CN-121981528-A - Risk assessment method for blockchain technology

CN121981528ACN 121981528 ACN121981528 ACN 121981528ACN-121981528-A

Abstract

The invention belongs to the technical field of blockchains, and discloses a risk assessment method for a blockchain technology, which captures systematic changes such as node increase and decrease, algorithm upgrading and the like through dynamic modeling of a knowledge graph, and combines a time sequence attention mechanism to mine risk along with a transmission rule of a common recognition period, so as to cover the whole stack risk of a technical layer, a data layer and an application layer, the whole period tracking and iterative optimization mechanism ensures that an assessment system continuously adapts to the dynamic characteristics of the blockchain, breaks through neglect of time dimension and association relation by the traditional method, forms complete coverage from risk factor identification to association analysis, analyzes strategy interaction of an attacker and an defender through a game theory model, combines scene risk grade division to generate a coping scheme adapting to a specific service scene, and the strategy generation process is integrated with multidimensional factors such as risk diffusion speed, implementation cost and the like, so as to change from a scientific decision based on data from a simple dependency history case, and solve the problems of strong subjectivity and unreasonable resource allocation of the coping strategy of the traditional method.

Inventors

QIAN JIN
FENG JUN
XI JIAN
Xie Shuichan
XIE YAN
JIANG XUEYING
DUAN YUHAI

Assignees

深圳市前海中基供应链有限公司

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (8)

1. A risk assessment method for a blockchain technology is characterized by comprising the following specific steps: The method comprises the steps of S1, dynamically modeling risk factors based on knowledge graphs, namely integrating the knowledge graphs of multi-source data construction entities, mining risk paths by GRAPHSAGE neighborhood aggregation algorithms, and automatically updating when a system changes to generate a dynamic topological graph, wherein the dynamic topological graph comprises risk factor nodes, associated edges and edge weights, and the risk factor nodes comprise intelligent contract holes, node faults, algorithm force attacks, double-flower attacks, data leakage, cross-chain transmission anomalies and authority override; S2, introducing risk quantification of a time sequence attention mechanism, namely acquiring multi-dimensional time sequence data according to a preset number of block periods on the basis of the step S1, and processing by combining an improved LSTM (least squares) with the time sequence attention mechanism, quantifying a risk value and strengthening key period characteristics; s3, constructing a multi-dimensional risk assessment matrix, namely, based on the quantized value in the step S2, constructing a three-dimensional nine-layer assessment matrix, and integrating each dimension and the comprehensive risk value by using a hierarchical analysis method and entropy weight method weights; s4, training a self-adaptive risk level classification model, namely dynamically classifying a plurality of risk levels according to scenes by utilizing the comprehensive risk value in the step S3 and combining a risk database of the scenes through a reinforcement learning training model; S5, dynamically adjusting a real-time risk early warning threshold value, namely dynamically adjusting the early warning threshold value by using an improved particle swarm algorithm according to the risk level in the step S4, and setting constraint conditions; S6, generating a risk coping strategy based on game theory, namely constructing an attack and defense game model aiming at the risk of the early warning in the step S5, constructing a profit matrix based on historical data, and solving Nash equilibrium to obtain an optimal strategy; and S7, performing risk tracking and model iterative optimization, namely recording related data based on the strategy effect of the step S6, using an incremental learning iterative model, setting performance indexes, and retraining when the performance indexes are not up to standard.
2. The risk assessment method for blockchain technology according to claim 1, wherein the specific steps of the knowledge-graph-based risk factor dynamic modeling in the step S1 are as follows: S11, multi-source data fusion and risk path mining, namely integrating blockchain project white paper and code audit report multi-source data, constructing a knowledge graph comprising technical components, network entities and operation parameters, training a graph neural network through GRAPHSAGE neighborhood aggregation algorithm, and sampling neighbor node characteristics to mine potential risk paths; And S12, outputting a dynamic updating mechanism and an associated framework, namely automatically triggering incremental updating of the knowledge graph and regenerating a risk factor topological graph when the system has node increasing and decreasing and algorithm upgrading events.
3. The method for risk assessment of blockchain technology according to claim 1, wherein the specific steps of introducing the risk quantification of the time series attention mechanism in step S2 are as follows: s21, acquiring time sequence data and capturing characteristics, namely acquiring multidimensional time sequence data of a node layer, a consensus layer and a contract layer according to every 10 block periods on the basis of a risk correlation frame in the step S1, inputting the data into an LSTM (link state transition) network connected with an added residual error, and strengthening gradient transfer in long-sequence training; S22, attention weighting and quantized value output, namely accessing a time sequence attention module at an LSTM output layer, endowing a key period of sudden increase of the computation with high weight, strengthening important time node characteristics, and normalizing a processing result to 0-10 minutes; positive mapping means that the severity of the risk factors is positively correlated with the score, for example, the delay is more than or equal to 100ms and corresponds to 8-10 minutes, negative mapping means that the severity of the risk factors is negatively correlated with the score, for example, the consensus success rate is more than or equal to 99.9% and corresponds to 1-3 minutes.
4. The method for risk assessment of blockchain technology according to claim 1, wherein the specific steps of constructing the multidimensional risk assessment matrix in step S3 are as follows: S31, constructing a three-dimensional nine-layer index system, namely constructing a technology, data and an application three-dimensional evaluation frame based on the quantized value in the step S2, wherein the technical layer comprises 3 indexes, the data comprises 3 indexes, and the application layer comprises 3 indexes; The technical layer indexes comprise consensus mechanism safety, node fault tolerance rate and intelligent contract execution efficiency, the data layer indexes comprise data encryption strength, privacy protection compliance and cross-chain data consistency, and the application layer indexes comprise contract calling success rate, user authority management precision and service scene adaptation degree; and S32, weight integration and comprehensive risk value generation, namely determining weights by adopting a hierarchical analysis method-entropy weight method, integrating all index quantized values by a weighted summation formula, and outputting all dimension scores and comprehensive risk values.
5. The risk assessment method for blockchain technology according to claim 1, wherein the specific steps of training the adaptive risk classification model in step S4 are as follows: S41, constructing a multi-scene risk database, namely constructing a risk event database of a scene based on the comprehensive risk value in the step S3, labeling a risk factor combination and a loss degree label for each case, and forming a standardized training sample through data cleaning and feature extraction; S42, training and adapting the reinforcement learning model, namely taking the risk quantized value vector as a state, taking 5 risk levels as an action space, introducing a reward function into a scene adaptation coefficient, and training the model through an epsilon-greedy strategy to enable the model to have scene adaptation capability.
6. The method for risk assessment of blockchain technology according to claim 1, wherein the specific steps of dynamically adjusting the real-time risk early warning threshold in step S5 are as follows: s51, constructing a real-time state monitoring engine, namely constructing a real-time monitoring system of node health degree and transaction heat running indexes according to the risk level in the step S4, and analyzing on-chain data through an API interface; S52, a dynamic threshold optimization and constraint mechanism is that a particle swarm optimization threshold adjustment formula is adopted, constraint conditions of single adjustment less than or equal to 15% and interval more than or equal to 1 hour are set to prevent oscillation, and emergency adjustment is triggered in an abnormal state.
7. The method for risk assessment of blockchain technology according to claim 1, wherein the specific steps of generating the risk coping strategy based on the game theory in step S6 are as follows: S61, constructing an attack and defense game model, namely constructing an attacker-defender double game model for the risk of early warning in the step S5, defining an attacker strategy and defender strategy, building a profit matrix based on historical cases, and quantifying the profit difference value of different strategy combinations; And S62, generating an optimal strategy and sequencing priorities, namely solving Nash equilibrium by an iterative deletion inferior strategy method, sequencing the strategy priorities by combining the risk diffusion speed and the influence range factors, and outputting a scheme comprising an implementation step and a responsibility main body.
8. The method for risk assessment of blockchain technology according to claim 1, wherein the specific steps of risk tracking and model iterative optimization in step S7 are as follows: S71, constructing a full-period risk tracking system, namely establishing a full-period tracking mechanism of a risk index change curve and early warning response time data for evaluating the strategy effect of the step S6, and recording index data once per hour to form a visual trend chart; S72, incremental learning and model optimization closed loop, namely adopting an incremental learning frame with a 30-day sliding window, freezing parameters of a bottom layer of a model, only updating a top layer classifier to reduce cost and increase efficiency, setting performance indexes with identification accuracy rate more than or equal to 90%, starting full-scale retraining when the identification accuracy rate does not reach the standard, and reserving a historical model through version management; The updated parameters of the top classifier comprise a full-connection layer weight matrix and bias items, the learning rate is 1/5 of the initial training learning rate of the model, and if the initial learning rate is 0.01, the updated learning rate is 0.002.

Description

Risk assessment method for blockchain technology Technical Field The invention belongs to the technical field of blockchain, and particularly relates to a risk assessment method for a blockchain technology. Background The prior blockchain risk assessment method has obvious limitation, and the invention patent with publication number of CN119809348A discloses a system for carrying out risk assessment on blockchain projects, which adopts a static weight distribution mechanism, depends on expert experience to preset index weights, cannot respond to dynamic changes of risk factors such as intelligent contract vulnerability triggering frequency and the like, and is difficult to adapt to the real-time state of the system. The threshold value adjustment adopts fixed standards, does not consider the differentiated requirements of scenes such as finance, supply chains and the like, is difficult to realize accurate early warning in the face of delayed response when the system changes such as node increase and decrease, algorithm upgrading and the like, and the risk response scheme is based on historical case experience summary, lacks quantitative analysis on strategy interaction of an attacker and an defender, and is strong in subjectivity of strategy generation and insufficient in resource allocation rationality. In addition, an effective model iteration mechanism is not established in the method, when the system is subjected to node change or algorithm update, the evaluation model parameters cannot be dynamically adapted, so that the recognition accuracy in long-term use is reduced, and the risk evaluation requirement under the rapid development of the blockchain technology is difficult to meet. Disclosure of Invention The present invention is directed to a risk assessment method for blockchain technology, so as to solve the above-mentioned problems in the background art. In order to achieve the above purpose, the invention provides a risk assessment method for a blockchain technology, which comprises the following specific steps: S1, risk factor dynamic modeling based on knowledge graph, namely integrating multi-source data to construct knowledge graph containing entities such as technical components, mining risk paths by GRAPHSAGE neighborhood aggregation algorithm, automatically updating when the system changes, and generating dynamic topological graph; S2, risk quantification for introducing a time sequence attention mechanism is carried out, wherein multi-dimensional time sequence data are collected according to a preset number of block periods (such as 10) on the basis of the step S1, an average detection window based on a block chain typical attack (such as a double-flower attack) is set in the period, the block interval characteristics of a main stream block chain (such as an Ethernet) are adapted, key time nodes of risk conduction can be covered, the risk values (0-10 minutes) are quantified by combining the improved LSTM with the time sequence attention mechanism, and the key period characteristics are strengthened; s3, constructing a multi-dimensional risk assessment matrix, namely constructing a three-dimensional nine-layer assessment matrix (three indexes are contained in a technology, data and an application layer respectively) based on the quantized value in the step S2, and integrating each dimension and the comprehensive risk value by using a hierarchical analysis method and entropy weight legal weight; s4, training a self-adaptive risk level classification model, namely dynamically classifying a plurality of risk levels (such as 5) according to scenes by using the comprehensive risk value in the step S3 and combining a risk database of 12 types of scenes through a reinforcement learning training model, wherein the threshold values of the scenes of finance and the supply chain are different; S5, dynamically adjusting a real-time risk early warning threshold value, namely dynamically adjusting the early warning threshold value by using an improved particle swarm algorithm according to the risk level in the fourth step, setting a constraint condition to prevent vibration, and emergently reducing the threshold value in abnormal conditions to improve the timeliness of risk response; s6, generating a risk coping strategy based on game theory, namely constructing an attack and defense game model aiming at the risk of the early warning in the step S5, constructing a profit matrix based on historical data, solving Nash equilibrium to obtain an optimal strategy, and sequencing an output scheme according to priority; and S7, performing risk tracking and model iterative optimization, namely tracking the strategy effect of the step S6, recording related data, using an incremental learning iterative model, setting performance indexes, retraining when the performance indexes are not up to standard, forming a closed loop through version management, and ensuring that an evaluation system is continuously adapted to