] Users are rated on their level of unwillingness to disclose various types of private data. A user privacy preference weight value is calculated based on this rating, and then this weight value is used to calculate the user privacy preference P. r The user privacy preference P r This refers to the degree of user preference for different objective information factors;

[0029] Based on changes in users' settings for various types of privacy data, privacy awareness P is obtained. a The privacy awareness P a To raise users' awareness of protecting their sensitive information from security threats during data sharing;

[0030] Based on the user's perception of the privacy protection level of the current business operation environment, data privacy attributes, and the electricity service attributes used, the privacy expectation P is obtained. e The privacy expectation P e This represents users' expectations regarding the level of data privacy protection they receive.

[0031] User privacy preferences for parameters P r Privacy awareness P a and privacy expectations P e Perform quantitative calculations to obtain the experience P of business participants.

[0032] Preferably, the sub-steps of step S4 include:

[0033] Establish a correlation mapping function f between two factors: the business operation environment (C) and the privacy and security experience (P) of business participants regarding sensitive information, and the objectivity and privacy (I) of sensitive information. I (I r I a I t |C,P);

[0034] Establish objectivity, privacy (I), and correlation function (f) I (I r I a I t A reinforcement learning model for the mapping relationship between |C,P);

[0035] By inputting the power business operation environment C and the business participant experience P into the reinforcement learning model, the sensitivity quantification index I′ of the objective information factor I under the influence of the power business operation environment C and the business participant experience P is obtained.

[0036] Calculate the learning incentive R for the reinforcement learning model. I and model loss L I The direction that causes the model loss to decrease the fastest is obtained through backpropagation;

[0037] The model is iteratively optimized, the model parameter coefficients are updated, and finally the quantized correlation function f is obtained. I ;

[0038] According to the quantization function f I Calculate the objective factor index value I′ under the influence of correlation;

[0039] Establish a reinforcement learning model to quantify the association function f C (C t C s |I, P), and according to the quantization function f C (C t C s|I,P) Calculate the sensitivity metrics C′ of the power business operation environment C under the influence of objective information factor I and business participant experience P;

[0040] Establish a reinforcement learning model to quantify the association function f P (P r P a P e |I, C), and according to the quantization function f P (P r P a P e |I,C) Calculate the sensitivity metrics P′ of the business participant experience P under the influence of the power business operation environment C and objective information factor I.

[0041] Preferably, in step S4, the sub-step of optimizing the dynamic evaluation model using deep reinforcement learning includes:

[0042] Construct information sensitivity and sensitivity impact index x i Neural network model of mapping relationship NN e ;

[0043] The quantified values ​​of the sensitivity assessment indicators I, C, and P, as well as the quantified values ​​of the sensitivity indicators I′, C′, and P′ under the influence of multiple factors, are input into the neural network model NN. e In the process, an estimate of information sensitivity is obtained;

[0044] For each indicator x i The numerical value is used to calculate the learning incentive FR in each iteration, based on the information sensitivity estimate obtained from the model. se And calculate the model loss.

[0045] The direction of the value that causes the model loss to decrease the fastest is obtained through backpropagation of the neural network, and the weights of each neuron are updated.

[0046] The neural network is dynamically modified through multiple iterations until the model converges, the optimal index weight matrix is ​​determined, and finally the dynamically optimized data sensitivity evaluation model is obtained.

[0047] Preferably, the specific expression of the dynamic evaluation model is:

[0048] ;

[0049] in, Λ represents sensitivity, and Λ is the weighting coefficient.

[0050] The present invention also provides a dynamic evaluation system for the information sensitivity of smart grids, including a data information acquisition module, an index quantification module, a multi-factor correlation analysis module, a model construction and optimization module, and a sensitivity calculation module;

[0051] The data information acquisition module is used to acquire raw power grid data information, which includes data circulation environment information of the power grid to be evaluated, privacy data information of historical business participants' experiences, and historical structured electricity consumption data information.

[0052] The index quantification module is used to perform calculation and analysis based on the original power grid data information to obtain quantified values ​​of information sensitivity evaluation indicators such as objective information factor I, power business operation environment C, and business participant experience P.

[0053] The multi-factor correlation analysis module is used to analyze the correlation mapping relationship between objective information factor I, power business operation environment C, and business participant experience P, and uses a self-learning method to obtain the correlation mapping function f under the influence of multiple factors. I f C f P And the sensitivity metrics I′, C′, and P′ under the influence of multiple factors;

[0054] The model construction and optimization module is used to construct a dynamic evaluation model of the data sensitivity of the power grid to be evaluated based on the quantitative values ​​I, C, P of the sensitivity evaluation index and the quantitative indexes I′, C′, P′ of the sensitivity under the influence of multiple factors, and to use deep reinforcement learning to intelligently optimize the dynamic evaluation model to obtain the final evaluation model of the data sensitivity of the power grid to be evaluated.

[0055] The sensitivity calculation module is used to input the data information generated by the power grid to be evaluated into the data sensitivity assessment model to obtain the data sensitivity of the power grid to be evaluated. ;

[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0057] 1. The present invention provides a dynamic evaluation method and system for the sensitivity of smart grid information. The multi-dimensional quantitative system for the sensitivity of smart grid business information fully considers the objective privacy attributes of the data carrier itself, the power business operation environment, and the subjective privacy threat experience of related participating entities, and more accurately describes the actual privacy and security requirements of sensitive information in power business scenarios.

[0058] 2. The present invention provides a dynamic evaluation method and system for information sensitivity of smart grids, which takes into account the multiple impacts of business operation environment and business participant experience on information sensitivity, making the evaluation of information sensitivity more universal and targeted.

[0059] 3. The present invention provides a dynamic evaluation method and system for information sensitivity of smart grids, which adopts a dynamic feedback mechanism based on deep reinforcement learning, realizes dynamic optimization of the information sensitivity evaluation model based on scenario adaptation, improves the accuracy of information sensitivity calculation of smart grids, and thus provides a secure sharing and circulation scheme that matches "security requirements - security capabilities" for sensitive information related to smart grid businesses. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 Flowchart for dynamic assessment of information sensitivity

[0062] Figure 2 Information sensitivity assessment index system and framework diagram

[0063] Figure 3 Quantification of multi-factor association function for self-learning data sensitivity in power business scenarios

[0064] Figure 4 Flowchart for the dynamic self-optimization of a data sensitivity assessment model based on deep reinforcement learning

[0065] Figure 5 A dependency graph model for sensitive information owners and business platforms in community electricity usage scenarios.

[0066] Figure 6 Dependency graph of sensitive information owners in community electricity use scenarios Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0068] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other embodiments, well-known structures, circuits, materials, or methods have not been specifically described in order to avoid obscuring the invention.

[0069] Throughout this specification, references to "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "an embodiment," "an example," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0070] In the description of this invention, the terms "front", "rear", "left", "right", "up", "down", "vertical", "horizontal", "high", "low", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of this invention.

[0071] Example 1

[0072] This embodiment discloses a dynamic evaluation method for the information sensitivity of a smart grid, such as... Figures 1-6 As shown, this embodiment takes a local community household electricity consumption scenario as an example, and the method steps include:

[0073] S1: Obtain raw power grid data information, which includes historical structured electricity consumption data of the power grid to be evaluated, business data circulation environment information, and privacy data information of historical business participants;

[0074] The historical structured electricity consumption data includes user location information, user electricity consumption data, user electricity consumption time, and user service request information. The privacy data information experienced by historical business participants includes user ratings on the degree to which they are unwilling to disclose various types of privacy data, changes in user settings for various types of privacy data, and information on the degree of privacy protection of the current business operating environment, data privacy attributes, and the electricity service attributes used.

[0075] In step S1, the raw power grid data collected is acquired under a fixed power consumption environment. The model established after analyzing the collected data is also used to analyze and calculate the sensitivity of the smart grid under this power consumption environment. In community power consumption scenarios, user electricity consumption data is mostly collected by smart meters. This data is collected frequently and has fine granularity, containing a large amount of user privacy information such as household habits and lifestyles.

[0076] S2: Based on the original power grid data information, perform calculation and analysis to obtain specific indicators for objective information factor I, power business operation environment C, and business participant experience P sensitivity assessment.

[0077] In the context of household electricity consumption in communities, user sensitive data is granular and includes highly sensitive data attributes such as addresses and accounts. Based on these characteristics, objective privacy metrics such as information attributes, information scale, and information timeliness are first calculated, using the household electricity consumption data collected in Table 1 as an example:

[0078] Table 1. Example of household electricity consumption data collected in the community.

[0079]

[0080] In step S2, the sub-step of analyzing and processing the acquired raw power grid data information to obtain objective information factor I includes: performing numerical processing on historical structured electricity consumption data information to obtain objective information attribute I. a It is a numerical mapping based on sensitive information attributes. It assigns values ​​according to the inherent sensitivity of different data attributes, following the principle that the higher the sensitivity of the data attribute, the larger the mapped value. Different business scenarios may involve the same attribute A. i The possible range of values ​​for A is different; specifically, for the same attribute A... j Multiple sensor data collection values ​​are used to assign larger mapping values ​​to more sensitive types of values.

[0081] Let the dataset D = {D1, D2...D} m}, where D i It is the collected data of individual i, D i ={d i，1 d i，2 ...d i，n}, where d i j is attribute A of individual i. j The values ​​are determined by selecting n non-negative values ​​and mapping them to f1, f2, ... f. n The structured data is then processed into numerical values ​​to obtain the information attributes I of the collected data D. a ,Right now: ;

[0082] In this embodiment, based on the structured household electricity consumption data D collected by the smart meter, the data can be classified according to data type.

[0083] Privacy attributes are categorized into two types: numerical attributes (e.g., A4-A8) and categorical attributes (e.g., A1-A3). Numerical attributes can be further divided into benefit-type (e.g., A4-A6, A8) and cost-type (e.g., A7) based on their impact on privacy sensitivity. Benefit-type numerical attributes indicate that the larger the attribute value, the greater the sensitivity of the data. Cost-type attributes have the opposite effect. Therefore, the information attributes of the above types are quantified: for numerical attributes, the original measurement value is mapped to the interval [0, 1] using the deviation relativization method; for categorical attributes, their objective privacy attributes are calculated based on the granularity and sensitivity of the values, using a... j Represents attribute A j For a given value, the specific calculation formulas for each type of information attribute are as follows:

[0084] a. Numerical benefit-type attributes The calculation formula is as follows:

[0085] ;

[0086] b. Numerical cost-type attributes The calculation formula is as follows:

[0087] ;

[0088] c. Classification attributes The calculation formula is as follows:

[0089] ;

[0090] in, , These represent the maximum and minimum values ​​of this attribute in the current dataset, respectively. , They represent The granularity and sensitivity are quantified based on different attribute values, while having the following limitations: , This maps the attribute sensitivity to the [0,1] interval. For example, since an address is a categorical attribute with finer data granularity and higher sensitivity, its attribute sensitivity calculated using this formula is closer to 1. In summary, the collected data... Information attributes ,:

[0091] ;

[0092] in, Representing data attributes The subscript 8 represents the number of attributes in the community household electricity consumption data collection (as shown in Table 1); This represents a specific user, and 9 represents the current electricity usage scenario in the community (e.g., ...). Figure 5 The number of users (as shown); If it represents a category among the three attribute classifications, then There are three possible values: ; Indicates attribute category Included attributes gather; Representing attributes The corresponding category information attribute calculation function;

[0093] The time intervals and time spans of collecting historical structured electricity consumption data are calculated and analyzed to obtain an objective information scale I. s ; Calculate the time interval for collecting electricity consumption data based on the smart meter's data collection rules. Time span and data volume And calculate the quantization weight coefficient accordingly. Because attackers can obtain a household's electricity usage schedule and appliance usage by analyzing household electricity data, they can further analyze the household's demographics, daily routines, and even occupational backgrounds. Therefore, for household electricity data, the more data collected and the longer the time span, the greater the risk of privacy exposure. Conversely, the shorter the collection interval, the greater the possibility of detecting hidden information in the electricity data, thus posing a higher privacy risk. Therefore, the time span... and data volume The larger the value, the greater the impact on information scale indicators. The greater the degree of influence, the longer the time interval. The smaller the influence ratio, the lower the weighting coefficient. With time span and data volume Inversely proportional to the time interval It is directly proportional, and the specific calculation formula is as follows: Thus, the weighting coefficients are obtained. Then, based on this coefficient Calculation of information scale indicators: .

[0094] Based on the characteristic that electricity consumption data changes over time, the timeliness of obtaining objective information is crucial. tThe validity of data information is affected over time, which is a function of time t. This function can be divided into three categories: 1. Users' perception of a certain sensitive attribute increases over time, showing greater emphasis on that attribute, but this emphasis has an upper limit. 1. It is a bounded increasing function; 2. As time increases, users tend to increasingly ignore a certain sensitive attribute, and the degree of ignoring has a lower limit, i.e. 3. The degree of importance a user places on a particular sensitive attribute will not be affected over time, i.e. This is a constant function. In this embodiment, people typically only focus on electricity consumption over a recent period. Therefore, the information validity and sensitivity of the community household electricity consumption sensitive dataset decrease over time, and there is a lower bound, meaning σ(t) is a bounded decreasing function. Let the information timeliness quantification function for community household electricity consumption be:

[0095] ;

[0096] Taking the initial data acquisition time as time 0, where This function represents the standardized time value between the data sharing time and the initial collection time, mapping the timeliness of dataset information to a range. Furthermore, its timeliness decreases over time;

[0097] Objective information attributes of parameters Scale of objective information and the timeliness of objective information To obtain objective information factors, a comprehensive measurement of sensitive information about the power grid to be evaluated is required. :

[0098] ;

[0099] in, Used to control information attributes separately Information scale Information timeliness The impact factor of the indicator on the privacy sensitivity of the current dataset.

[0100] In step S2, the sub-step of analyzing and processing the original power grid data information to obtain the power business operation environment C includes: using the correlation analysis method of mutual information to construct an association dependency graph model;

[0101] In the process of cross-domain sharing and circulation of power data, data producers, data owners, data intermediaries, data users, and business platforms collectively constitute the data circulation environment. The entire lifecycle of data, from production and transmission to use, relies on this environment. The logical interconnections and data convergence relationships among the various entities involved in the power business together form a network of power data convergence and integration. Therefore, the power business operating environment mainly refers to the operating and support environment of the power business and the data circulation network environment across the entire business chain.

[0102] Therefore, considering the dependencies between data flow participants in the current power business operation environment, a correlation analysis using mutual information is employed to generate a dependency graph G′ for the power business operation environment. In this graph G′, if the current node (denoted as random variable G) is involved... i The set of directly associated nodes is called a Group. i Then node Gi and its directly associated node set Group i Form a privacy-related chain Group i →G i The relationship, namely the node G that is the owner of the power information. i Its privacy data in the current business operating environment and the Group i There is a chain-like flow relationship between the nodes, and the group is a collection. i The nodes in the array directly or indirectly depend on node G. i The data.

[0103] In this embodiment, in a community electricity consumption scenario, user electricity data is mostly collected by smart meters. This data is collected frequently and has fine granularity, containing a large amount of user privacy information such as household habits and lifestyles. Therefore, to protect household electricity privacy, a secure data aggregation method is often adopted in practical applications to upload and aggregate smart meter data. During the data aggregation process, firstly, the smart meters are randomly grouped, and the transmission path and aggregation node for the collected information are selected according to the group settings and the trust status of the node services in the current scenario. Then, according to business needs, the data is aggregated and reported level by level through the aggregation nodes, and finally delivered to the business platform for unified analysis and processing. Thus, in a community electricity consumption scenario, the logical interconnection and data aggregation relationship between the owner of sensitive information and the business platform due to the secure aggregation of sensitive information can be abstracted as follows: Figure 5 The association dependency model shown.

[0104] In the diagram, data production nodes within the same group analyze the status of candidate aggregation nodes (aggregation nodes indicated by dashed arrows in the diagram), select a primary aggregation node (i.e., the aggregation node indicated by solid arrows in the diagram) to share its data, and then each aggregation node aggregates and uploads the data to the data-using nodes for the business platform to use for data fusion analysis. As the business progresses, the status and trustworthiness of each node, as well as other factors such as the privacy and security experience of community users, will all affect this network relationship. This forms the basis for subsequent quantification of various data dimensions and the implementation of scenario-adaptive information sensitivity assessment.

[0105] In the power business operation environment, data sharing and circulation is a dynamic process. The data flow path and method depend on the data security protection capabilities provided by business participants and platforms within the power business operation environment, as well as the security status of data circulation nodes in that environment. Simultaneously, the security of the connections between participating entities in the data circulation environment is also a key factor leading to the leakage of power business data and individual privacy. Therefore, the impact of the power business operation environment on the privacy and security of sensitive information should be considered from the perspective of the entire data business chain circulation network environment, taking into account the service trustworthiness C between the participating entities in the data circulation. t Dependency Relevance C r And the correlation sensitivity C s Based on the state of each node in the data flow path within the aforementioned dependency graph model, the service reliability C in the power business scenario is obtained. t In the process of power business data sharing, data owners always choose entities with higher stability and security as service providers for their data sharing and use.

[0106] Therefore, its level of trust in the current service operation environment is determined by the state of each node in its data flow chain. That is, the service trustworthiness of the power business operation scenario represented by the dependency graph G′ is determined by the state of nodes G in the data owner node set. i The level of trust in the current business operating environment is used to represent:

[0107] ;

[0108] Where m is the number of data owner nodes in the current business operation environment, θ i These are the weight coefficients for each node. Represents node G i The level of trust in the current business operating environment. Specifically, The probability of normal service and the importance of each node in the node privacy association chain Gi→Groupi are used as the metric, i.e.:

[0109] ;

[0110] in, Group i A certain node G x The importance of a node is calculated using the in-degree and out-degree of that node in the dependency graph G′ of the power business scenario.

[0111] Right now ;n x Node G represents the process of sharing historical data. x The number of abnormal failures or attacks, n x This represents the total number of times the node provides data services, and the ratio of these two values ​​indicates the probability of abnormal data sharing behavior for that node. In summary, the service reliability C in the power business scenario... t The calculation formula is:

[0112] ;

[0113] In this embodiment, the constructed interconnection communication model describes the data dependencies between entities participating in data flow, thereby enabling the calculation of service trustworthiness. .like Figure 5 As shown, now assume node If it is a reference entity, then in the diagram... Vertices have adjacency-related groups of paths It constitutes a privacy-related chain. , i.e., node With sets There are direct or indirect information dependencies between nodes. Specifically, in Figure 5 In the middle, node , With nodes There is a direct dependency relationship, and the nodes Through nodes , With nodes There is an indirect dependency relationship. Therefore, during the data aggregation process in this pattern, nodes... For example, the service trustworthiness calculation in the current environment is as follows:

[0114]

[0115] ;

[0116] Similarly, the service trustworthiness of other nodes in the current environment. All of these can be quantified using the methods described above, ultimately yielding the service reliability under the current business operating environment, as shown below:

[0117] ;

[0118] in As the weight coefficients for each node, this embodiment of the invention uses... Figure 5 middle Node importance To calculate;

[0119] Mutual information is used to identify the correlation between nodes in the aforementioned dependency graph model, obtaining the dependency correlation degree Cr in the power business scenario. Since mutual information, as a measure of correlation analysis, has the advantage of low sensitivity compared to other statistical correlation identification methods, it can overcome the limitations of linear and nonlinear relationships. Furthermore, the correlation of nodes in a graph data structure reflects the approximate dependency strength of tuple nodes. Therefore, based on key pathways to privacy leakage and considering the security impact measurement of multi-entity associations with fine granularity, this method uses mutual information to identify the correlation between nodes in the power business operation environment, defining node G accordingly. i G j The dependency relationship between them is:

[0120] ;

[0121] Among them, D i D j Representing node G respectively i G j Data sets that can be shared between them Represents node G j For dataset D i Data item d i The shared dependency probability function similar; Represents node G i G j At the same time, the probability function that each node G has a shared dependency on data items in its own dataset, i.e.: j In the dataset D i Data item d i While there is a need for shared dependencies, node G i For dataset D j The data item dj also has shared dependency requirements. If using G... i and G j Inter-node dependency correlation C r (G i G j ) represents vertex G in graph G′ i To G j edge (G) i G j The weight w i，j Therefore, the dependency graph G′ of the power business operation environment is a weighted graph. Thus, the dependency graph model can be described by the dependency relevance adjacency matrix M as follows:

[0122] ;

[0123] Where m is the number of nodes in the graph, and the weight w i，j For the dependence of relevance G r (G i G j ), and 1≤i, j≤m, w i，j =0 indicates an independent node with no relation.

[0124] In this embodiment, the sensitive information owners and business platform node sets in the data dependency relationship model of each participating entity in the current community electricity consumption scenario are used as a basis. Using mutual information correlation analysis, a corresponding dependency graph of sensitive information owners and business platforms is generated (e.g., Figure 6 (As shown). The dependency correlation is obtained by calculating the mutual information between the sensitive data of each node. , representing vertices in the dependency graph arrive edge weights For example, using nodes and Dependency Relevance The weight of the edge connecting the two is:

[0125] ;

[0126] Therefore, the edges in the graph have dependency weights, forming a weighted directed graph. The dependencies between nodes constitute the edge set, resulting in a dependency graph among owners of sensitive information in the community electricity consumption scenario, such as... Figure 6 As shown.

[0127] Therefore, the dependency graph model can be used with a dependency relevance adjacency matrix. The description is as follows:

[0128] ;

[0129] Among them, based on dependency relevance The definitions include: That is, the rectangle is a symmetric matrix.

[0130] Then, according to The dependency correlation of the node adjacency related groups is calculated. Association sensitivity That is, in Figure 6 In the middle, the picture shows... Vertices have adjacency-related groups of paths This constitutes a privacy-related chain. Calculated based on this Association sensitivity :

[0131] ;

[0132] In the adjacency matrix of dependency correlation In the representation, Use the weights between matrix nodes The calculation yields: ,but The specific calculation formula is as follows:

[0133] ;

[0134] Similarly, other nodes' sensitivity to the current environment All of these can be quantified using the methods described above, ultimately yielding the correlation sensitivity under the current business operating environment. :

[0135] ;

[0136] Based on parameter service reliability and correlation sensitivity Calculate and obtain the power business operating environment .

[0137] ;

[0138] in These represent the service credibility. and correlation sensitivity Weight distribution coefficients in measuring the impact of sensitive information on privacy and security in the power business operation environment.

[0139] In step S2, the sub-step of analyzing and processing the acquired raw power grid data information to obtain the business participant experience P includes:

[0140] Users are rated on their level of unwillingness to disclose various types of private data. A user privacy preference weight value is calculated based on this rating, and then this weight value is used to calculate the user privacy preference P. r Suppose that, based on the community household electricity consumption data collected in Table 1, a user has a specific privacy attribute A. j The degree to which users are unwilling to disclose their data is scored, and the final weight value p of each data attribute is calculated. j Then, the privacy preference weight vector of user i in the current environment is composed of the weight values ​​of the degree of disclosure of each privacy attribute data. for: Users' privacy preferences regarding their shared data He comprehensively measures the degree of privacy preference for different data items and attributes, that is, the user's... Privacy preferences for:

[0141] ;

[0142] in Indicates user Privacy preference vectors for their data items The weight of the degree of unwillingness to disclose This refers to the single information attribute calculated after classifying the data attributes in step 2.

[0143] Based on changes in users' settings for various types of privacy data, privacy awareness is gained. First, a survey was conducted to summarize the data collection settings of all users' smart meters in the current community environment. Changes in user privacy settings were recorded, and the strictness of user privacy settings was scored and ranked. Based on this, user privacy awareness was quantified.

[0144] ;

[0145] in, For power business data attributes Visibility, The value can be 0 or 1, where 0 means the property is not visible and 1 means the property is visible; This indicates the amount of change in the visibility of user-defined data. For attributes Privacy preference weights; , These represent the proportions of more lenient and stricter privacy settings for other users in the current power business operating environment compared to this user;

[0146] Based on the user's perception of the privacy protection level of the current business operating environment, data privacy attributes, and the electricity service they use, privacy expectations are obtained. First, we collect the privacy protection expectations of each user in the current community environment. Then, based on the privacy expectation quantification table, we quantify the privacy expectations of family users. Finally, we perform unified numerical processing and map the value to... Interval.

[0147] Table 2 Quantification of Privacy Expectations

[0148] Privacy Expectations Privacy protection level [0,3) weak [3,5) Low [5,7) middle [7,9) higher [9,10] high

[0149] User privacy preferences for parameters Privacy awareness and privacy expectations Perform quantitative calculations to obtain the experience of business participants. Based on the above analysis, the privacy and security experience of power business participants regarding sensitive information can be assessed using privacy preferences (P). r Privacy awareness P a and privacy expectations P e To achieve comprehensive quantization, the expression is as follows:

[0150] P=γ1P r +γ2P a +γ3P e ;

[0151] Wherein, γ1+γ2+γ3=1 is used to control the weighting factors of each indicator for measuring the privacy and security experience of sensitive information of the above-mentioned business participants.

[0152] S3: Analyze the correlation mapping relationship between objective information factor I, the power business operation environment C, and the experience of business participants P, and use a self-learning method to obtain the correlation mapping function f under the influence of multiple factors. I f C f P And the sensitivity metrics I′, C′, and P′ under the influence of multiple factors.

[0153] In diverse and complex power business scenarios, the various dimensions of sensitivity indicators for business information are not entirely independent but rather interconnected. Firstly, the objective privacy measurement of power business data is influenced by the subjective privacy threat experiences of relevant entities within the business operation environment. On one hand, the collection, sharing, and transmission of power business data depend on the business needs of different power scenarios; on the other hand, the sensitivity of the data largely depends on the subjective privacy experience of the data owner. Therefore, the business operation environment and business participants have the following relationship regarding the privacy measurement of sensitive power information: I′=f I (I r I a I t |C,P),f I This represents the measurement function of the objective information factor I under the influence of the business operation environment C and the sensitive information privacy and security experience P of business participants.

[0154] Secondly, the impact of the power business operating environment on the privacy and security of sensitive information is also influenced to some extent by the subjective experience of business participants and the objective privacy measurement of data. For example, when data owners subjectively perceive a significant risk of privacy leakage of their sensitive data, they will adjust the way they share their private data and the privacy visibility settings, thereby leading to changes in the data circulation network environment across the entire business chain and the objective amount of data privacy. Therefore, the subjective privacy and security experience of business participants and the objective attributes of sensitive information have the following relationship with the power business operating environment: C′=f C (C t C s |I, P), where f C This represents the privacy and security measurement function of the business operation environment C on sensitive information, under the influence of the business participants' sensitive information privacy and security experience P and the sensitive information objectivity privacy measurement I.

[0155] Furthermore, as the main actors sharing private data, participants in the power industry are significantly influenced by the privacy and security experiences of the power data itself, as well as the security status of the power business operating environment. For example, even datasets with the same sensitive attributes may have different privacy preferences and expectations depending on the operating environment. Additionally, the objectivity and privacy of power data also affect data owners' expectations for the protection of sensitive data. Therefore, the power business environment and the objective privacy of sensitive information have the following relationship with the privacy and security experiences of business participants: P′=f P (P r P a P e |I,C), where f P This represents a measurement function for the sensitive information privacy and security experience P of business participants under the influence of the objectivity and privacy measurement of sensitive information (I) and the business operation environment (C).

[0156] In this embodiment, the quantification of the multi-factor correlation function of data sensitivity in the community electricity consumption scenario mainly targets the dynamic measurement system of information sensitivity under the influence of multiple factors. It establishes a self-learning feedback mechanism to quantify the interaction between various factors. Therefore, during the self-learning process, it determines the mapping function f between each factor. I (I r I a I t |C,P)f C (C t C s |I,P),f P (P r P a P e|I,C) as the learning action (A), minimizing the difference between the indicator evaluation value and the actual result as the decision to take action (Policy,P), and evaluating the effect of the model as the incentive (Reward,R).

[0157] In this embodiment, the objective attributes of household electricity consumption sensitive data are used. and community electricity service environment Improving the privacy and security of electricity use for community users Taking the influence relationship as an example, this influence function is explained in detail. Multinomial regression self-learning optimization, other correlation functions , The quantification method is similar, and the specific process is as follows: Figure 3 As shown, the specific steps for implementation are as follows:

[0158] ① Objective attributes of household electricity consumption sensitive data and community electricity service environment Privacy and security experience influenced by factors such as It can be represented as: ;

[0159] ② and Factors such as user privacy and security experience The magnitude of their correlation impact is constrained by their respective secondary indicator factors, and in reality, these secondary indicator factors have a significant impact on the privacy and security experience. There are both independent influences and correlated influences resulting from the combined effects of multiple indicators. Therefore, user privacy and security experience... of and Multifactor association function The following polynomial regression equation can be used for fitting:

[0160] ;

[0161] in The factors that affect user privacy and security experience include the following five: information attributes. Information scale Information timeliness Service credibility and correlation sensitivity etc.; cross terms Indicator Factors , The combined impact on user privacy and security experience; specifically, when Time factor The power term indicates that the factor has a stronger power-law impact on the privacy and security experience.

[0162] To simplify the expression, the interaction terms of the polynomial regression equation are... Replace with variable ,Right now Indicates related factors , The combined effect of these factors. Therefore, in and User privacy and security experience influenced by factors such as The polynomial regression equation can be reformulated as:

[0163] ;

[0164] This expression can be represented in matrix form as follows: ;in These are the weight coefficients for each factor item. It is an input data matrix composed of independent factor terms, power terms, and cross terms. For constant terms;

[0165] ③ Regarding user privacy and security experience Multifactor association function The aim is to fit the model and optimize the parameters of the polynomial regression equation using reinforcement learning methods. and This is implemented in the form of [formula missing]. Since the highest power of the polynomial above is 2, a base model with a degree of 2 must be used. Meanwhile, the learning rate determines the convergence performance of the machine learning model: an excessively large learning rate will prevent the model from converging, while an excessively small learning rate will slow down the convergence speed. Therefore, the initial learning rate for the regression model should ideally be set between 0.01 and 0.001.

[0166] ④ Set an incentive function to calculate the multi-factor association function. The learning incentives during the fitting process can be influenced by the user's privacy and security experience in the current scenario. The sum of the differences between the actual and predicted values ​​is expressed as follows:

[0167] ;

[0168] in, It is the actual value of each user's privacy and security experience. It is the model prediction value corresponding to it;

[0169] ⑤ Calculate the model loss degree based on the squared difference function, which can be obtained from the target stimulus value. Compared with actual incentive value The differences are expressed as follows: ;

[0170] ⑥ The decision in this learning process is to minimize the model loss. That is, by using stochastic gradient descent, we find the direction in which the model loss decreases the fastest. Then, based on the gradient descent direction obtained from the decision, we calculate the parameter vector, thereby updating the weight coefficients and obtaining the correlation function model after this round of optimization.

[0171] ⑦ To ensure more accurate model convergence, the learning rate is dynamically adjusted during each training round, and steps ④ through ⑥ are repeated iteratively to optimize the model. Finally, the converged model is saved, which is the correlation function obtained through self-learning optimization. .

[0172] Based on the multi-factor correlation mapping function f obtained by quantification I (I r I a I t |C,P)f C (C t C s |I,P),f P (P r P a P e |I,C), calculate the sensitivity quantification index I′ of objective information factor I under the influence of power business operation environment C and business participant experience P; calculate the sensitivity quantification index C′ of power business operation environment C under the influence of objective information factor I and business participant experience P; calculate the sensitivity quantification index P′ of business participant experience P under the influence of power business operation environment C and objective information factor I.

[0173] S4: Based on the quantified values ​​I, C, and P of the sensitivity assessment index and the quantified values ​​I′, C′, and P′ of the sensitivity under the influence of multiple factors, construct a dynamic assessment model of the data sensitivity of the power grid to be assessed, and use deep reinforcement learning to dynamically optimize the dynamic assessment model to obtain the final assessment model of the data sensitivity of the power grid to be assessed.

[0174] This learning process uses optimizing the weight coefficients of each level of the comprehensive evaluation model as actions (A), minimizing the difference between the sensitivity assessment value and the user's privacy expectations as the decision to take action (P), and the model evaluation effect as the incentive (R). That is, it adopts an "action-incentive-decision-action" cyclical feedback method to optimize the comprehensive data sensitivity assessment model. The specific implementation of actions and decisions is based on neural network model training. The quantified values ​​of information sensitivity assessment indicators are used as input to the neural network, including objective measurement indicators and indicators under the influence of multiple factors. After learning and mapping by the neural network, the data sensitivity assessment result can be obtained. The coefficients of each neuron in the neural network constitute the weight values ​​of the sensitivity assessment model.

[0175] In step S4, such as Figure 4 As shown, the sub-steps for optimizing the dynamic evaluation model using deep reinforcement learning include:

[0176] ①The dynamic comprehensive evaluation model for data sensitivity in community electricity consumption scenarios can be represented by the following linear regression equation:

[0177]

[0178] in These are the weighting coefficients for each indicator. The quantitative values ​​for each indicator include the steps. The independent index values ​​calculated in step 1 and the index values ​​quantified by the multi-factor association function in step 5 are used to form the regression equation. The regression equation can then be represented in matrix form as follows: ,in It is to fit the parameter matrix of each index using reinforcement learning. It is the input data matrix. For constant terms;

[0179] ② Construct a neural network model based on the linear regression equation above. The input layer has 11 neurons (including secondary indicators of 8 measurement systems and 3 quantitative indicators under the influence of multiple factors), the first hidden layer has 3 neurons (the number of primary indicators), and the output layer has 1 neuron. Its output is the comprehensive evaluation of information sensitivity. ;

[0180] ③ Initialize the model using the combined weights of a random normal distribution as the weights of each neuron in the neural network. Similarly, to ensure model convergence, a learning rate in the range of 0.01 to 0.001 is selected. An activation function is set to map the input data to the (0,1) interval, thereby obtaining the normalized information sensitivity. The output value is used as input to train the model, taking the quantitative values ​​of each indicator and the influence of the multi-factor correlation mapping function on the indicator.

[0181] ④ Calculate the learning motivation for this learning process based on the incentive value function. It obtains information sensitivity through model calculation. and privacy expectations The differences are expressed as follows:

[0182] ;

[0183] ⑤ Then, the expected learning incentive value is calculated using the loss function. and actual incentive value The difference is used to obtain the model loss as follows: ;

[0184] ⑥ The decision-making process of the information sensitivity comprehensive model in its self-learning process is to minimize the model loss. That is, to find the model loss using stochastic gradient descent. The direction of fastest descent. Then, the coefficients of each neuron in the neural network are calculated through backpropagation, thereby updating the model weight coefficients and obtaining the optimized information sensitivity comprehensive model for this round.

[0185] ⑦ Repeat steps ④ to ⑥ to dynamically update the model parameters by optimizing the weight coefficients of each neuron until the model converges. The input layer neuron coefficients then represent the weights of each secondary indicator and its associated influences, while the hidden layer neuron coefficients represent the weights of the primary indicators in the comprehensive evaluation model. This results in the optimized data sensitivity evaluation model for the community electricity consumption scenario. .

[0186] S5: Input the data information generated by the power grid to be evaluated into the dynamic evaluation model to obtain the sensitivity of the power grid to be evaluated. The specific expression of the dynamic evaluation model is as follows:

[0187]

[0188]

[0189]

[0190] Where Λ is the weighting coefficient; f I (I r I a I t |C,P)f C (C t C s |I,P),f P(P r P a P e |I, C) represent the correlation and influence between various factors. They describe the adaptive mechanism of business scenarios under the influence of multiple factors from the perspective of function model. I, C, and P are the quantitative values ​​of each primary indicator. They, together with the corresponding indicator weight vectors Θ = {Θ1, Θ2, Θ3} and Φ = {Φ1, Φ2, Φ3}, and the weight factors of their respective secondary indicators (e.g., α1, β1, γ1, etc.), realize the dynamic optimization of the information sensitivity assessment model from the data aspect.

[0191] The dynamic assessment method for information sensitivity of smart grids provided in this embodiment can achieve dynamic optimization of the scenario-adaptive information sensitivity assessment model under the premise of comprehensively considering multiple factors. This makes the assessment of information sensitivity more universal and targeted, while improving the accuracy of information sensitivity calculation for smart grids. As a result, it provides support for a secure sharing and circulation scheme that matches "security requirements - security capabilities" for sensitive information related to smart grid businesses.

[0192] Example 2: This example discloses a dynamic evaluation system for the information sensitivity of a smart grid. This example is designed to implement the dynamic evaluation method described in Example 1, and includes a data information acquisition module, an index quantification module, a multi-factor correlation analysis module, a model construction and optimization module, and a sensitivity calculation module.

[0193] The data information acquisition module is used to acquire raw power grid data information, which includes data circulation environment information of the power grid to be evaluated, privacy data information of historical business participants' experiences, and historical structured electricity consumption data information.

[0194] The index quantification module is used to perform calculation and analysis based on the original power grid data information to obtain the quantified values ​​of objective information factor I, power business operation environment C, and business participant experience P information sensitivity evaluation indicators.

[0195] The multi-factor correlation analysis module is used to analyze the correlation mapping relationship between objective information factor I, power business operation environment C, and business participant experience P, and uses a self-learning method to obtain the correlation mapping function f under the influence of multiple factors. I f C f P And the sensitivity metrics I′, C′, and P′ under the influence of multiple factors;

[0196] The model construction and optimization module is used to construct a dynamic evaluation model of the data sensitivity of the power grid to be evaluated based on the quantitative values ​​I, C, P of the sensitivity evaluation index and the quantitative indexes I′, C′, P′ of the sensitivity under the influence of multiple factors, and to use deep reinforcement learning to intelligently optimize the dynamic evaluation model to obtain the final evaluation model of the data sensitivity of the power grid to be evaluated.

[0197] The sensitivity calculation module is used to input the data information generated by the power grid to be evaluated into the data sensitivity assessment model to obtain the data sensitivity of the power grid to be evaluated. .

[0198] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic evaluation method for the information sensitivity of a smart grid, characterized in that, The method steps include: S1: Obtain raw power grid data information, which includes historical structured electricity consumption data of the power grid to be evaluated, business data circulation environment information, and privacy data information of historical business participants; S2: Based on the original power grid data information, perform calculation and analysis to obtain the quantitative values ​​of objective information factor I, power business operation environment C, and business participant experience P information sensitivity evaluation index; S3: Analyze the correlation mapping relationship between objective information factor I, the power business operation environment C, and the experience of business participants P, and use a self-learning method to obtain the correlation mapping function under the influence of multiple factors. And the sensitivity metrics I′, C′, and P′ under the influence of multiple factors; S4: Based on the quantified values ​​I, C, and P of the sensitivity assessment index and the quantified values ​​I′, C′, and P′ of the sensitivity index under the influence of multiple factors, construct a dynamic assessment model of the data sensitivity of the power grid to be assessed, and use deep reinforcement learning to intelligently optimize the dynamic assessment model to obtain the data sensitivity assessment model of the power grid to be assessed. S5: Input the data information generated by the power grid to be evaluated into the data sensitivity assessment model to obtain the data sensitivity of the power grid to be evaluated. ; The historical structured electricity consumption data includes user location information, user electricity consumption data, and user service request information. The business data circulation environment information includes business operation and support environment information and data full business chain circulation network environment information. The privacy data information of historical business participants includes user rating information on the degree of unwillingness to disclose privacy data, user settings for privacy data, and user's expected level of privacy protection. In step S2, the sub-step of analyzing and processing the original power grid data to obtain objective information factor I includes: Historical structured electricity consumption data is digitized to obtain objective information attribute I. a The information attribute I a Numerical mapping of attribute I, which is an objective information factor; The time intervals and time spans of collecting historical structured electricity consumption data are calculated and analyzed to obtain an objective information scale I. s The information scale I s For objective information factor I, the size of the data; Based on the characteristic that electricity consumption data changes over time, the timeliness of obtaining objective information is crucial. t The information timeliness I t The impact of time variations on objective information factor I data attributes; Attribute I of objective information of parameters a Scale of objective information I s and the timeliness of objective information I t The sensitive information of the power grid to be evaluated is comprehensively measured to obtain objective information factor I; In step S2, the sub-step of analyzing and processing the acquired raw power grid data information to obtain the power business operation environment C includes: A correlation dependency graph model is constructed using the mutual information correlation analysis method. Based on the state of each node in the data flow path in the aforementioned dependency graph model, the service reliability in the power business scenario is obtained. The credibility of the service The level of trust in the ability to provide data privacy protection for the power business operating environment under the current services; Mutual information is used to identify the relationships between nodes in the dependency graph model, thereby obtaining the dependency relevance in the power business scenario. The dependency relevance This refers to the interrelationships between nodes that have dependencies in the power business operation environment. Through the dependency correlation Quantify the correlation sensitivity of the power business operating environment. The correlation sensitivity The level of privacy and security of sensitive information caused by the interrelationships and dependencies between nodes in the power business operation environment; Based on parameter service reliability and correlation sensitivity Calculate and obtain the power business operating environment C; In step S2, the sub-step of analyzing and processing the original power grid data information to obtain the business participant experience P includes: Users are rated on their level of unwillingness to disclose various types of private data. A user privacy preference weight value is calculated based on this rating, and then this weight value is used to calculate the user privacy preference P. r The user privacy preference P r This refers to the degree of user preference for different objective information factors; Based on changes in users' settings for various types of privacy data, privacy awareness P is obtained. a The privacy awareness P a To raise users' awareness of protecting their sensitive information from security threats during data sharing; Based on the user's perception of the privacy protection level of the current business operation environment, data privacy attributes, and the electricity service attributes used, the privacy expectation P is obtained. e The privacy expectation P e This represents users' expectations regarding the level of data privacy protection they receive. User privacy preferences for parameters P r Privacy awareness P a and privacy expectations P e Perform quantitative calculations to obtain the experience P of business participants.

2. The method for dynamically evaluating the information sensitivity of a smart grid according to claim 1, characterized in that, The sub-steps of step S4 include: Establish a correlation mapping function between two factors—the power business operation environment (C) and the business participant experience (P)—and the objectivity of sensitive information factor (I). ; Establish objective information factor I and correlation mapping function A reinforcement learning model for the mapping relationship between them; By inputting the power business operation environment C and the business participant experience P into the reinforcement learning model, the sensitivity quantification index I′ of the objective information factor I under the influence of the power business operation environment C and the business participant experience P is obtained. Calculate the learning incentives for the reinforcement learning model. and model loss The direction that causes the model loss to decrease the fastest is obtained through backpropagation; The model is iteratively optimized, the model parameter coefficients are updated, and finally the quantized correlation mapping function is obtained. ; According to the quantized correlation mapping function Calculate the objective factor index value I′ under the influence of correlation; Establish a reinforcement learning model to quantize the association function And quantize the association function according to the reinforcement learning model. Calculate the sensitivity quantification C′ of the power business operating environment C under the influence of objective information factor I and business participant experience P; Establish a reinforcement learning model to quantize the association function And quantize the association function according to the reinforcement learning model. The sensitivity quantification of the business participant experience P under the influence of the power business operating environment C and objective information factor I is P′.

3. The method for dynamically evaluating the information sensitivity of a smart grid according to claim 1, characterized in that, In step S4, the sub-step of optimizing the dynamic evaluation model using deep reinforcement learning includes: Constructing information sensitivity and sensitivity impact indicators Neural network model of inter-mapping relationship ; The quantified values ​​of the sensitivity assessment indicators I, C, and P, as well as the quantified values ​​of sensitivity under the influence of multiple factors I′, C′, and P′, are input into the neural network model. In the process, an estimate of information sensitivity is obtained; For each indicator The numerical value is used to calculate the learning incentive in each iteration, based on the information sensitivity estimate obtained from the model. And calculate the model loss. The direction of the value that causes the model loss to decrease the fastest is obtained through backpropagation of the neural network, and the weights of each neuron are updated. The neural network is dynamically modified through multiple iterations until the model converges, the optimal index weight matrix is ​​determined, and finally the dynamically optimized data sensitivity evaluation model is obtained.

4. The method for dynamically evaluating the information sensitivity of a smart grid according to claim 2, characterized in that, The specific expression for the dynamic evaluation model is as follows: ； For sensitivity, Λ is the weighting coefficient. , These represent the indicator weights of I, C, and P, respectively, and the indicator weights of the sensitivity metrics I′, C′, and P′.

5. A dynamic evaluation system for the information sensitivity of a smart grid, characterized in that, The system is used to perform the dynamic evaluation method for information sensitivity of smart grids according to any one of claims 1 to 4, wherein the system includes a data information acquisition module, an index quantification module, a multi-factor correlation analysis module, a model construction and optimization module, and a sensitivity calculation module. The data information acquisition module is used to acquire raw power grid data information, which includes data circulation environment information of the power grid to be evaluated, privacy data information of historical business participants, and historical structured electricity consumption data information. The index quantification module is used to perform calculation and analysis based on the original power grid data information to obtain the quantified values ​​of objective information factor I, power business operation environment C, and business participant experience P information sensitivity evaluation indicators. The multi-factor correlation analysis module is used to analyze the correlation mapping relationship between objective information factor I, power business operation environment C, and business participant experience P, and uses a self-learning method to obtain the correlation mapping function under the influence of multiple factors. And the sensitivity metrics I′, C′, and P′ under the influence of multiple factors; The model construction and optimization module is used to construct a dynamic evaluation model of the data sensitivity of the power grid to be evaluated based on the quantitative values ​​I, C, P of the sensitivity evaluation index and the quantitative indexes I′, C′, P′ of the sensitivity under the influence of multiple factors, and to use deep reinforcement learning to intelligently optimize the dynamic evaluation model to obtain the final evaluation model of the data sensitivity of the power grid to be evaluated. The sensitivity calculation module is used to input the data information generated by the power grid to be evaluated into the data sensitivity assessment model to obtain the data sensitivity of the power grid to be evaluated. .

Images