Multi-dimensional dynamic trust evaluation method and system for new energy power terminal

Abstract

The invention discloses a multi-dimensional dynamic trust evaluation method and system for a new energy power terminal, and the method comprises the steps: extracting the universal features of the terminal based on terminal log interaction data, extracting special knowledge evidence according to an equipment level perception layer and a network layer, and converting all features into 0-1 standardized values; when a terminal accesses for the first time, the static attribute of the terminal is complementarily collected, so that an initial trust baseline is established and an initial trust value is given; firstly, an IoT device interaction diagram is constructed, nodes are terminal devices, attributes are knowledge evidence vectors, edges are device direct interaction relations, and weights of the edges are defined by historical interaction quality; fusing the equipment knowledge evidence and the interaction relationship through a node2vec algorithm to generate a low-dimensional embedded vector; and finally, calculating the cosine similarity of the device vectors, multiplying the cosine similarity by the interaction frequency proportion to obtain the direct trust degree of the 0-1 interval, and quantifying the direct trust relationship between the devices. Performing trust evaluation on the message flow dimension; and historical interaction behavior trust evaluation and timeliness optimization.

Description

Technical Field

[0001] This invention relates to a multi-dimensional dynamic trust assessment method and system for new energy power terminals, belonging to the field of network security technology. The method integrates key technologies such as zero-trust security architecture, terminal behavior analysis, and dynamic risk assessment, and particularly relates to a method for continuous trust assessment and malicious behavior detection of terminal nodes. Background Technology

[0002] With the rapid development of the new energy power industry, the application of new energy power terminals (such as photovoltaic inverters, energy storage controllers, and various sensors) in power systems is becoming increasingly widespread and important. As key nodes in power production, transmission, distribution, and management, the safe and reliable operation of these terminals directly affects the stability and security of the new energy power system. However, current new energy power terminals face numerous security challenges. On the one hand, new energy power systems exhibit a high degree of Internet of Things (IoT) characteristics, with a large number of widely distributed terminal devices and frequent and complex interactions between them. Traditional trust assessment methods based solely on single-dimensional identity authentication and simple communication monitoring are no longer sufficient to comprehensively and accurately assess the trust status of terminals. For example, traditional identity authentication methods such as digital certificates and keys cannot effectively identify abnormal behaviors caused by attacks or malfunctions during terminal operation, such as communication anomalies, unauthorized commands, and firmware tampering. On the other hand, the trust status of terminals is dynamic, changing with operating time, interaction history, and the environment. Static trust assessment methods cannot reflect the current trust level of terminals in real time, potentially leading to untimely risk warnings and subsequently causing power system failures or even safety accidents. Meanwhile, in new energy power systems, newly connected terminal equipment lacks historical interaction data, making it difficult to effectively assess using traditional direct trust calculation methods, resulting in a gap in trust assessment. Furthermore, new attack methods targeting new energy power terminals are constantly emerging, such as firmware tampering attacks and network traffic spoofing attacks. Existing trust assessment technologies are insufficient in identifying these new attacks, failing to guarantee the system's security protection requirements. Therefore, a multi-dimensional, dynamic, accurate, and comprehensive trust assessment method for new energy power terminals is needed to address the problems of existing technologies, such as limited trust assessment dimensions, insufficient dynamism, difficulty in assessing new terminals, and weak ability to identify new attacks, thereby providing strong support for the safe and stable operation of new energy power systems. Summary of the Invention

[0003] This invention provides a multi-dimensional dynamic trust assessment method for new energy power terminals. Addressing the "one-time authentication, always trust" problem caused by the lack of trust management technology for various terminal access behaviors in existing new energy systems, this invention addresses the issue of single-dimensional assessment in traditional methods. It extracts multi-dimensional knowledge evidence (such as the accuracy of perception layer data, network layer command response latency, and the number of communication anomalies at the behavior layer) from basic terminal attributes, historical behavior, and message traffic to achieve a comprehensive trust assessment, avoiding misjudgments based on a single dimension. It also solves the problem of evaluation gaps due to the lack of historical interaction data for newly connected unfamiliar terminals. By collecting static attributes and combining them with traditional identity authentication to establish an initial trust baseline, and leveraging the node2vec framework, it addresses the difficulty in establishing interaction relationships between IoT terminals. To address the modeling issue, node2vec is used to construct a terminal interaction graph, fusing high-dimensional knowledge evidence and interaction relationships into low-dimensional vectors to accurately capture device-associated trust features. Neighbor node feedback is used to indirectly assess the trust level of new terminals. Overcoming the limitations of insufficient dynamics and timeliness in trust status, real-time logs and traffic data are used, combined with a 24-hour sliding window to filter recent interaction records and time decay factors, to achieve real-time updates of trust values ​​and immediate triggering of risk management. Simultaneously, trust levels are categorized, quantifying direct trust, traffic trust, and behavioral trust values ​​into standard values ​​to calculate a comprehensive trust value. Three threshold levels—"high (≥80 points), medium (60-79 points), and low (<60 points)"—and corresponding handling rules are clearly defined, providing a clear basis for operation and maintenance.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] A multi-dimensional dynamic trust assessment method for new energy power terminals includes the following steps:

[0006] Step 1: Multi-dimensional data processing and standardization:

[0007] Based on terminal log interaction data, we extract the common features of the terminal, and at the same time extract specific knowledge evidence according to the device level perception layer and network layer. We convert all features into 0-1 standardized values ​​to provide a unified data foundation for subsequent calculations.

[0008] Step 2: Establishing the initial trust baseline for new access terminals:

[0009] When a terminal first connects, in addition to traditional digital certificates and key verification for identity authentication, its static attributes (hardware characteristics, configuration information, environmental parameters) are collected to establish an initial trust baseline and assign an initial trust value, thus solving the problem of trust assessment starting point when a new terminal has no historical data.

[0010] Step 3: Calculate direct trust based on node2vec:

[0011] First, an IoT device interaction graph is constructed, where nodes represent terminal devices, attributes are knowledge evidence vectors, and edges represent direct interaction relationships between devices. The weight of each edge is defined by the quality of historical interactions. Then, the node2vec algorithm is used to fuse device knowledge evidence and interaction relationships to generate low-dimensional embedding vectors. Finally, the cosine similarity (knowledge evidence matching degree) of the device vectors is calculated and multiplied by the interaction frequency ratio (number of interactions between device A and B / total number of interactions between A) to obtain the direct trust degree in the 0-1 range, thus quantifying the direct trust relationship between devices.

[0012] Step 4: Trust assessment at the message traffic level:

[0013] Analyze and store the number of packets sent (PS), received (PR), packet loss rate (PDR), and transmission / reception time (TS / TR) of terminal data packets. By fusing the formula of packet ratio (PR / PS), packet loss rate, and time difference (TR-TS), calculate the terminal trust value in this dimension to supplement the trust judgment basis at the traffic level.

[0014] Calculate the trust score using the following formula:

[0015]

[0016] Where PS represents the number of data packets sent, PR represents the number of data packets received (dividing them gives the packet ratio), PDR represents the packet loss rate, TR represents the time to receive data packets, and TS represents the time to send data packets. Delay is the data packet delay, Delay_thresh is the delay threshold, and Jitter is the delay jitter, Jitter_thresh is the jitter threshold. , , , Assigning weights to each indicator. + + + =1;

[0017] Step 5: Trust assessment and timeliness optimization of historical interaction behavior:

[0018] Based on the number of successful / failed interactions in history, the expected trust value is calculated using a Beta distribution. A 24-hour sliding window is used to filter recent interaction records, and a time decay factor is introduced (the larger the K value, the faster the old data decays) to ensure that the trust assessment focuses on recent behavior and improves the timeliness and accuracy of the assessment.

[0019] Step 6: Set multi-level trust thresholds:

[0020] Set multi-level trust thresholds; comprehensively process the direct trust level, terminal trust value, and trust value expectation of the terminal, and determine the trust level of the terminal based on the set thresholds and update it in real time.

[0021] This application presents a multi-dimensional dynamic trust assessment method for new energy power terminals, which is built around the core of "multi-dimensional data support - initial trust foundation - hierarchical trust calculation - timeliness optimization".

[0022] In step 1 above, the general characteristics of the terminal include: terminal type, number of communication anomalies, frequency of unauthorized commands, firmware tampering flags, hardware fault codes and deviation rate of neighboring node behavior, etc.

[0023] In step 1 above, the perception layer extracts specific knowledge evidence including: the device's data accuracy, data integrity, online stability, and time synchronization; the network layer extracts specific knowledge evidence including: the device's command response latency, command forwarding success rate, and bandwidth utilization.

[0024] Step 3 above includes the following steps:

[0025] Step 1: Construct the IoT device interaction graph G = (V, E, W):

[0026] Node V: Each IoT device is a node, and the node attribute is the "knowledge evidence" of that device. Construct a knowledge evidence vector about that node;

[0027] Edge E: If device A and device B have direct interaction, then edge E (A,B) exists; if there is no interaction, then there is no edge.

[0028] Edge weight W: Defined using the "historical interaction quality" of device A and device B, a higher weight indicates better interaction quality;

[0029] Step 2: Transform into a low-dimensional vector based on node2vec:

[0030] In IoT scenarios, node2vec is used to learn low-dimensional embedding vectors of nodes (terminal devices). It integrates knowledge evidence such as device attributes with the relationship between interactive behavior and transforms them into low-dimensional vectors.

[0031] Step 3: Calculate the direct trust level of the device (quantify the trust relationship).

[0032] After obtaining the low-dimensional vectors of the devices, a calculation model is designed based on the real-time performance and security weights of the terminal equipment in the new energy power plant. The cosine similarity between the vector Vec_A of device A and the vector Vec_B of device B is calculated, and then multiplied by the interaction frequency weight to avoid misjudgment of low-frequency interactions. That is:

[0033]

[0034] Direct trust is the product of cosine similarity (knowledge evidence matching degree) and the proportion of interaction frequency:

[0035]

[0036] Let A be the number of interactions between A and B. The total number of interactions between device A and device B is DT. The value of DT ranges from 0 to 1. The closer the value is to 1, the higher the trust level of device A in device B. and These represent the i-th component in the low-dimensional vectors of device A and device B, respectively.

[0037] Add to this formula Dynamic feature weights are used to achieve higher dynamic realism. Let A be the time interval between the most recent interaction between A and B. Here, k represents the time decay coefficient, enhancing the timeliness of interaction trust. The final formula is:

[0038] .

[0039] Step 5 above includes:

[0040] Trust assessment based on interaction behavior considers that the number of successful and failed interactions in the past influences the next interaction. The expected trust value is calculated using a Beta distribution based on the historical number of successful and failed interactions.

[0041]

[0042] To ensure timeliness, a sliding window selection process is used to select historical interaction records within 24 hours for trust value calculation. A time decay factor is also introduced.

[0043]

[0044] Where m represents the number of successful interactions within the sliding window, n represents the total number of interactions, K is a constant (the larger the K value, the faster the historical data decays), s represents the number of successful interactions in the past, f represents the number of failed interactions in the past, AC is the trust score calculation result, and Exp represents the expected value calculation. This represents the number of successful interactions within the time window. This represents the number of failed interactions within the time window. This represents the i-th time window.

[0045] In step 6 above, a three-level trust threshold is set:

[0046] The above trust values ​​are processed comprehensively:

[0047]

[0048] Where DT represents direct trust level; T represents terminal trust value; and AC represents expected trust value.

[0049] S≥80 indicates high trust, terminal is operating normally, no intervention;

[0050] When S is 60-79, it indicates medium trust, triggering an alert and increasing monitoring frequency;

[0051] If S<60, it is considered a low trust level and is identified as a potential malicious terminal. Measures such as restricting permissions, disconnecting the connection, and starting the isolation mechanism will be implemented.

[0052] Terminal trust levels are categorized, and the trust values ​​of terminals are updated in real time for the next trust assessment.

[0053] A multi-dimensional dynamic trust assessment system for new energy power terminals includes:

[0054] Multi-dimensional data processing and standardization module: Based on terminal log interaction data, it extracts the general features of the terminal, and at the same time extracts specific knowledge evidence according to the device-level perception layer and network layer, and transforms all features into 0-1 standardized values ​​to provide a unified data foundation for subsequent calculations;

[0055] The initial trust baseline construction module for new access terminals is used to collect static attributes (hardware characteristics, configuration information, and environmental parameters) on the basis of traditional digital certificates and key verification authentication when a terminal accesses the network for the first time. This establishes an initial trust baseline and assigns an initial trust value, solving the problem of the starting point for trust assessment when a new terminal has no historical data.

[0056] The direct trust calculation module based on node2vec is used to first construct an IoT device interaction graph, where nodes are terminal devices, attributes are knowledge evidence vectors, and edges represent direct interaction relationships between devices. The weight of each edge is defined by the quality of historical interactions. Then, the node2vec algorithm is used to fuse device knowledge evidence and interaction relationships to generate low-dimensional embedding vectors. Finally, the cosine similarity (knowledge evidence matching degree) of the device vectors is calculated and multiplied by the interaction frequency ratio (number of interactions between device A and B / total number of interactions between A) to obtain the direct trust degree in the 0-1 range, quantifying the direct trust relationship between devices.

[0057] The message traffic dimension trust assessment module is used to analyze and store the number of packets sent (PS), received (PR), packet loss rate (PDR), and transmission / reception time (TS / TR) of terminal data packets. By fusing the packet ratio (PR / PS), packet loss rate, and time difference (TR-TS) formula, it calculates the terminal trust value under this dimension, supplementing the trust judgment basis at the traffic level.

[0058] Calculate the trust score using the following formula:

[0059]

[0060] Where PS represents the number of data packets sent, PR represents the number of data packets received (dividing them gives the packet ratio), PDR represents the packet loss rate, TR represents the time to receive data packets, and TS represents the time to send data packets. Delay is the data packet delay, Delay_thresh is the delay threshold, and Jitter is the delay jitter, Jitter_thresh is the jitter threshold. , , , Assigning weights to each indicator. + + + =1;

[0061] The historical interaction behavior trust assessment and timeliness optimization module is used to calculate the expected trust value based on the number of successful / failed interactions in history using the Beta distribution; it uses a 24-hour sliding window to filter recent interaction records and introduces a time decay factor (the larger the K value, the faster the old data decays) to ensure that the trust assessment focuses on recent behavior and improves the timeliness and accuracy of the assessment.

[0062] Multi-level trust threshold setting module: used to set multi-level trust thresholds; comprehensively process the direct trust level of the terminal, the terminal trust value, and the expected trust value, and determine the trust level of the terminal based on the set thresholds and update it in real time.

[0063] The common characteristics of the aforementioned terminals include: terminal type, number of communication anomalies, frequency of unauthorized commands, firmware tampering flags, hardware fault codes and deviation rate of neighboring node behavior, etc.

[0064] The aforementioned perception layer extracts specific knowledge evidence including: device data accuracy, data integrity, online stability, and time synchronization; the network layer extracts specific knowledge evidence including: device command response latency, command forwarding success rate, and bandwidth utilization.

[0065] The aforementioned direct trust calculation module based on node2vec includes:

[0066] Construct an IoT device interaction graph G = (V, E, W) unit:

[0067] Node V: Each IoT device is a node, and the node attribute is the "knowledge evidence" of that device. Construct a knowledge evidence vector about that node;

[0068] Edge E: If device A and device B have direct interaction, then edge E (A,B) exists; if there is no interaction, then there is no edge.

[0069] Edge weight W: Defined using the "historical interaction quality" of device A and device B, a higher weight indicates better interaction quality;

[0070] Transform low-dimensional vector units based on node2vec:

[0071] In IoT scenarios, node2vec is used to learn low-dimensional embedding vectors of nodes (terminal devices). It integrates knowledge evidence such as device attributes with the relationship between interactive behavior and transforms them into low-dimensional vectors.

[0072] The unit of direct trust in computing devices (quantifying trust relationships):

[0073] After obtaining the low-dimensional vectors of the devices, a calculation model is designed based on the real-time performance and security weights of the terminal equipment in the new energy power plant. The cosine similarity between the vector Vec_A of device A and the vector Vec_B of device B is calculated, and then multiplied by the interaction frequency weight to avoid misjudgment of low-frequency interactions. That is:

[0074]

[0075] Direct trust is the product of cosine similarity (knowledge evidence matching degree) and the proportion of interaction frequency:

[0076]

[0077] Let A be the number of interactions between A and B. The total number of interactions between device A and device B is DT. The value of DT ranges from 0 to 1. The closer the value is to 1, the higher the trust level of device A in device B. and These represent the i-th component in the low-dimensional vectors of device A and device B, respectively.

[0078] Add to this formula Dynamic feature weights are used to achieve higher dynamic realism. Let A be the time interval between the most recent interaction between A and B. Here, k represents the time decay coefficient, enhancing the timeliness of interaction trust. The final formula is:

[0079] .

[0080] Any techniques not mentioned in this invention are based on existing technologies.

[0081] The present invention has the following beneficial effects:

[0082] 1) Effectively solves the problems of single traditional assessment dimensions, lack of assessment for new terminals, lagging trust status updates, weak identification of new threats, and low operation and maintenance efficiency. By integrating multi-dimensional features (static attributes, dynamic behavior, layered knowledge evidence, traffic data) with the node2vec algorithm, it achieves a three-dimensional and accurate assessment of terminal trust, avoiding misjudgment based on a single dimension.

[0083] 2) When new terminals are connected, the initial trust baseline is used to fill the evaluation gap where there is no historical data. At the same time, the trust relationship between IoT devices is captured through interaction graph modeling to adapt to the system expansion requirements.

[0084] 3) By leveraging real-time data, a 24-hour sliding window, and a time decay factor, the trust value is dynamically updated. Combined with three-level thresholds, it can trigger early warnings or isolation in real time, thereby improving the speed of anomaly response.

[0085] 4) By iteratively optimizing the algorithm using the Bayesian model, the ability to identify new types of attacks is enhanced, and system security protection is strengthened; in addition, the clear trust classification and automated handling mechanism reduces manual intervention, improves operation and maintenance efficiency, and ultimately provides a reliable guarantee for the safe and stable operation of the new energy power system terminal. Attached Figure Description

[0086] Figure 1 This is a flowchart of the multi-dimensional dynamic trust assessment method for new energy power terminals according to the present invention;

[0087] Figure 2 Generate a graph for the terminal vector; Detailed Implementation

[0088] To better understand the present invention, the following embodiments further illustrate the content of the present invention, but the content of the present invention is not limited to the following embodiments.

[0089] This invention proposes a multi-dimensional dynamic trust assessment method for new energy power terminals. It evaluates the terminal's basic attributes and historical behavior from multiple dimensions, and applies a "direct trust + recommended trust" calculation framework (combined with the node2vec algorithm) to the trust assessment of Internet of Things (IoT) devices, enabling real-time updates and risk warnings of terminal trust levels. The method includes the following steps:

[0090] Step 1: Multi-dimensional data processing and standardization:

[0091] Based on existing terminal log interaction data, features are extracted from each sample, including terminal type, number of communication anomalies, frequency of unauthorized commands, firmware tampering markers, hardware fault codes, and deviation rate from neighboring nodes. Simultaneously, specific knowledge evidence is extracted at the device level (perception layer and network layer). This knowledge evidence extraction includes data accuracy, data integrity, online stability, and time synchronization for perception layer devices; and command response latency, command forwarding success rate, and bandwidth utilization for network layer devices. These are then converted into 0-1 standardized values ​​to ensure subsequent computational feasibility.

[0092] Step 2: Establishing the initial trust baseline for new access terminals:

[0093] When a terminal is first connected to the system, in addition to traditional identity authentication such as digital certificates and key verification, its static attributes (hardware characteristics, configuration information and environmental parameters) are collected as an initial trust baseline and an initial trust value is given to solve the problem of trust assessment starting point when a new terminal has no historical data.

[0094] Step 3: Calculate direct trust based on node2vec:

[0095] First, an IoT device interaction graph is constructed, where terminal devices are the "nodes" and the direct interaction relationships between devices are the "edges," with the edge weights defined by the average interaction success rate. Then, the node2vec algorithm is used to fuse device knowledge evidence and interaction relationships to generate low-dimensional embedding vectors. Finally, the cosine similarity (knowledge evidence matching degree) of the device vectors is calculated and multiplied by the interaction frequency ratio (number of interactions between device A and B / total number of interactions between A) to obtain the direct trust degree in the 0-1 range, quantifying the direct trust relationship between devices. The specific operations are as follows:

[0096] Step 1: Construct the IoT device interaction graph (G = (V, E, W))

[0097] V (Node): Each IoT device is treated as a node, and the node attribute is the "knowledge evidence" of that device. A knowledge evidence vector is constructed about that node.

[0098] E (edge): If device A and device B have direct interaction, then there exists an edge E (A,B); if there is no interaction, then there is no edge.

[0099] W (edge ​​weight): Defined by the "historical interaction quality" between the two, such as the "average data accuracy × average integrity" of the data reported by node S1 to node G1 (0.95 × 0.98 = 0.931). The higher the weight, the better the interaction quality.

[0100] Step 2: Use node2vec to learn the low-dimensional embedding vector of the node (device). In IoT scenarios, node2vec is used to integrate the device attributes and other evidence with the relationship of interaction behavior, and transform them into a low-dimensional vector.

[0101] Step 3: Calculate the direct trust level of the device (quantify the trust relationship).

[0102] After obtaining the low-dimensional vectors of the equipment, a calculation model is designed based on the real-time performance and safety weights of the new energy power plant equipment. The cosine similarity between the vector of equipment A (Vec_A) and the vector of equipment B (Vec_B) is calculated, and then multiplied by the interaction frequency weight to avoid misjudgment of low-frequency interactions. That is:

[0103]

[0104] Direct trust is the product of cosine similarity (knowledge evidence matching degree) and the proportion of interaction frequency:

[0105]

[0106] Let A be the number of interactions between A and B. Let DT be the total number of interactions between device A and device B. The value of DT ranges from (0, 1). The closer the value is to 1, the higher the trust level between device A and device B. Based on this formula, add... Dynamic feature weights are used to achieve higher dynamic realism. Let A be the time interval between the most recent interaction between A and B. This is a time decay coefficient, enhancing the timeliness of interactive trust; the final formula is:

[0107]

[0108] Step 4: Evaluation based on terminal packet traffic information:

[0109] Analyze and store the number of packets sent (PS), received (PR), packet loss rate (PDR), and transmission / reception time (TS / TR) of terminal data packets. By fusing the formula of packet ratio (PR / PS), packet loss rate, and time difference (TR-TS), calculate the terminal trust value in this dimension to supplement the trust judgment basis at the traffic level.

[0110] The trust value is calculated using the following formula.

[0111]

[0112] Where PS represents the number of data packets sent, PR represents the number of data packets received (dividing them gives the packet ratio), PDR represents the packet loss rate, TR represents the time to receive data packets, and TS represents the time to send data packets. Delay is the data packet delay, Delay_thresh is the delay threshold, and Jitter is the delay jitter, Jitter_thresh is the jitter threshold. , , , Weights for each indicator ( + + + =1).

[0113] Step 5: Trust assessment and timeliness optimization of historical interaction behavior:

[0114] Based on the number of successful / failed interactions in history, the expected trust value is calculated using a Beta distribution. A 24-hour sliding window is used to filter recent interaction records, and a time decay factor is introduced (the larger the K value, the faster the old data decays) to ensure that the trust assessment focuses on recent behavior and improves the timeliness and accuracy of the assessment.

[0115] Trust assessment based on interaction behavior considers that the number of successful and failed interactions in the past influences the next interaction. The expected trust value is calculated using a Beta distribution based on the number of successful and failed interactions in the past.

[0116] A

[0117] To ensure timeliness, a sliding window selection process is used to select historical interaction records within 24 hours for trust value calculation, and a time decay factor is also introduced.

[0118]

[0119] Where m is the number of successful interactions in the history within the sliding window, n is the number of historical interactions, and K is a constant. The larger the value of K, the faster the historical data decays.

[0120] Step 6: Set the three-level trust threshold:

[0121] The above trust values ​​are processed comprehensively:

[0122]

[0123] High trust (≥80 points): The terminal is operating normally without intervention;

[0124] Zhongxin Trust (60-79 points): Trigger an early warning and increase monitoring frequency;

[0125] Low Trust (<60 points): Determined as a potentially malicious terminal, and measures such as restricting permissions, disconnecting from the terminal, and activating isolation mechanisms are implemented.

[0126] The aforementioned method classifies terminal trust levels and updates terminal trust values ​​in real time for subsequent trust assessments. It effectively addresses issues such as the traditional single-dimensional assessment, gaps in new terminal assessment, delayed trust status updates, weak identification of new threats, and low operational efficiency. By fusing multi-dimensional features (static attributes, dynamic behavior, layered knowledge evidence, and traffic data) with the node2vec algorithm, it achieves a comprehensive and accurate assessment of terminal trust, avoiding misjudgments based on a single dimension. When a new terminal connects, it fills the assessment gaps due to the lack of historical data based on the initial trust baseline. Simultaneously, it captures the trust relationships associated with IoT devices through interaction graph modeling, adapting to system expansion needs. Real-time data-driven approaches, a 24-hour sliding window, and time decay factors enable dynamic updates of trust values. Combined with three-level thresholds, it instantly triggers warnings or isolation, improving anomaly response speed. It enhances the ability to identify new attacks and strengthens system security through Bayesian model iterative optimization algorithms. Furthermore, the clear trust classification and automated handling mechanisms reduce manual intervention, improve operational efficiency, and ultimately provide reliable guarantees for the safe and stable operation of new energy power system terminals.

Claims

1. A multi-dimensional dynamic trust assessment method for new energy power terminals, characterized in that: Comprising the following steps: Step 1: Multi-dimensional data processing and standardization: Based on terminal log interaction data, extract general features of the terminal, and extract special knowledge evidence according to the device level perception layer and network layer, and convert all features into 0-1 standardized numerical values; Step 2: Construction of initial trust baseline for new access terminal: When the terminal accesses for the first time, based on the traditional digital certificate, key verification identity authentication, supplement the collection of its static attributes, establish the initial trust baseline and give the initial trust value, solve the trust evaluation starting point problem when the new terminal has no historical data; Step 3: Direct trust degree calculation based on node2vec: First, build an IoT device interaction graph, nodes are terminal devices, attributes are knowledge evidence vectors, and edges are device direct interaction relationships. The weight of the edge is defined by the historical interaction quality. Then, through the node2vec algorithm, the device knowledge evidence and interaction relationship are fused to generate a low-dimensional embedding vector. Finally, the cosine similarity of the device vector is calculated and multiplied by the interaction frequency ratio to obtain the direct trust degree in the 0-1 interval, quantifying the direct trust relationship between devices; Step 4: Message flow dimension trust evaluation: Analyze and store the number of terminal data packets sent, received, and lost, and the sending / receiving time. Through the formula of fusing packet ratio, packet loss rate, and time difference, the terminal trust value in this dimension is calculated, and the trust judgment basis in the flow layer is supplemented; The trust value is calculated by the following equation: ; Where PS represents the number of data packets sent, PR represents the number of data packets received, and the packet ratio is obtained by dividing, PDR represents the packet loss rate, TR represents the time of receiving data packets, and TS represents the time of sending data packets; Delay is the data packet delay, Delay_thresh is the delay threshold; Jitter is the delay jitter, Jitter_thresh is the jitter threshold; α, β, γ, δ are the weights of each index, α+β+γ+δ=1; Step 5: Historical interaction behavior trust evaluation and timeliness optimization: Based on the number of historical interaction successes / failures, the trust value expectation is calculated using the Beta distribution. A 24-hour sliding window is used to filter recent interaction records, and a time decay factor is introduced to ensure that the trust evaluation focuses on recent behavior, improving the timeliness and accuracy of the evaluation; Step 6: Set multi-level trust threshold: Set multi-level trust threshold; Comprehensively process the direct trust degree of the terminal, the terminal trust value, and the trust value expectation, and according to the set threshold, judge the trust level of the terminal, and update it in real time.

2. The multi-dimensional dynamic trust evaluation method for new energy power terminals according to claim 1, characterized in that: In step 1, the general features of the terminal include: terminal type, communication exception times, instruction overreach frequency, firmware tamper mark, hardware fault code, and neighbor node behavior deviation rate.

3. The multi-dimensional dynamic trust evaluation method for new energy power terminals according to claim 1 or 2, characterized in that: In step 1, the special knowledge evidence extracted from the perception layer includes: data accuracy, data integrity, online stability, and time synchronization of the device; the special knowledge evidence from the network layer includes: instruction response delay, instruction forwarding success rate, and bandwidth utilization rate of the device.

4. The multi-dimensional dynamic trust evaluation method for new energy power terminals according to claim 1 or 2, characterized in that: In step 3, the following steps are included: First step: Build IoT device interaction graph G = (V, E, W): Node V: Each IoT device as a node, node attribute is the "knowledge evidence" of the device, build knowledge evidence vector about the node; Edge E: If device A and device B have direct interaction, there is an edge E (A, B); no interaction, no edge; Edge weight W: defined by the "historical interaction quality" of device A and device B, the higher the weight, the better the interaction quality; Step 2: Convert low-dimensional vector based on node2vec: Use node2vec to learn the low-dimensional embedding vector of the node In the IoT scenario, use node2vec to integrate the knowledge evidence and interaction behavior relationship of the device, and convert it into a low-dimensional vector; Step 3: Calculate the direct trust degree of the device: After getting the low-dimensional vector of the device, combine the real-time and security weight of the terminal device in the new energy power plant to design the calculation model, calculate the cosine similarity of the vector Vec_A of device A and the vector Vec_B of device B, and then multiply the interaction frequency weight to avoid misjudgment of low-frequency interaction, that is: The direct trust degree is the product of the cosine similarity and the interaction frequency proportion: is the number of interactions between A and B, is the total number of interactions of A, DT ranges from 0 to 1, the closer the value is to 1, the higher the trust degree of device A to B is; and respectively represent the i-th component in the low-dimensional vector of device A and device B. On the basis of this formula, the dynamic characteristic weight of is added to achieve higher dynamic authenticity, is the time interval of the last interaction between A and B, is the time decay coefficient, which enhances the timeliness of interaction trust, and k represents all other devices that have interactions with device A. The final formula is: 。 5. The multi-dimensional dynamic trust evaluation method for new energy power terminals according to claim 1 or 2, characterized in that: Step 5 includes: According to the number of historical successful and failed interactions, use the Beta distribution to calculate the trust value expectation: In order to pursue timeliness, the historical interaction times are selected by sliding window, and the interaction records within 24 hours are selected for trust value calculation, and a time decay factor is also introduced: ； wherein m is the number of successful historical interactions within the sliding window, n is the number of historical interactions, K is a constant, the greater the value of K, the faster the historical data decays; s represents the number of successful interactions in the historical interactions, f represents the number of failed interactions in the historical interactions, AC is the trust calculation result, Exp represents expectation calculation, represents the number of successful interactions within the time window, represents the number of failed interactions within the time window, represents the i th time window.

6. The multi-dimensional dynamic trust evaluation method for new energy power terminals according to claim 1 or 2, characterized in that: In step 6, set the three-level trust threshold: Comprehensive processing of trust value: Where, DT is the direct trust degree; T is the terminal trust value, AC is the trust value expectation; S≥80 points for high trust, the terminal is running normally, no intervention; S is 60-79 points for medium trust, trigger warning, increase monitoring frequency; S<60 points for low trust, determine as a potential malicious terminal, execute measures such as limiting authority, disconnecting connection, and starting isolation mechanism; Real-time update of terminal trust value for next trust evaluation.

7. A multi-dimensional dynamic trust assessment system for new energy power terminals, characterized in that: Including: Multi-dimensional data processing and standardization module: based on terminal log interaction data, extract terminal general features, and extract special knowledge evidence according to device level perception layer and network layer, and convert all features to 0-1 standardized values; New access terminal initial trust baseline construction module: when the terminal accesses for the first time, based on traditional digital certificate, key verification identity authentication, supplement the collection of its static attributes (hardware features, configuration information, environment parameters), to establish the initial trust baseline and give the initial trust value, solve the trust evaluation starting point problem when there is no historical data of new terminal; Direct trust degree calculation module based on node2vec: first build IoT device interaction graph, node is terminal device, attribute is knowledge evidence vector, edge is device direct interaction relationship, and edge weight is defined by historical interaction quality; Then integrate device knowledge evidence and interaction relationship through node2vec algorithm to generate low-dimensional embedding vector; Finally, calculate the cosine similarity of device vector, and multiply the interaction frequency proportion to get the direct trust degree in the interval [0, 1], quantize the direct trust relationship between devices; Packet flow dimension trust evaluation module: used for analyzing and storing the sending quantity, receiving quantity, packet loss rate and sending / receiving time of terminal data packet, calculating the terminal trust value in this dimension by fusing the formula of packet ratio, packet loss rate and time difference, and supplementing the trust judgment basis in the flow layer; The trust value is calculated by the following equation: ; wherein PS represents the number of data packet sent, PR represents the number of data packet received, the division of which obtains a data packet ratio, PDR represents a packet loss rate, TR represents a time of receiving data packet, TS represents a time of sending data packet; Delay is a data packet delay, Delay_thresh is a delay threshold value; Jitter is a delay jitter, Jitter_thresh is a jitter threshold value; 、 、 、 is a weight of each index, + + + =1. Historical interaction behavior trust evaluation and timeliness optimization module: used for calculating the trust value expectation based on the historical interaction success / failure times by using Beta distribution; using a 24-hour sliding window to filter recent interaction records and introducing a time decay factor to ensure that the trust evaluation focuses on recent behavior and improves the timeliness and accuracy of the evaluation; Multi-level trust threshold setting module: used for setting multi-level trust thresholds; The direct trust degree of the terminal, the terminal trust value and the trust value expectation are comprehensively processed, and the trust level of the terminal is judged according to the set threshold and updated in real time.

8. The multi-dimensional dynamic trust assessment system for new energy power terminal according to claim 7, characterized in that: The general characteristics of the terminal include: terminal type, communication anomaly times, instruction overreach frequency, firmware tampering mark, hardware fault code and neighbor node behavior deviation rate.

9. The multi-dimensional dynamic trust assessment system for new energy power terminal according to claim 7 or 8, characterized in that: The perception layer extracts special knowledge evidence, including: data accuracy, data integrity, online stability and time synchronization of the device; the network layer extracts special knowledge evidence, including: instruction response delay, instruction forwarding success rate and bandwidth utilization rate of the device.

10. The multi-dimensional dynamic trust assessment system for new energy power terminal according to claim 7 or 8, characterized in that: The direct trust degree calculation module based on node2vec includes: Constructing IoT device interaction graph G = (V, E, W) unit: Node V: each IoT device as a node, node attribute as the "knowledge evidence" of the device, constructing the knowledge evidence vector about the node; Edge E: if device A and device B have direct interaction, there is an edge E (A, B); if there is no interaction, there is no edge; Edge weight W: defined by the "historical interaction quality" of device A and device B, the higher the weight, the better the interaction quality; Low-dimensional vector conversion unit based on node2vec: In the IoT scenario, the knowledge evidence and interaction behavior relationship of the device are fused by using node2vec to convert into a low-dimensional vector; Calculate the direct trust degree of the device (quantitative trust relationship) unit: After obtaining the low-dimensional vector of the device, a calculation model is designed combining the real-time and security weight of the terminal device in the new energy power plant, the cosine similarity of the vector Vec_A of device A and the vector Vec_B of device B is calculated, and then multiplied by the interaction frequency weight to avoid misjudgment of low-frequency interaction, that is: The direct trust degree is the product of the cosine similarity and the interaction frequency ratio: is the number of interactions between A and B, is the total number of interactions of A, DT ranges from 0 to 1, the closer the value is to 1, the higher the trust degree of device A to B is; and respectively represent the i-th component in the low-dimensional vector of device A and device B. On the basis of this formula, the dynamic feature weight is added to achieve higher dynamic authenticity, is the time interval of the last interaction between A and B, is the time decay coefficient, which enhances the timeliness of interaction trust, and k represents all other devices that have interactions with device A. The final formula is: 。

Images