A cloud-based pumped storage power station intelligent inspection equipment credit evaluation method and device

Abstract

This invention provides a cloud-based method and apparatus for reputation assessment of intelligent inspection equipment in pumped-storage power stations. The method includes the following steps: S1, the cloud platform collects reputation assessment indicators for each intelligent inspection equipment point, and uses a BP neural network to calculate the reputation of each equipment point; S2, based on the reputation of each equipment point, a directed weighted network model for point reputation is constructed to calculate the credibility of each equipment and determine whether it is a single-point attacker; S3, based on the equipment entity attributes, a knowledge graph is used to calculate the behavioral-semantic fusion similarity between equipment and determine whether it is a collusive attacker. This invention solves the problems of existing related technologies having insufficient objectivity, low accuracy, and limited application scenarios in reputation assessment under single-point attack and collusive attack scenarios.

Description

Technical Field

[0001] This invention relates to the field of IoT security for pumped storage power stations, and more particularly to a cloud-based method and apparatus for credit assessment of intelligent inspection equipment for pumped storage power stations. Background Technology

[0002] With the rapid development of IoT technology, various intelligent inspection devices are widely used in pumped storage power stations to assist staff in conducting 24 / 7, multi-directional anomaly detection. The IoT in pumped storage power stations is highly open, decentralized, and dynamic, making it vulnerable to attacks from both external and internal sources. While authentication, firewalls, and cryptographic technologies can effectively defend against external attacks, they are ineffective against internal attacks launched by malicious devices that have gained access to the network.

[0003] To avoid interaction with malicious devices joining the network and block internal attack routes, intelligent inspection devices can securely and reliably share information based on trust relationships between network-connected devices. In existing reputation assessment methods, device A can assess the point reputation of device B based on the quality of information sent to it, providing a reference for determining device B's trustworthiness. Furthermore, the point reputation of a device from multiple assessments can be aggregated into the device's global reputation, used to identify potential malicious devices.

[0004] Currently, there is no research on reputation assessment methods for intelligent inspection equipment in pumped storage power stations. Applying reputation assessment methods to intelligent inspection systems for pumped storage power stations presents three pressing issues: 1) Due to limited computing and storage resources, as the number of devices and the amount of interactive data increase, it becomes difficult to guarantee the sufficiency of collected information and the accuracy of reputation calculations; 2) Single-point attacks on reputation, where a malicious device assigns a false reputation value to another device to mislead the trustworthiness judgments and detection of other devices; 3) Collusive attacks on reputation, where multiple malicious devices collude to maliciously lower or raise the overall reputation of certain devices.

[0005] The aforementioned issues directly affect the objectivity and accuracy of equipment credibility. Solving these problems is a necessary condition for ensuring the security of the IoT-based intelligent inspection system for pumped storage power stations. Summary of the Invention

[0006] The first objective of this invention is to provide a cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations, addressing the aforementioned problems.

[0007] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0008] A cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations, characterized by the following steps:

[0009] S1. The cloud platform collects reputation evaluation indicators for each intelligent inspection equipment point and uses a BP neural network to calculate the reputation of each equipment point.

[0010] S2. Based on the point reputation of each device, construct a point reputation directed weighted network model, calculate the credibility of each device, and determine whether it is a single point attacker;

[0011] If the judgment result is yes, the point reputation of this device will not be included in the global reputation calculation of other devices;

[0012] Otherwise, participate in the global reputation calculation;

[0013] S3. Based on device entity attributes, use knowledge graph to calculate the behavior-semantic fusion similarity between devices and determine whether they are colluding attackers.

[0014] When the judgment result is yes, the point reputation of the colluding equipment assessment will not participate in the global reputation calculation of other equipment;

[0015] Otherwise, it will participate in the global reputation calculation.

[0016] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0017] As a preferred technical solution of the present invention: in step S1, the credit evaluation index of the intelligent inspection equipment includes: inherent attribute index and performance attribute index;

[0018] The inherent attribute indicators are: standardization, protocol, and descriptiveness;

[0019] The performance attributes are: response time, throughput, latency, connection success rate, availability, and reliability.

[0020] As a preferred technical solution of the present invention: In step S1, specifically, on the cloud platform, the point reputation evaluation index of each intelligent inspection device is input into the BP neural network to obtain the point reputation of each intelligent inspection device.

[0021] As a preferred technical solution of the present invention: In step S2, specifically, on the cloud platform, a point reputation directed weighted network is generated based on the point reputation evaluation relationship between intelligent inspection devices and the point reputation obtained from the evaluation.

[0022] In the directed weighted network of point reputation, each node represents a smart inspection device in a pumped storage power station, and the directed edges between nodes represent the point reputation evaluation relationship between the smart inspection devices, pointing from the point reputation evaluator to the evaluated device. The weight of the directed edge represents the point reputation being evaluated.

[0023] The credibility of each intelligent inspection device is calculated based on a point-reputation directed weighted network.

[0024] Credibility cr i The calculation formula is as follows:

[0025]

[0026] Among them, cr i Indicates device u i Credibility, Indicates the device u i The evaluated set of devices, r ij Indicates device u i Evaluation equipment u j Points of reputation, Indicates device u j Average point reputation;

[0027] Average Points Reputation The calculation formula is as follows:

[0028]

[0029] in, Indicates the evaluation equipment u j A collection of devices with a reputation for accuracy, r kj Indicates device u k Evaluation equipment u j Points of reputation;

[0030] The criteria for determining a single point of attack are:

[0031]

[0032] Where n is the number of nodes in the point reputation directed weighted network. If the credibility of a node is less than μ, the intelligent inspection device corresponding to that node is identified as a single point attacker.

[0033] As a preferred technical solution of the present invention: In steps S2 and S3, the global reputation calculation formula is as follows:

[0034]

[0035] Among them, R j Indicates device u j Global reputation Indicates the evaluation equipment u j A collection of devices that provide credit rating. For device u i The credibility after normalization, r ij Indicates device u i Evaluation equipment u j Points of credibility.

[0036] As a preferred technical solution of the present invention: in step S3, the device entity attributes include: identification tag, global reputation, credibility, number of evaluations, trusted object, untrusted object, and communication object.

[0037] As a preferred technical solution of the present invention: In step S3, specifically, on the cloud platform, a knowledge graph is generated based on the entity attributes of the intelligent inspection equipment to obtain a vectorized representation of entities and relationships;

[0038] The formula for calculating the similarity between device behaviors and semantics is as follows:

[0039] fs ij =α·ks ij +(1-α)·bs ij (6)

[0040] Among them, fs ij bs ij ks ij Representing device u respectively i and equipment u j The similarity between them is calculated based on fusion similarity, behavioral similarity, and semantic similarity, with α representing the weight coefficient.

[0041] The criteria for identifying colluding attackers are as follows: based on the actual situation of the intelligent inspection equipment network in the pumped storage power station and the scale of the colluding attack, a fusion similarity threshold ε and a quantity threshold δ are set. When the fusion similarity between devices is greater than the threshold ε, the two devices are marked as suspicious colluding devices.

[0042] When the number of suspicious devices with similar fusion properties exceeds the threshold δ, these are identified as colluding attackers.

[0043] As a preferred technical solution of the present invention: behavioral similarity bs ij The calculation formula is as follows:

[0044]

[0045] Where n represents the total number of intelligent inspection devices in the pumped storage power station, and r ik and r jk Representing device u respectively i and equipment u j The equipment being evaluated k Points of credibility.

[0046] As a preferred technical solution of the present invention: semantic similarity ks ij The calculation formula is as follows:

[0047]

[0048] Where d represents the entity and relation represent the dimension of the vector space, l ki and l kj Representing device u respectively i and equipment u j The value of the corresponding vector in the k-th dimension.

[0049] Another objective of this invention is to provide a cloud-based intelligent inspection equipment reputation assessment device for pumped storage power stations.

[0050] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0051] A cloud-based intelligent inspection equipment reputation assessment device for pumped storage power stations, characterized in that: the device comprises:

[0052] The cloud platform module is used to collect reputation evaluation indicators of various intelligent inspection devices and perform data processing tasks with high computing resource requirements.

[0053] The point reputation assessment module calculates the point reputation of each device based on the reputation assessment indicators of each intelligent inspection device and using a BP neural network.

[0054] The single-point attacker identification module constructs a directed weighted network model of point reputation based on the point reputation of each device, calculates the credibility of each device, and determines whether it is a single-point attacker. If the determination result is yes, the point reputation evaluated by the device does not participate in the global reputation calculation of other devices; otherwise, it participates in the global reputation calculation.

[0055] The collusion attacker identification module uses a knowledge graph based on device entity attributes to calculate the behavior-semantic fusion similarity between devices and determine whether they are collusion attackers. If the identification result is yes, the point reputation of the collusion device is not included in the global reputation calculation of other devices; otherwise, it is included in the global reputation calculation.

[0056] The global reputation assessment module calculates the global reputation of each intelligent inspection device based on its point reputation and normalized credibility.

[0057] This invention provides a cloud-based method and apparatus for credit assessment of intelligent inspection equipment in pumped storage power stations. The method includes the following steps: S1. A cloud platform collects credit assessment indicators for each intelligent inspection device and uses a BP neural network to calculate the credit of each device. S2. Based on the credit of each device, a directed weighted network model for credit is constructed to calculate the credibility of each device and determine whether it is a single-point attacker. If the determination result is yes, the credit of the device is not included in the global credit calculation of other devices; otherwise, it is included in the global credit calculation. S3. Based on the entity attributes of the devices, a knowledge graph is used to calculate the behavioral-semantic fusion similarity between devices and determine whether they are colluding attackers. If the determination result is yes, the credit of the colluding device is not included in the global credit calculation of other devices; otherwise, it is included in the global credit calculation. Compared with existing technologies, this invention comprehensively considers multi-dimensional indicators related to information interaction with intelligent inspection equipment to evaluate the point reputation of the equipment. It uses a BP neural network to represent the nonlinear mapping from multi-dimensional indicators to the point reputation of the equipment, thereby improving the objectivity and accuracy of point reputation evaluation. By using a directed weighted network of point reputation to calculate the credibility of the equipment, it can effectively detect single-point attacks. By using a knowledge graph to calculate the similarity between behaviors and semantics of the devices, it can effectively detect collusion attacks, ensuring the accuracy of the global reputation of the equipment. This invention solves the problems of existing related technologies being not objective enough, having low accuracy in reputation evaluation under single-point attack and collusion attack scenarios, and having limited application scenarios. Attached Figure Description

[0058] Figure 1 This is a system model diagram.

[0059] Figure 2 The flowchart illustrates the cloud-based intelligent inspection equipment reputation assessment method for pumped storage power stations provided by this invention.

[0060] Figure 3 For point reputation directed weighted graph;

[0061] Figure 4 The figures show the performance of the method of this invention and the comparison method under different proportions of malicious devices in a single-point attack scenario; in the figures, (a) is a comparison of the mean absolute error (MAE) of the global reputation assessment results; (b) is a comparison of the mean squared error (MSE) of the global reputation assessment results; (c) is a comparison of the precision of the global reputation assessment results; and (d) is a comparison of the recall of the global reputation assessment results.

[0062] Figure 5 The figures show the performance of the method of this invention and the comparison method under different numbers of devices in a single-point attack scenario; in the figures, (a) is the MAE comparison chart; (b) is the MSE comparison chart; (c) is the Precision comparison chart; and (d) is the Recall comparison chart.

[0063] Figure 6 The figures show the performance of the method of this invention and the comparative method in a collusion attack scenario; in the figures, (a) is the MAE comparison figure; (b) is the MSE comparison figure; (c) is the Precision comparison figure; and (d) is the Recall comparison figure. Detailed Implementation

[0064] The present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

[0065] Figure 1 As shown in the system model diagram, to obtain sufficient information for analyzing the status of energy facilities and promptly detecting anomalies, various intelligent inspection devices located at the equipment layer interact with each other to provide services, share the observed information data of pumped storage power stations, and upload the collected multi-dimensional evaluation indicators to the cloud platform for complex reputation calculations. This invention deploys the designed method on a cloud platform.

[0066] like Figure 2 As shown, a cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations includes the following steps:

[0067] S1. The cloud platform collects reputation evaluation indicators for each intelligent inspection equipment point and uses a BP neural network to calculate the reputation of each equipment point.

[0068] S2. Based on the point reputation of each device, construct a point reputation directed weighted network model, calculate the credibility of each device, and determine whether it is a single point attacker;

[0069] If the judgment result is yes, the point reputation of this device will not be included in the global reputation calculation of other devices;

[0070] Otherwise, participate in the global reputation calculation;

[0071] S3. Based on device entity attributes, use knowledge graph to calculate the behavior-semantic fusion similarity between devices and determine whether they are colluding attackers.

[0072] When the judgment result is yes, the point reputation of the colluding equipment assessment will not participate in the global reputation calculation of other equipment;

[0073] Otherwise, it will participate in the global reputation calculation.

[0074] Steps S1, S2, and S3 are executed sequentially;

[0075] Step S1, specifically, includes:

[0076] S101: The device receives data sent by other devices (i.e., service provider devices); it transforms the interaction status with the service provider devices and the quality of services provided into reputation evaluation indicators for the service provider devices, and uploads them to the cloud platform for reputation evaluation;

[0077] To objectively and accurately assess equipment reputation, this invention proposes a multi-dimensional reputation evaluation index system for intelligent inspection equipment in pumped storage power stations, including inherent attribute indicators and performance attribute indicators. Inherent attribute indicators refer to the relevance of the services provided by the transmitting equipment, including standardization, protocol compliance, and descriptiveness; performance attribute indicators refer to the characteristics exhibited by the equipment during interaction, including response time, throughput, latency, connection success rate, availability, and reliability. Specific definitions are shown in Table 1 below:

[0078] Table 1

[0079] Credit assessment indicators definition Standardization The degree to which the service description language documentation conforms to the service description language specification. Protocol Degree The extent to which the service conforms to the Network Service Protocol Summary Descriptive Measurement of Service Description Language documents Response time Time span from when a node sends an interaction request to when it receives a response Throughput Maximum number of requests processed per unit time Delay Time required to process a given request Connection success rate The ratio of response messages to request messages Availability The ratio of successful calls to total calls reliability The ratio of correct messages to total messages

[0080] S102: Generate dataset. On the cloud platform, the collected device reputation evaluation metrics are standardized into sample data, divided into training and test sets;

[0081] S103: Backpropagation Neural Network Design and Implementation. Design a reasonably structured backpropagation neural network model and train it using a training set. Design details are as follows:

[0082] The input to the BP neural network model is the nine multidimensional evaluation indicators defined in Table 1 for the intelligent inspection equipment, and the output is the one-point reputation of the equipment.

[0083] The BP neural network model is configured with a 3-layer structure, including one input layer, one hidden layer, and one output layer; the relevant parameter configurations of the BP neural network of this invention are shown in Table 2:

[0084] Table 2

[0085]

[0086]

[0087] The BP neural network model was tested using a test set; the trained BP neural network model was then deployed on the cloud platform of the pumped storage power station to objectively and reliably evaluate the point reputation of each intelligent inspection device.

[0088] Step S2, specifically, includes:

[0089] S201: As Figure 3 As shown, a directed weighted network of point reputation is constructed. The directed weighted network of point reputation is denoted by G = (U, E, R);

[0090] Among them, U={u1,u2,...,u n} represents the set of intelligent inspection device nodes in the Internet of Things (IoT) of a pumped storage power station, E = {e1, e2, ..., e m} represents the set of point reputation evaluation relationships among intelligent inspection devices, (u i ,u j )∈E represents the device u i Pointer to device u j Directed weighted edges, r ij This represents the weight of the directed weighted edge, i.e., the weight of the device u. i For equipment u j Evaluation of point reputation;

[0091] S202: Calculate device trustworthiness. Calculate the trustworthiness of each intelligent inspection device based on a point-reputation directed weighted network;

[0092] The formula for calculating credibility is as follows:

[0093]

[0094] Among them, cr i Indicates device u i Credibility, Indicates the device u i The evaluated set of devices, r ij Indicates device u i Evaluation equipment u j Points of reputation, Indicates device u j Average point reputation;

[0095] The formula for calculating average point credit is as follows:

[0096]

[0097] in, Indicates the evaluation equipment u j A collection of devices with a reputation for accuracy, r kj Indicates device u k Evaluation equipment u j Points of reputation;

[0098] S203: Single-point attacker detection. The criteria for single-point attacker detection are:

[0099]

[0100] Where n is the number of nodes in the point reputation directed weighted network. If the credibility of a node is less than μ, the smart inspection device corresponding to that node is judged as a single point attacker.

[0101] S204: Global Reputation Calculation. The formula for global reputation calculation is as follows:

[0102]

[0103] Among them, Rj Indicates device u j Global reputation; For device u i The confidence level after normalization is calculated using the following formula:

[0104]

[0105] The global reputation calculation formula does not include point reputation evaluated by a single point attacker;

[0106] Step S3, specifically, includes:

[0107] S301: Knowledge Graph Attribute Settings. Considering the characteristics of malicious intelligent inspection devices colluding to attack, this invention selects seven basic attributes of the device as entity attributes for constructing the knowledge graph, as shown in Table 3.

[0108] Table 3

[0109] Attribute Name type Identification Label Data attributes Global Reputation Data attributes Credibility Data attributes Number of reviews Data attributes Trusted Object object properties Untrusted objects object properties Communication object object properties

[0110] S302: Knowledge Graph Construction. The knowledge graph was constructed using the Protégé tool and stored in the Neo4j database.

[0111] To alleviate the data sparsity problem and reduce computational complexity, this invention uses the TransE algorithm to map entities and relationships between entities in a knowledge graph to a d-dimensional low-dimensional vector space.

[0112] S303: Behavior-Semantic Fusion Similarity Calculation. The formula for calculating behavior-semantic fusion similarity between devices is as follows:

[0113] fs ij =α·ks ij +(1-α)·bs ij (6)

[0114] Among them, fs ij bs ij ks ij Representing device u respectively i and equipment u j The similarity between them is calculated based on fusion similarity, behavioral similarity, and semantic similarity, with α representing the weight coefficient.

[0115] The formula for calculating behavioral similarity is as follows:

[0116]

[0117] Where n represents the total number of intelligent inspection devices in the pumped storage power station, and r ik and r jk Representing device u respectively i and equipment uj The equipment being evaluated k Points of credibility.

[0118] The formula for calculating semantic similarity is as follows:

[0119]

[0120] Where d represents the entity and relation represent the dimension of the vector space, l ki and l kj Representing device u respectively i and equipment u j The value of the corresponding vector in the k-th dimension.

[0121] S304: Collusive Attacker Judgment. The criteria for judging collusive attackers are as follows: Based on the actual situation of the intelligent inspection equipment network in the pumped storage power station and the scale of the collusive attack, a fusion similarity threshold ε and a quantity threshold δ are set. When the fusion similarity between devices is greater than the threshold ε, the two devices are marked as suspicious collusive devices.

[0122] When the number of suspicious devices with similar fusion scores exceeds the threshold δ, these are identified as colluding attackers.

[0123] S205: Global Reputation Update. In the global reputation calculation in steps 2-4, the point reputation of the colluding attackers is removed.

[0124] The credit assessment method of this invention is abbreviated as CDI-NPRE;

[0125] In this embodiment, the comparison method used includes:

[0126] 1) NPRE: It only has the ability to identify single-point attacks, but not the ability to identify collusive attacks;

[0127] 2) RM: Average Point Reputation as Global Reputation;

[0128] 3) TE_1: Corrects global reputation by assigning lower weights to extreme evaluations;

[0129] 4) TE_2: Corrects global reputation by assigning high weights to normal assessments;

[0130] 5) BS-RE: Collusive attacker identification only considers the similarity of behavior between devices.

[0131] A cloud-based intelligent inspection equipment reputation assessment device for pumped storage power stations, comprising:

[0132] Cloud platform module: Used to collect reputation evaluation indicators of various intelligent inspection devices and perform data processing tasks with high computing resource requirements;

[0133] Point reputation assessment module: Based on the reputation assessment indicators of each intelligent inspection device, a BP neural network is used to calculate the point reputation of each device.

[0134] Single-point attacker identification module: Based on the point reputation of each device, construct a directed weighted network model of point reputation, calculate the credibility of each device, and determine whether it is a single-point attacker;

[0135] If the judgment result is yes, the point reputation of this device will not be included in the global reputation calculation of other devices;

[0136] Otherwise, participate in the global reputation calculation;

[0137] Collusive attacker identification module: Based on device entity attributes and using knowledge graph, calculate the behavior-semantic fusion similarity between devices to determine whether they are collusive attackers;

[0138] When the judgment result is yes, the point reputation of the colluding equipment assessment will not participate in the global reputation calculation of other equipment;

[0139] Otherwise, participate in the global reputation calculation;

[0140] Global Reputation Assessment Module: Calculates the global reputation of each intelligent inspection device based on its point reputation and normalized credibility.

[0141] Figure 4 The graph shows the performance of the present invention and the comparative method under different proportions of malicious devices in a single-point attack scenario. Figure 4 In the figures, (a) shows a comparison of the mean absolute error (MAE) of the global reputation assessment results; (b) shows a comparison of the mean squared error (MSE) of the global reputation assessment results; (c) shows a comparison of the precision of the global reputation assessment results; and (d) shows a comparison of the recall of the global reputation assessment results. The horizontal axis of the four figures represents the ratio of the number of malicious intelligent inspection devices to the total number of intelligent inspection devices in the pumped storage power station. The vertical axis of the four figures represents MAE, MSE, Precision, and Recall, respectively. The four broken lines represent the method of this invention, the comparative method RM, the comparative method TE_1, and the comparative method TE_2, respectively. It can be seen that, for single-point attack scenarios, under different proportions of malicious devices, the method of this invention has higher accuracy in assessing the global reputation of the devices and smaller error compared with the true global reputation of the devices. Furthermore, as the proportion of malicious devices increases, the accuracy performance advantage of the method of this invention compared with the comparative method becomes more obvious.

[0142] Figure 5 The graph shows the performance of the present invention and the comparative method under different numbers of devices in a single-point attack scenario. Figure 5In the diagram, (a) shows the MAE comparison chart; (b) shows the MSE comparison chart; (c) shows the Precision comparison chart; and (d) shows the Recall comparison chart. The horizontal axis of the four charts represents the number of intelligent inspection devices in the pumped storage power station. The vertical axis of the four charts represents MAE, MSE, Precision, and Recall, respectively. The four broken lines represent the method of this invention, the comparison method RM, the comparison method TE_1, and the comparison method TE_2, respectively. It can be seen that, for single-point attack scenarios, under different numbers of devices, the method of this invention has higher accuracy in evaluating the global reputation of the devices, and the error between it and the actual global reputation of the devices is smaller. Furthermore, as the number of devices increases, the accuracy performance of the comparison method decreases significantly, while the accuracy performance of the method of this invention remains at a high level.

[0143] Figure 6 Performance graphs of the present invention and the comparative method in a collusive attack scenario; Figure 6 In the diagram, (a) shows the MAE comparison chart; (b) shows the MSE comparison chart; (c) shows the Precision comparison chart; and (d) shows the Recall comparison chart. The horizontal axis of the four charts represents the ratio of the number of intelligent inspection devices participating in the collusion attack to the total number of intelligent inspection devices in the pumped storage power station. The vertical axis of the four charts represents MAE, MSE, Precision, and Recall, respectively. The three broken lines represent the method of this invention, the comparative method BS-RE, and the comparative method NPRE, respectively. It can be seen that, for collusion attack scenarios, under different proportions of colluding devices, the method of this invention has higher accuracy in evaluating the global reputation of devices, smaller error compared to the true global reputation of devices, and as the proportion of malicious devices increases, the accuracy performance advantage of the method of this invention becomes more obvious compared to the comparative method, while the recall performance remains at a high level.

[0144] The performance metrics used in this invention include the mean absolute error (MAE), mean squared error (MSE), precision, and recall of the global reputation assessment results, calculated using the following formulas:

[0145]

[0146]

[0147]

[0148]

[0149] Where n is the total number of intelligent inspection devices in the pumped storage power station, and R i It is device u i True overall reputation The equipment u is evaluated by various methods iGlobal reputation. MAE can intuitively reflect the magnitude of global reputation error, while MSE can intuitively reflect changes in global reputation.

[0150] The above specific embodiments are used to explain and illustrate the present invention, and are only preferred embodiments of the present invention, not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A cloud-based credibility evaluation method for intelligent inspection equipment of pumped storage power stations, characterized in that: The method includes the following steps: S1. The cloud platform collects reputation evaluation indicators for each intelligent inspection equipment point and uses a BP neural network to calculate the reputation of each equipment point. S2. Based on the point reputation of each device, construct a point reputation directed weighted network model, calculate the credibility of each device, and determine whether it is a single point attacker; If the judgment result is yes, the point reputation of this device will not be included in the global reputation calculation of other devices; Otherwise, participate in the global reputation calculation; S3. Based on device entity attributes, use knowledge graph to calculate the behavior-semantic fusion similarity between devices and determine whether they are colluding attackers. When the judgment result is yes, the point reputation of the colluding equipment assessment will not participate in the global reputation calculation of other equipment; Otherwise, participate in the global reputation calculation; In step S2, the credibility of each intelligent inspection device is calculated based on a point reputation directed weighted network; Credibility The calculation formula is as follows: （1） in, Indicates equipment Credibility, Indicates the device The evaluated equipment set, Indicates equipment Evaluation equipment Points of reputation, Indicates equipment Average point reputation; Average Points Reputation The calculation formula is as follows: （2） in, Indicates evaluation equipment A collection of devices that provide credit rating. r kj Indicates equipment Evaluation equipment Points of reputation; The criteria for determining a single point of attack are: （3） in, In a directed weighted network of point reputation, the number of nodes is determined by the degree of trust of a node. If so, the intelligent inspection device corresponding to that node is identified as a single-point attacker; In steps S2 and S3, the global reputation calculation formula is as follows: （4） in, Indicates equipment Global reputation Indicates evaluation equipment A collection of devices that provide credit rating. For equipment Credibility after normalization Indicates equipment Evaluation equipment Points of reputation; In step S3, specifically, on the cloud platform, a knowledge graph is generated based on the entity attributes of the intelligent inspection equipment to obtain a vectorized representation of entities and relationships; The formula for calculating the similarity between device behaviors and semantics is as follows: （6） in, , , Representing the equipment and equipment The similarity between them includes fusion similarity, behavioral similarity, and semantic similarity. Indicates the weighting coefficient; The criteria for identifying collusive attackers are: setting a fusion similarity threshold based on the actual networking of intelligent inspection equipment in pumped storage power stations and the scale of the collusive attack. and quantity threshold When the fusion similarity between devices is greater than a threshold The two devices were flagged as suspected collusion devices; When the number of suspicious devices with similar fusion properties exceeds a threshold These individuals are then identified as colluding attackers. semantic similarity The calculation formula is as follows: （8） in, Entities and relations represent the dimensions of a vector space. and Representing the equipment and equipment The corresponding vector is at the th The value that can be taken in the dimension.

2. The cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations according to claim 1, characterized in that: In step S1, the credit evaluation indicators for intelligent inspection equipment points include: inherent attribute indicators and performance attribute indicators; The inherent attribute indicators are: standardization, protocol, and descriptiveness; The performance attributes are: response time, throughput, latency, connection success rate, availability, and reliability.

3. The cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations according to claim 1, characterized in that: In step S1, specifically, on the cloud platform, the point reputation evaluation index of each intelligent inspection device is input into the BP neural network to obtain the point reputation of each intelligent inspection device.

4. The cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations according to claim 1, characterized in that: In step S2, specifically, on the cloud platform, a directed weighted network of point reputation is generated based on the point reputation assessment relationship between intelligent inspection devices and the point reputation obtained from the assessment. In the directed weighted network of point reputation, each node represents a smart inspection device in a pumped storage power station. The directed edges between nodes represent the point reputation assessment relationship between the smart inspection devices, pointing from the point reputation assessor to the assessed device. The weight of the directed edge represents the point reputation being assessed.

5. The cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations according to claim 1, characterized in that: In step S3, the device entity attributes include: identification tag, global reputation, trustworthiness, number of evaluations, trusted object, untrusted object, and communication object.

6. The cloud-based intelligent inspection equipment reputation evaluation method for pumped storage power stations according to claim 1, characterized in that: Behavioral similarity The calculation formula is as follows: （7） in, This indicates the total number of intelligent inspection devices in a pumped storage power station. and Representing the equipment and equipment Equipment being evaluated Points of credibility.

7. A cloud-based intelligent inspection equipment reputation evaluation device for pumped storage power stations, characterized in that: The device is based on the cloud-based intelligent inspection equipment reputation assessment method for pumped storage power stations as described in claim 1, and includes: The cloud platform module is used to collect reputation evaluation indicators of various intelligent inspection devices and perform data processing tasks with high computing resource requirements. The point reputation assessment module calculates the point reputation of each device based on the reputation assessment indicators of each intelligent inspection device and using a BP neural network. The single-point attacker identification module constructs a directed weighted network model of point reputation based on the point reputation of each device, calculates the credibility of each device, and determines whether it is a single-point attacker. If the determination result is yes, the point reputation evaluated by the device does not participate in the global reputation calculation of other devices; otherwise, it participates in the global reputation calculation. The collusion attacker identification module uses a knowledge graph based on device entity attributes to calculate the behavior-semantic fusion similarity between devices and determine whether they are collusion attackers. If the identification result is yes, the point reputation of the collusion device is not included in the global reputation calculation of other devices; otherwise, it is included in the global reputation calculation. The global reputation assessment module calculates the global reputation of each intelligent inspection device based on its point reputation and normalized credibility.

Images