A digital asset load early warning management method and system based on machine learning

Abstract

The present invention provides a digital asset load early warning management method and system based on machine learning, which belongs to the field of financial technology and includes the following steps: S101: establishing an intelligent perception layer, capturing multi-dimensional heterogeneous data, and establishing four-dimensional heterogeneous cube data; S102: establishing an intelligent analysis layer, generating and exporting perception optimization data and prediction and early warning data based on a dual-model coupling architecture, outputting the perception optimization data to the intelligent perception layer, and outputting the prediction and early warning data and the perception optimization data to the decision control layer; S103: establishing a decision control layer, synthesizing the final dynamic threshold based on a dual-track collaborative threshold mechanism; S104: establishing an early warning response layer, executing a grading strategy based on the final dynamic threshold; S105; establishing an evolutionary feedback layer for optimizing the intelligent analysis layer and the decision control layer; S106: recording relevant operating data and early warning information; the method can provide early warnings for digital asset transactions and improve the security of transactions.

Description

Technical Field

[0001] The present invention belongs to the field of financial technology, and more specifically, relates to a digital asset load early warning management method and system based on machine learning. Background Art

[0002] Driven by the wave of digitalization, digital assets have become one of the important core assets of enterprises and institutions. From the massive financial data held by financial institutions, to the user information and business data accumulated by Internet companies, to the experimental data of various scientific research institutions, these digital assets carry huge value and play a key role in corporate operations, market decision-making and innovative development. In the transaction of digital assets, there will be a large amount of data exchange. Traditional digital asset load warning management methods are often less secure and less accurate in the transaction of data assets. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides a digital asset load warning management method and system based on machine learning to solve the technical problems in the existing technology that the traditional digital asset load warning management method has low security and low warning accuracy.

[0004] The purpose and effectiveness of the digital asset load early warning management method and system based on machine learning of the present invention are achieved by the following specific technical means:

[0005] A digital asset load early warning management method based on machine learning includes the following steps:

[0006] S101: Establish an intelligent perception layer to capture multivariate heterogeneous data. Multivariate heterogeneous data is used to represent the diversified data of digital asset transaction loads. After the multivariate heterogeneous data is captured, denoising is performed and four-dimensional heterogeneous cube data is created based on the multivariate heterogeneous data.

[0007] S102: Establishing an intelligent analysis layer. The intelligent analysis layer is established based on a dual-model dynamic coupling architecture. Based on the four-dimensional heterogeneous cube data and the dual-model coupling architecture, it generates and exports perception optimization data and prediction and warning data. The intelligent analysis layer and the intelligent perception layer form a closed loop. The perception optimization data is output to the intelligent perception layer for optimization of the intelligent perception layer. The prediction and warning data and perception optimization data are output to the decision control layer.

[0008] S103: Establishing a decision control layer. The decision control layer is established based on a dual-track collaborative threshold mechanism. The dual-track collaborative threshold mechanism includes a static threshold and a dynamic threshold. The static threshold is generated based on four-dimensional heterogeneous cube data, and the dynamic threshold is generated based on the PPO algorithm. The final dynamic threshold is synthesized based on the static and dynamic thresholds. At the same time, the final dynamic threshold is output to the early warning response layer.

[0009] S104: Establishing an early warning response layer, which executes a grading strategy based on the final dynamic threshold and sends relevant early warning information to the administrator. At the same time, the early warning information is output to the evolution feedback layer.

[0010] S105: Establish an evolutionary feedback layer. The evolutionary feedback layer generates analysis optimization data and decision optimization data based on the past five minutes of slices, warning information, and final dynamic thresholds of the four-dimensional heterogeneous cube data. The analysis optimization data is output to the intelligent analysis layer, and the decision optimization data is sent to the decision control layer.

[0011] S106: During the execution of S101 to S105, relevant operating data and warning information are recorded to facilitate maintenance and optimization by the administrator.

[0012] As a further solution of the present invention, S101 specifically includes:

[0013] The intelligent perception layer includes three links when it is established: native link, market link and physical link;

[0014] Multi-dimensional heterogeneous data refers to diverse data obtained from different sources, forms, and dimensions. Specifically, multi-dimensional heterogeneous data is obtained by acquiring TPS data, smart contract call frequency, gas price volatility, and cross-chain transaction latency based on native links. This data is used to reflect the real-time status of digital asset transactions. Market data and market sentiment index are obtained based on market links. This data is used to reflect the market environment and the potential impact of market sentiment. Physical links are used to obtain server CPU / memory / bandwidth utilization and server environment parameters. This data is used to reflect the physical carrying capacity boundaries of servers used to process digital asset transactions.

[0015] After the multivariate heterogeneous data is acquired, data cleaning operations are performed on the multivariate heterogeneous data to ensure the normality of the multivariate heterogeneous data. The multivariate heterogeneous data that has completed data cleaning is then denoised based on the sliding window filtering algorithm to remove short-term noise in the multivariate heterogeneous data.

[0016] When performing the denoising operation, the window size is dynamically adjusted and the sensitivity coefficient is set to 0.85;

[0017] When acquiring multivariate heterogeneous data, the time information and regional location information of the multivariate heterogeneous data are acquired simultaneously;

[0018] Based on the structured integration of multi-dimensional heterogeneous data, time information and regional location information, four-dimensional heterogeneous cube data is established. The four-dimensional heterogeneous cube data includes time dimension, space dimension, chain layer dimension and physical dimension. Based on the four-dimensional heterogeneous cube data, multi-dimensional data fusion is achieved;

[0019] The intelligent analysis layer outputs the four-dimensional heterogeneous cube data to the intelligent analysis layer, decision control layer and evolution feedback layer respectively.

[0020] As a further solution of the present invention, S102 specifically includes:

[0021] The intelligent analysis layer connects data with the intelligent perception layer, and the intelligent analysis layer receives the four-dimensional heterogeneous cube data output from the intelligent perception layer;

[0022] The dual-model dynamic coupling architecture is coupled by the TCN-LSTM hybrid model and the association analysis model;

[0023] The TCN-LSTM hybrid model is used to generate perception optimization data. The time, physical, and chain-level dimensions of four-dimensional heterogeneous cube data are fed into the TCN-LSTM hybrid model. The TCN-LSTM hybrid model extracts the past two hours of load time series data from these dimensions. The load time series data consists of 30-second TPS data, CPU / memory / bandwidth utilization, and server environment parameters. Abnormal data in the load time series data is removed and normalized. The TCN-LSTM hybrid model generates and exports perception optimization data based on the load time series data. The perception optimization data represents predicted abnormal transaction operations within the next hour.

[0024] After the perception optimization data is generated and exported, the intelligent analysis layer outputs the perception optimization data to the intelligent perception layer, and the intelligent perception layer performs optimization based on the perception optimization data;

[0025] The association analysis model is used to generate prediction and warning data. The association analysis model is built based on a hypergraph neural network. The four-dimensional heterogeneous cube data is imported into the association analysis model as input. The association analysis model establishes a digital asset transaction hypergraph based on the four-dimensional heterogeneous cube data. The prediction and warning data is generated and exported based on the digital asset transaction hypergraph. After the prediction and warning data is generated and exported, the intelligent analysis layer outputs the prediction and warning data to the decision control layer.

[0026] As a further solution of the present invention, the intelligent perception layer performs optimization based on the perception optimization data, including:

[0027] After receiving the data from the intelligent perception layer, the intelligent perception layer performs judgment and comparison operations based on the optimized threshold;

[0028] If the perception optimization data is less than the optimization threshold, it means that there are fewer abnormal trading operations predicted in the next hour. In this case, the intelligent perception layer will reduce the crawling rate of multi-dimensional heterogeneous data by 0.7, and at the same time partially release the server resources used for data crawling operations.

[0029] If the perception optimization data is greater than the optimization threshold, it means that there are many abnormal transaction operations predicted in the next hour. In this case, the intelligent perception layer increases the crawling rate of multi-dimensional heterogeneous data. At the same time, the intelligent perception layer dispatches excess server resources for data crawling operations.

[0030] When the perception optimization data is greater than the optimization threshold, the intelligent perception layer performs hierarchical control. When the perception optimization data exceeds the optimization threshold by 15%, the crawling rate increases to 1.2 times, and the intelligent perception layer dispatches a small amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 30%, the crawling rate increases to 1.5 times, and the intelligent perception layer dispatches an appropriate amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 50%, the crawling rate increases to 2 times, and the intelligent perception layer dispatches a large amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 80%, the expansion contract and review contract are triggered.

[0031] As a further solution of the present invention, the expansion contract and the review contract include:

[0032] The expansion contract automatically connects the hot backup server, and the intelligent perception layer and intelligent analysis layer schedule the resources of the hot backup server for expansion operation, allocating a maximum of 50% of the hot backup server's resources. At the same time, the operation of the evolutionary feedback layer is stopped, and the resources of the evolutionary feedback layer are dispatched to the intelligent perception layer and intelligent analysis layer to maximize the intelligent perception layer's crawling of multi-heterogeneous data and the intelligent analysis layer's data generation operations. Until the perception optimization data drops to within 50% of the optimization threshold, the expansion contract is stopped, the hot backup server is disconnected, and the operation of the evolutionary feedback layer is resumed.

[0033] The review contract specifically sends an abnormal alert to the administrator, who will manually review the relevant data in the intelligent analysis layer and the intelligent perception layer. The administrator can manually stop the expansion contract, thereby disconnecting the connected hot backup server and restoring the operation of the evolution feedback layer.

[0034] As a further solution of the present invention, S103 specifically includes:

[0035] The decision control layer is connected to the intelligent analysis layer and the intelligent perception layer respectively. The decision control layer receives the perception optimization data and prediction and warning data from the intelligent analysis layer, and receives the four-dimensional heterogeneous cube data from the intelligent perception layer.

[0036] The data source for the static threshold is the peak CPU / memory / bandwidth utilization in the physical dimension and the peak TPS data in the chain layer dimension in the four-dimensional heterogeneous cube data. Before obtaining the static threshold, an attenuation compensation factor is generated based on the attenuation of hardware performance. The static threshold is calculated based on the attenuation compensation factor and the static threshold data source: ;

[0037] in, Expressed as a static threshold, 、 、 and Represented as weight factors of CPU / memory / bandwidth utilization and TPS data respectively, Expressed as the peak CPU utilization, Expressed as peak memory utilization, Expressed as the peak bandwidth utilization, Expressed as TPS data peak, Expressed as a constant, Expressed as attenuation compensation factor;

[0038] The data sources for the dynamic threshold are perception optimization data, forecast and warning data, and the historical maximum TPS data of the chain layer dimension in the four-dimensional heterogeneous cube data, as well as the market data and market public opinion sentiment index of the chain layer dimension. Before obtaining the dynamic threshold, the market correction factor is obtained based on the market data and market public opinion sentiment index. The calculation formula for the market correction factor is: ;

[0039] in, Expressed as the market correction factor, Represented as market data, Expressed as the market sentiment index, the market correction factor changes dynamically with the market data captured by the intelligent perception layer and the market sentiment index;

[0040] Dynamically obtain dynamic thresholds based on market correction factors, PPO algorithm and dynamic threshold data source;

[0041] Based on the static threshold and dynamic threshold, dual-track collaborative synthesis is performed to generate the final dynamic threshold, which is then output to the early warning response layer.

[0042] As a further solution of the present invention, S104 specifically includes:

[0043] The early warning response layer connects with the decision control layer and receives the final dynamic threshold from the decision control layer.

[0044] After the early warning response layer receives the final dynamic threshold from the decision control layer, it executes the classification strategy based on the final dynamic threshold. The classification strategy includes blue warning, yellow warning, orange warning and red warning.

[0045] Compare the final dynamic threshold to 1:

[0046] If the final dynamic threshold is between 40% and 60%, a blue warning message will be sent to the administrator, and preventive resource allocation will be performed at the same time. The hot backup server will be automatically connected and resource allocation will be performed based on the hot backup server.

[0047] If the final dynamic threshold is between 60% and 75%, a yellow warning message will be sent to the administrator, and the digital asset transaction process will be optimized. The free transaction links will be allocated based on the spatial dimension to carry out the digital asset transaction process.

[0048] If the final dynamic threshold is between 75% and 90%, an orange warning message will be sent to the administrator, and digital asset disaster recovery operations will be implemented. High-risk operations in the transaction link will be stopped based on the spatial dimension, transaction links will be scheduled based on the spatial dimension, and backup transaction links will be activated for digital asset transactions.

[0049] If the final dynamic threshold is greater than 90% and digital asset preservation operations are being executed at the same time, a red alert message will be sent to the administrator and asset preservation operations will be executed at the same time, immediately freezing digital assets that are not being traded based on the spatial dimension;

[0050] If the final dynamic threshold is not within the above range, no relevant action will be performed;

[0051] When the early warning response layer is running, a visual progress bar is generated. Administrators can directly watch the progress bar to understand the range of the final dynamic threshold. The visual progress bar is rectangular, and each quarter of the progress bar is separated and filled with blue, yellow, orange and red respectively. The final dynamic threshold is displayed at one end of the progress bar.

[0052] As a further solution of the present invention, S105 specifically includes:

[0053] The evolutionary feedback layer is connected to the decision control layer, the execution response layer, and the intelligent perception layer respectively. The evolutionary feedback layer receives the final dynamic threshold from the decision control layer, the past five-minute slices of the four-dimensional heterogeneous cube data of the intelligent perception layer, and the warning information of the warning response layer, and generates analysis optimization data and decision optimization data based on the above information. After generation, the evolutionary feedback layer outputs and imports the analysis optimization data and decision optimization data into the intelligent analysis layer and the decision control layer respectively, thereby performing feedback optimization on the dual-model dynamic coupling architecture of the intelligent analysis layer and the dual-track collaborative threshold mechanism of the decision control layer.

[0054] A digital asset load early warning management system based on machine learning, comprising:

[0055] The intelligent perception layer includes three links: native link, market link, and physical link. It is used to capture multi-dimensional heterogeneous data and build four-dimensional heterogeneous cube data based on multi-dimensional heterogeneous data.

[0056] Native link, used to obtain TPS data, smart contract call frequency, gas price volatility and cross-chain transaction latency;

[0057] Market link, used to obtain market data and market sentiment index;

[0058] Physical link, used to obtain server CPU / memory / bandwidth utilization and server environment parameters;

[0059] The intelligent analysis layer is built based on a dual-model dynamic coupling architecture. The dual-model dynamic coupling architecture is coupled by a TCN-LSTM hybrid model and an association analysis model. The intelligent analysis layer generates and exports perception optimization data based on the TCN-LSTM hybrid model and generates and exports warning prediction data based on the association analysis model.

[0060] The decision-making control layer is established based on the dual-track collaborative threshold mechanism. The dual-track collaborative threshold mechanism is used to generate static thresholds and dynamic thresholds. The decision-making control layer generates the final dynamic threshold based on the dual-track collaborative synthesis;

[0061] The early warning response layer is used to execute the hierarchical strategy and send relevant early warning information to the administrator based on the hierarchical strategy;

[0062] The evolutionary feedback layer is used to generate analysis and decision optimization data, thereby providing feedback optimization for the dual-model dynamic coupling architecture of the intelligent analysis layer and the dual-track collaborative threshold mechanism of the decision control layer;

[0063] Hot backup server, used as a backup server for resource scheduling;

[0064] Output execution modules for preventive resource allocation, digital asset transaction process optimization, digital asset disaster recovery, or asset preservation operations;

[0065] The data recording module is used to record relevant operating data and warning information during the execution of steps S101 to S105 to facilitate maintenance and optimization by the administrator.

[0066] As a further solution of the present invention, the output end of the intelligent perception layer is unidirectionally connected to the input ends of the intelligent analysis layer, the decision control layer and the evolutionary feedback layer based on the communication link; the output end of the intelligent analysis layer is unidirectionally connected to the input end of the decision control layer and the intelligent perception layer based on the communication link; the intelligent analysis layer and the intelligent perception layer form a closed loop; the output end of the decision control layer is unidirectionally connected to the early warning response layer and the input end of the evolutionary feedback layer based on the communication link; the output end of the early warning response layer is unidirectionally connected to the evolutionary feedback layer and the input end of the output execution module based on the communication link; the output end of the evolutionary feedback layer is unidirectionally connected to the input end of the decision control layer and the intelligent analysis layer based on the communication link; the hot backup server is bidirectionally connected to the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer based on the communication link; the input end of the data recording module is unidirectionally connected to the output ends of the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer based on the communication link.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] First, establish the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer, and connect the data of each layer, obtain multi-dimensional heterogeneous data through the intelligent perception layer, and establish four-dimensional heterogeneous cube data based on the multi-dimensional heterogeneous data. Then, output the four-dimensional heterogeneous cube data to the intelligent analysis layer, evolutionary feedback layer and decision control layer. The intelligent analysis layer generates and exports perception optimization data and prediction and early warning data based on the four-dimensional heterogeneous cube data, and imports the perception optimization data into the intelligent perception layer, so that the intelligent perception layer triggers hierarchical control or capacity expansion contracts, thereby improving the data settlement rate of the intelligent perception layer and increasing the amount of data obtained, thereby improving the accuracy of the prediction and early warning data obtained by the intelligent analysis layer, and at the same time outputs the prediction and early warning data and perception optimization data to the decision control layer. Afterwards, the decision control layer contracts the static threshold and the dynamic threshold based on the dual-track collaborative threshold mechanism. Thus, a final dynamic threshold is generated, and the final dynamic threshold is output to the early warning response layer. The early warning response layer performs a hierarchical strategy based on the final dynamic threshold, thereby performing relevant operations from preventive resource allocation, digital asset transaction process optimization operations, digital asset disaster recovery operations, asset preservation operations or not performing related actions. The administrator can directly understand the range of the final dynamic threshold through a visual progress bar, which improves convenience. While the method process is in progress, analysis optimization data and decision optimization data can be generated based on the evolutionary feedback layer and output to the intelligent analysis layer and the decision control layer respectively, thereby optimizing the decision control layer and the intelligent analysis layer. This method can optimize the resource scheduling of the server load in the ongoing digital transaction based on the hierarchical strategy. When the warning level is high, the digital assets of the trading users can be protected, which improves security. In addition, this method can improve the accuracy of digital asset transaction warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a flowchart of the steps of a digital asset load early warning management method based on machine learning in the present invention;

[0070] Figure 2 It is a schematic diagram of a digital asset load early warning management system based on machine learning of the present invention;

[0071] Figure 3 This is an example diagram of a visual progress bar in a digital asset load early warning management method based on machine learning in the present invention;

[0072] Figure 4 This is a schematic diagram of the intelligent perception layer in a digital asset load early warning management system based on machine learning in the present invention. DETAILED DESCRIPTION

[0073] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the technical solutions of the present invention, but are not intended to limit the scope of protection of the present invention.

[0074] Example 1:

[0075] As attached Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 As shown:

[0076] The present invention provides a digital asset load early warning management method based on machine learning, which is suitable for performance evaluation and includes the following steps:

[0077] S101: Establish an intelligent perception layer to capture multivariate heterogeneous data. Multivariate heterogeneous data is used to represent the diversified data of digital asset transaction loads. After the multivariate heterogeneous data is captured, denoising is performed and four-dimensional heterogeneous cube data is established based on the multivariate heterogeneous data.

[0078] Specifically, the intelligent perception layer includes three links when it is established, namely the native link, the market link and the physical link. During the establishment, a dedicated data blockchain network based on DPoS+BFT is established, and the data collection rules and verification logic are defined through smart contracts to ensure the integrity and non-tamperability of the native link. The market link is stored in the block after verification through the zero-knowledge proof of the oracle network. The physical link is signed by the edge node and then stored on the chain, forming a full-link traceable data collection system.

[0079] Multi-dimensional heterogeneous data refers to diversified data obtained from different sources, forms and dimensions. The specific method of obtaining multi-dimensional heterogeneous data is to obtain TPS data, smart contract call frequency, gas price volatility and cross-chain transaction delay based on the native link. The above data is used to reflect the real-time status of digital asset transactions. Market data and market public opinion sentiment index are obtained based on the market link. The above data are used to reflect the market environment and the potential impact of market public opinion. The physical link is used to obtain server CPU / memory / bandwidth utilization and server environment parameters. The above data is used to reflect the physical carrying capacity boundaries of the server used to process digital asset transactions.

[0080] Among them, the data obtained by the native link respectively represent: TPS data reflects the transaction processing capacity of the blockchain network in real time and is used to calculate the instantaneous load rate; the frequency of smart contract calls is used to identify resource competition bottlenecks that may be caused by high-frequency contracts; Gas price volatility is used to predict network congestion trends and provide early warnings for speculative behavior in the associated market; cross-chain transaction delay is used to evaluate cross-chain bridging efficiency and discover bottlenecks in collaboration between heterogeneous chains.

[0081] The data obtained from the market link are respectively represented as follows: market data is used to reflect the fluctuations of the digital asset trading market, and the market public opinion sentiment index is used to quantify the potential driving force of market sentiment on user trading behavior.

[0082] The data obtained from the physical link represents: CPU / memory / bandwidth utilization is used to dynamically evaluate the physical carrying capacity boundaries of the digital asset trading server, and server environmental parameters are used to prevent sudden performance degradation caused by hardware overheating.

[0083] After the multivariate heterogeneous data is acquired, data cleaning operations are performed on the multivariate heterogeneous data to ensure the normality of the multivariate heterogeneous data. The multivariate heterogeneous data that has completed data cleaning is denoised based on the sliding window filtering algorithm to remove the short-term noise of the multivariate heterogeneous data. When performing the denoising operation, the window size is dynamically adjusted, and the sensitivity coefficient is set to 0.85.

[0084] Specifically, when performing data cleaning operations on multivariate heterogeneous data, an improved isolation forest algorithm is used to set differentiated detection and cleaning strategies for different types of data based on different data within the multivariate heterogeneous data. For data collected by the native link, the integrity of the data can be verified based on the hash value of the transaction signature. For data collected by the market link, the integrity of the data can be verified based on oracle signature verification and multi-party comparison. For data collected by the physical link, a static physical threshold interval can be set to verify the integrity of the data. After the cleaning of the multivariate heterogeneous data is completed, the missing values ​​therein are processed. For relational data such as market data and market public opinion sentiment index, potential associations can be inferred and data can be completed based on graph neural networks. For time series data such as gas price volatility and TPS data, data integrity can be filled based on bidirectional LSTM prediction, and administrators can also manually fill in missing data. When removing noise from multivariate heterogeneous data, median filtering or Gaussian filtering can also be used to denoise the multivariate heterogeneous data.

[0085] Furthermore, when acquiring multivariate heterogeneous data, time information and regional location information of the multivariate heterogeneous data are acquired simultaneously.

[0086] Based on the structured integration of multi-dimensional heterogeneous data, time information and regional location information, four-dimensional heterogeneous cube data is established. The four-dimensional heterogeneous cube data includes time dimension, space dimension, chain layer dimension and physical dimension. The fusion of multi-dimensional data is achieved based on the four-dimensional heterogeneous cube data.

[0087] Among them, the time dimension includes the time information of the data, the spatial dimension includes the regional location information of the digital asset trading server, the chain layer dimension includes the data information collected by the native link and the market link, and the physical dimension includes the data information collected by the physical link.

[0088] Specifically, the scattered native links, market links, physical links, time information, and regional location information serve as the input sources of four-dimensional heterogeneous cube data. The original data is converted into computable knowledge units through spatiotemporal labeling, dimensional classification, and associative coding. The functions of the computable knowledge units are then coupled to establish four-dimensional heterogeneous cube data.

[0089] The intelligent analysis layer outputs the four-dimensional heterogeneous cube data to the intelligent analysis layer, decision control layer and evolution feedback layer respectively.

[0090] S102: Establish an intelligent analysis layer. The intelligent analysis layer is established based on a dual-model dynamic coupling architecture. Based on the four-dimensional heterogeneous cube data and the dual-model coupling architecture, perception optimization data and prediction and warning data are generated and exported. The intelligent analysis layer and the intelligent perception layer form a closed loop. The perception optimization data is output to the intelligent perception layer for optimizing the intelligent perception layer. The prediction and warning data and perception optimization data are output to the decision control layer.

[0091] Specifically, the intelligent analysis layer is connected to the intelligent perception layer data. The intelligent analysis layer receives the four-dimensional heterogeneous cube data output from the intelligent perception layer. The dual-model dynamic coupling architecture is coupled by the TCN-LSTM hybrid model and the association analysis model.

[0092] The TCN-LSTM hybrid model is used to generate perception optimization data. The time, physical, and chain-layer dimensions of four-dimensional heterogeneous cube data are imported into the TCN-LSTM hybrid model as input. The TCN-LSTM hybrid model dispatches the past two hours of load time series data from the physical and chain-layer dimensions based on the time dimension, removes abnormal data from the load time series data, and normalizes the load time series data. During normalization, RobustScaler can be used to standardize the data. The load time series data consists of 30-second TPS data, CPU / memory / bandwidth utilization, and server environment parameters. Fourier transform is used to extract the components of each daily / weekly / monthly cycle of the data to generate a periodic encoding vector. The TCN-LSTM hybrid model generates and exports optimization data based on the periodic encoding vector generated by the load time series data. The perception optimization data is represented by predicted abnormal transaction operations within the next hour.

[0093] After the perception optimization data is generated and exported, the intelligent analysis layer outputs the perception optimization data to the intelligent perception layer, and the intelligent perception layer performs optimization based on the perception optimization data.

[0094] Furthermore, after receiving the data from the intelligent perception layer, the intelligent perception layer performs a judgment and comparison operation based on the optimized threshold, and the judgment and comparison operation includes:

[0095] If the perception optimization data is less than the optimization threshold, it means that there are fewer abnormal transaction operations predicted in the next hour. In this case, the intelligent perception layer reduces the capture rate of multi-dimensional heterogeneous data by 0.7, and at the same time partially releases the server resources used for data capture operations, so that the data transaction server can better allocate its resources, thereby improving the speed of data transactions.

[0096] If the perception optimization data is greater than the optimization threshold, it means that there are many abnormal operations in the predicted transactions in the next hour. In this case, the intelligent perception layer increases the capture rate of multi-dimensional heterogeneous data. At the same time, the intelligent perception layer dispatches excess server resources for data capture operations. When there are many abnormal operations in the predicted transactions in the next hour, it means that there may be abnormal behaviors such as network attacks or server overloads. More resources will be dispatched for the intelligent perception layer to capture data, increase the amount of data obtained, thereby improving the accuracy of the intelligent analysis layer during analysis and providing better responses for subsequent steps.

[0097] Furthermore, when the perception optimization data is greater than the optimization threshold, the intelligent perception layer performs hierarchical control, which includes:

[0098] When the perception optimization data exceeds the optimization threshold by 15%, the crawling rate increases to 1.2 times, and the intelligent perception layer dispatches a small amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 30%, the crawling rate increases to 1.5 times, and the intelligent perception layer dispatches an appropriate amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 50%, the crawling rate increases to 2 times, and the intelligent perception layer dispatches a large amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 80%, the expansion contract and the review contract are triggered. When there is too much abnormal operation data, it means that abnormal behavior is very likely to occur, and further resource scheduling is required to ensure the data crawling volume of the intelligent perception layer and at the same time improve the analysis accuracy of the intelligent analysis layer.

[0099] Specifically, the expansion contract and review contract include:

[0100] The expansion contract is specifically to automatically connect the hot backup server, and the intelligent perception layer and the intelligent analysis layer schedule the resources of the hot backup server for expansion operation, with a maximum allocation of 50% of the resources of the hot backup server. The maximum allocation of 50% of the resources of the hot backup server can reserve spare resources for the subsequent early warning response layer to perform related actions. At the same time, the operation of the evolutionary feedback layer is stopped, and the resources of the evolutionary feedback layer are scheduled to the intelligent perception layer and the intelligent analysis layer to maximize the intelligent perception layer's capture operation of multi-heterogeneous data and the intelligent analysis layer's data generation operation, until the perception optimization data is reduced to within 50% of the optimization threshold, the expansion contract is stopped, the connected hot backup server is disconnected, and the operation of the evolutionary feedback layer is restored.

[0101] The review agreement specifically sends an abnormal alert to the administrator, who then manually reviews the relevant data in the intelligent analysis layer and the intelligent perception layer. Based on manual review, the data can be better analyzed and viewed, thereby eliminating the possibility of data errors caused by abnormal operation of the data transaction server itself. The administrator can manually stop the expansion agreement, thereby disconnecting the connected hot backup server and restoring the operation of the evolution feedback layer.

[0102] The association analysis model is used to generate prediction and warning data. The association analysis model is built based on a hypergraph neural network. The four-dimensional heterogeneous cube data is imported into the association analysis model as input. The association analysis model establishes a digital asset transaction hypergraph based on the four-dimensional heterogeneous cube data. The prediction and warning data is generated and exported based on the digital asset transaction hypergraph. After the prediction and warning data is generated and exported, the intelligent analysis layer outputs the prediction and warning data to the decision control layer.

[0103] Specifically, the association analysis model is constructed based on the four dimensions of the four-dimensional heterogeneous cube data. During the construction, the timestamps are aligned based on the time dimension, and the relevant data of the timestamp unlinked layer dimension within 24 hours are removed at the same time. Then, the relevant data in the four-dimensional heterogeneous cube data are relationally connected to construct a digital asset transaction hypergraph. Finally, the association analysis model generates prediction and warning data based on the digital asset transaction hypergraph and the hypergraph convolution formula.

[0104] S103: Establish a decision control layer. The decision control layer is established based on a dual-track collaborative threshold mechanism. The dual-track collaborative threshold mechanism includes a static threshold and a dynamic threshold. The static threshold is generated based on four-dimensional heterogeneous cube data, and the dynamic threshold is generated based on the PPO algorithm. The final dynamic threshold is synthesized based on the static threshold and the dynamic threshold. At the same time, the final dynamic threshold is output to the early warning response layer.

[0105] Specifically, the decision control layer is connected to the intelligent analysis layer and the intelligent perception layer respectively. The decision control layer receives the perception optimization data and prediction and warning data from the intelligent analysis layer, and receives the four-dimensional heterogeneous cube data from the intelligent perception layer.

[0106] The data source for the static threshold is the peak CPU / memory / bandwidth utilization in the physical dimension and the peak TPS data in the chain layer dimension in the four-dimensional heterogeneous cube data. Before obtaining the static threshold, an attenuation compensation factor is generated based on the attenuation of hardware performance. The calculation formula for the attenuation compensation factor can be expressed as:

[0107] ;

[0108] in, Expressed as the attenuation compensation factor, Expressed as the maximum value of the attenuation compensation factor, the above formula indicates that the hardware performance attenuates by 2% per month. In actual applications, fitting changes can be made based on historical data of physical dimensions.

[0109] The static threshold is calculated based on the attenuation compensation factor and the static threshold data source. The calculation formula can be expressed as:

[0110] ;

[0111] in, Expressed as a static threshold, 、 、 and Represented as weight factors of CPU / memory / bandwidth utilization and TPS data respectively, Expressed as the peak CPU utilization, Expressed as peak memory utilization, Expressed as the peak bandwidth utilization, Expressed as TPS data peak, Expressed as a constant, Expressed as the attenuation compensation factor.

[0112] The data sources for the dynamic threshold are perception optimization data, forecast and warning data, and the historical maximum TPS data of the chain layer dimension in the four-dimensional heterogeneous cube data, as well as the market data and market public opinion sentiment index of the chain layer dimension. Before obtaining the dynamic threshold, the market correction factor is obtained based on the market data and market public opinion sentiment index. The calculation formula for the market correction factor is: ;

[0113] in, Expressed as the market correction factor, Represented as market data, Expressed as the market sentiment index, the market correction factor changes dynamically with the market data captured by the intelligent perception layer and the market sentiment index.

[0114] Dynamic thresholds are dynamically obtained based on market correction factors, PPO algorithm and dynamic threshold data sources.

[0115] Specifically, after the market correction factor is obtained, the PPO algorithm is calibrated based on the reward function. The 1-hour perception optimization data of TCN-LSTM is input as the baseline, and the output adjustment of the PPO algorithm is superimposed. The calculation formula of the dynamic threshold is:

[0116] ;

[0117] in, Expressed as a dynamic threshold, Expressed as the market correction factor, Expressed as the output adjustment amount of the PPO algorithm, Represented as perception optimization data, For early warning forecast data.

[0118] Based on the static threshold and dynamic threshold, dual-track collaborative synthesis is performed to generate the final dynamic threshold, which is then output to the early warning response layer.

[0119] Specifically, when performing the final dynamic threshold, a dynamic weight factor is introduced to perform dual-track collaborative synthesis. The calculation formula of the dynamic weight factor is:

[0120] ;

[0121] in, Expressed as a dynamic weight factor, Data represented as physical dimensions, Represented as chain-level dimension data, Data represented as a time dimension.

[0122] The calculation formula for the dual-track collaborative contract is:

[0123] ;

[0124] in, Denoted as the final dynamic threshold, Expressed as a dynamic weight factor, Expressed as a static threshold, Expressed as a dynamic threshold.

[0125] During the dual-track collaborative synthesis, a safety boundary mechanism is introduced to prevent the final dynamic threshold generated from being offset, which may cause errors in subsequent steps. The safety boundary mechanism imposes interval constraints on the final dynamic threshold. The interval constraint formula is:

[0126] ;

[0127] When the final dynamic threshold exceeds the interval constraint, the intelligent analysis layer is activated to regenerate the early warning prediction data, and at the same time, the intelligent perception layer is triggered to review the data.

[0128] S104: Establish an early warning response layer, which executes a grading strategy based on the final dynamic threshold and sends relevant early warning information to the administrator. At the same time, the early warning information is output to the evolution feedback layer.

[0129] Specifically, the early warning response layer is connected to the decision control layer data, and the early warning response layer receives the final dynamic threshold from the decision control layer.

[0130] After the early warning response layer receives the final dynamic threshold from the decision control layer, it executes a grading strategy based on the final dynamic threshold. The grading strategy includes blue warning, yellow warning, orange warning and red warning.

[0131] Compare the final dynamic threshold to 1:

[0132] If the final dynamic threshold is between 40% and 60%, a blue warning message will be sent to the administrator, and preventive resource allocation will be performed at the same time. The hot backup server will be automatically connected, and resource allocation will be performed based on the hot backup server. For example, when the server load of the Asia-Pacific node that is currently conducting transactions is large, the hot backup server will be connected to allocate resources and share the load. When the final dynamic threshold is lowered, recovery will be performed.

[0133] If the final dynamic threshold is between 60% and 75%, a yellow warning message will be sent to the administrator, and the digital asset transaction process optimization operation will be executed at the same time. The free transaction links will be allocated based on the spatial dimension to carry out the digital asset transaction process. For example, when the load of the server where the Eastern European node is trading approaches the load limit, a small number of free links of other nodes will be dispatched based on the Kongjia dimension to share the load. When the final dynamic threshold is reduced, recovery will be carried out.

[0134] If the final dynamic threshold is between 75% and 90%, an orange warning message will be sent to the administrator, and digital asset disaster recovery operations will be performed at the same time. High-risk operations in the transaction link will be stopped based on the spatial dimension, transaction links will be scheduled based on the spatial dimension, and backup transaction links will be enabled for digital asset transactions. For example, when the server of the Eastern European node that is currently conducting digital transactions encounters a small-scale network attack, the remaining nodes' free links will be scheduled based on the spatial dimension, and the backup transaction link will be started. The ongoing transactions and subsequent transactions will be migrated to the remaining nodes' free links and backup transaction links. When the final dynamic threshold is lowered, recovery will be performed.

[0135] If the final dynamic threshold is greater than 90% and the digital asset preservation operation is performed at the same time, a red alert message will be sent to the administrator, and the asset preservation operation will be performed at the same time. The digital assets that are not being traded will be immediately frozen based on the spatial dimension. For example, when the server for digital transactions at the Asian node encounters a large-scale network intrusion, the digital transaction link of the node will be immediately stopped, and the data on the digital transaction link where the transaction is stopped at the node will be synchronously transferred to the remaining safe nodes based on the spatial dimension, so that the transaction can continue. For the digital assets that are not being traded at the node, the freezing operation will be immediately performed to ensure the safety of the digital assets of the node until the node returns to normal and the freeze on the digital assets is lifted.

[0136] Furthermore, in actual applications, users can also set up hierarchical strategies based on strategy combinations accordingly, such as combining preventive resource allocation with digital asset trading process optimization operations, so that when the final dynamic threshold is in a certain range of the hierarchical strategy, the early warning response layer will execute preventive resource allocation + digital asset trading process optimization operations, or establish a strategy pool based on machine algorithms, so that each warning level corresponds to 3-5 strategy combinations. When a certain warning level is reached, the machine algorithm will evaluate the strategy combinations in the strategy pool in real time, and select the strategy combination with the highest success rate and the lowest latency for execution. During the execution process, the probability distribution of strategy selection will be updated within the set unit time.

[0137] If the final dynamic threshold is not within the above range, no relevant action is performed.

[0138] When the early warning response layer is running, a visual progress bar is generated. Administrators can directly watch the progress bar to understand the range of the final dynamic threshold. The visual progress bar is rectangular, and each quarter of the progress bar is separated and filled with blue, yellow, orange and red respectively. The final dynamic threshold is displayed at one end of the progress bar.

[0139] S105: Establish an evolutionary feedback layer. The evolutionary feedback layer generates analysis optimization data and decision optimization data based on the past five minutes of slices, warning information and final dynamic thresholds of the four-dimensional heterogeneous cube data, and outputs the analysis optimization data to the intelligent analysis layer and sends the decision optimization data to the decision control layer.

[0140] Specifically, the evolutionary feedback layer is connected to the decision control layer, the execution response layer, and the intelligent perception layer respectively. The evolutionary feedback layer receives the final dynamic threshold from the decision control layer, the past five-minute slices of the four-dimensional heterogeneous cube data of the intelligent perception layer, and the warning information of the warning response layer, and generates analysis optimization data and decision optimization data based on the above information. After generation, the evolutionary feedback layer outputs and imports the analysis optimization data and decision optimization data into the intelligent analysis layer and the decision control layer respectively, thereby performing feedback optimization on the dual-model dynamic coupling architecture of the intelligent analysis layer and the dual-track collaborative threshold mechanism of the decision control layer.

[0141] S106: During the execution of S101 to S105, relevant operating data and warning information are recorded to facilitate maintenance and optimization by the administrator.

[0142] A digital asset load early warning management system based on machine learning, comprising:

[0143] The intelligent perception layer includes three links: native link, market link, and physical link. It is used to capture multi-dimensional heterogeneous data and build four-dimensional heterogeneous cube data based on multi-dimensional heterogeneous data.

[0144] like Figure 4 As shown in the figure, the intelligent perception layer includes native links, market links, and physical links:

[0145] Native link, used to obtain TPS data, smart contract call frequency, gas price volatility and cross-chain transaction latency;

[0146] Market link, used to obtain market data and market sentiment index;

[0147] Physical link, used to obtain server CPU / memory / bandwidth utilization and server environment parameters;

[0148] The intelligent analysis layer is built based on a dual-model dynamic coupling architecture. The dual-model dynamic coupling architecture is coupled by a TCN-LSTM hybrid model and an association analysis model. The intelligent analysis layer generates and exports perception optimization data based on the TCN-LSTM hybrid model and generates and exports warning prediction data based on the association analysis model.

[0149] The decision-making control layer is established based on the dual-track collaborative threshold mechanism. The dual-track collaborative threshold mechanism is used to generate static thresholds and dynamic thresholds. The decision-making control layer generates the final dynamic threshold based on the dual-track collaborative synthesis;

[0150] The early warning response layer is used to execute the hierarchical strategy and send relevant early warning information to the administrator based on the hierarchical strategy;

[0151] The evolutionary feedback layer is used to generate analysis and decision optimization data, thereby providing feedback optimization for the dual-model dynamic coupling architecture of the intelligent analysis layer and the dual-track collaborative threshold mechanism of the decision control layer;

[0152] Hot backup server, used as a backup server for resource scheduling;

[0153] Output execution modules for preventive resource allocation, digital asset transaction process optimization, digital asset disaster recovery, or asset preservation operations;

[0154] The data recording module is used to record relevant operating data and warning information during the execution of steps S101 to S105 to facilitate maintenance and optimization by the administrator.

[0155] like Figure 2As shown, the output end of the intelligent perception layer is unidirectionally connected to the input ends of the intelligent analysis layer, the decision control layer and the evolutionary feedback layer based on the communication link. The output end of the intelligent analysis layer is unidirectionally connected to the input end of the decision control layer and the intelligent perception layer based on the communication link. The intelligent analysis layer and the intelligent perception layer form a closed loop. The output end of the decision control layer is unidirectionally connected to the early warning response layer and the input end of the evolutionary feedback layer based on the communication link. The output end of the early warning response layer is unidirectionally connected to the evolutionary feedback layer and the input end of the output execution module based on the communication link. The output end of the evolutionary feedback layer is unidirectionally connected to the decision control layer and the input end of the intelligent analysis layer based on the communication link. The hot backup server is bidirectionally connected to the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer based on the communication link. The input end of the data recording module is unidirectionally connected to the output end of the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer based on the communication link.

[0156] Furthermore, the administrator can connect the system across regions to establish different transaction nodes. For example, in the Asia-Pacific region, three groups of Asia-Pacific nodes can be established in the Asia-Pacific region, and in Eastern Europe, two groups of Eastern European nodes can be established in Eastern Europe. When one of the Asia-Pacific nodes in the Asia-Pacific region is attacked by a network attack or suffers abnormal behaviors such as server overload, the transaction link of the attacked node in the Asia-Pacific region can be dispatched to the vacant links of the other two groups of Asia-Pacific nodes, or the scheduling can be carried out across regions, thereby improving the real-time performance of digital asset transactions, and ensuring that there are secure links for trading users to conduct normal transactions in the event of an attack, thereby ensuring the security of trading users' digital asset transactions.

[0157] The specific usage and function of this embodiment 1 are as follows:

[0158] First, establish the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer, and connect the data of each layer, obtain multi-dimensional heterogeneous data through the intelligent perception layer, and establish four-dimensional heterogeneous cube data based on the multi-dimensional heterogeneous data. Then, output the four-dimensional heterogeneous cube data to the intelligent analysis layer, evolutionary feedback layer and decision control layer. The intelligent analysis layer generates and exports perception optimization data and prediction and early warning data based on the four-dimensional heterogeneous cube data, and imports the perception optimization data into the intelligent perception layer, so that the intelligent perception layer triggers hierarchical control or capacity expansion contracts, thereby improving the data settlement rate of the intelligent perception layer and increasing the amount of data obtained, thereby improving the accuracy of the prediction and early warning data obtained by the intelligent analysis layer, and at the same time outputs the prediction and early warning data and perception optimization data to the decision control layer. Afterwards, the decision control layer contracts the static threshold and the dynamic threshold based on the dual-track collaborative threshold mechanism. Thus, a final dynamic threshold is generated, and the final dynamic threshold is output to the early warning response layer. The early warning response layer performs a hierarchical strategy based on the final dynamic threshold, thereby performing relevant operations from preventive resource allocation, digital asset transaction process optimization operations, digital asset disaster recovery operations, asset preservation operations or not performing related actions. The administrator can directly understand the range of the final dynamic threshold through a visual progress bar, which improves convenience. While the method process is in progress, analysis optimization data and decision optimization data can be generated based on the evolutionary feedback layer and output to the intelligent analysis layer and the decision control layer respectively, thereby optimizing the decision control layer and the intelligent analysis layer. This method can optimize the resource scheduling of the server load in the ongoing digital transaction based on the hierarchical strategy. When the warning level is high, the digital assets of the trading users can be protected, which improves security. In addition, this method can improve the accuracy of digital asset transaction warnings.

[0159] An electronic device, comprising:

[0160] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method proposed in the first embodiment of the present invention.

[0161] The following is a detailed introduction to the various components of electronic equipment:

[0162] The processor is the control center of an electronic device and can be a single processor or a collective term for multiple processing elements. For example, the processor can be one or more central processing units (CPUs), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the first embodiment of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0163] The processor can execute various functions of the electronic device by running or executing software programs stored in the memory and calling data stored in the memory.

[0164] The memory is used to store the software program for executing the solution of the present invention, and the execution is controlled by the processor. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0165] The memory may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only (CD-ROM), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory may be integrated with the processor or exist independently and be coupled to the processor via an interface circuit of the electronic device, and this is not specifically limited in the embodiments of the present invention.

[0166] The above embodiments can be implemented in whole or in part via software, hardware (e.g., circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wireless communication (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer, or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0167] It should be understood that the term "and / or" as used herein simply describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent the existence of A alone, the existence of both A and B, or the existence of B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the related objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0168] It should be understood that in the embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0169] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A digital asset load early warning management method based on machine learning, characterized in that: The following steps are included: S101: Establish an intelligent perception layer to capture multivariate heterogeneous data. Multivariate heterogeneous data is used to represent the diversified data of digital asset transaction loads. After the multivariate heterogeneous data is captured, denoising is performed and four-dimensional heterogeneous cube data is created based on the multivariate heterogeneous data. S102: Establishing an intelligent analysis layer. The intelligent analysis layer is established based on a dual-model dynamic coupling architecture. Based on the four-dimensional heterogeneous cube data and the dual-model coupling architecture, it generates and exports perception optimization data and prediction and warning data. The intelligent analysis layer and the intelligent perception layer form a closed loop. The perception optimization data is output to the intelligent perception layer for optimization of the intelligent perception layer. The prediction and warning data and perception optimization data are output to the decision control layer. The intelligent analysis layer connects data with the intelligent perception layer, and the intelligent analysis layer receives the four-dimensional heterogeneous cube data output from the intelligent perception layer; The four-dimensional heterogeneous cube data includes time dimension, space dimension, chain layer dimension and physical dimension. The time dimension includes the time information of the data, the space dimension includes the regional location information of the digital asset trading server, the chain layer dimension includes the data information collected by the native link and the market link, and the physical dimension includes the data information collected by the physical link. The dual-model dynamic coupling architecture is coupled by the TCN-LSTM hybrid model and the association analysis model; The TCN-LSTM hybrid model is used to generate perception optimization data. The time, physical, and chain-level dimensions of four-dimensional heterogeneous cube data are fed into the TCN-LSTM hybrid model. The TCN-LSTM hybrid model extracts the past two hours of load time series data from these dimensions. The load time series data consists of 30-second TPS data, CPU / memory / bandwidth utilization, and server environment parameters. Abnormal data in the load time series data is removed and normalized. The TCN-LSTM hybrid model generates and exports perception optimization data based on the load time series data. The perception optimization data represents predicted abnormal transaction operations within the next hour. After the perception optimization data is generated and exported, the intelligent analysis layer outputs the perception optimization data to the intelligent perception layer, and the intelligent perception layer performs optimization based on the perception optimization data; After receiving the data from the intelligent perception layer, the intelligent perception layer performs judgment and comparison operations based on the optimized threshold; If the perception optimization data is less than the optimization threshold, it means that there are fewer abnormal trading operations predicted in the next hour. In this case, the intelligent perception layer will reduce the crawling rate of multi-dimensional heterogeneous data by 0.7, and at the same time partially release the server resources used for data crawling operations. If the perception optimization data is greater than the optimization threshold, it means that there are many abnormal transaction operations predicted in the next hour. In this case, the intelligent perception layer increases the crawling rate of multi-dimensional heterogeneous data. At the same time, the intelligent perception layer dispatches excess server resources for data crawling operations. The association analysis model is used to generate prediction and warning data. It is built based on a hypergraph neural network and imports four-dimensional heterogeneous cube data as input into the association analysis model. The association analysis model then builds a digital asset transaction hypergraph based on the four-dimensional heterogeneous cube data. Based on the digital asset transaction hypergraph, prediction and warning data are generated and exported. After the prediction and warning data is generated and exported, the intelligent analysis layer outputs the prediction and warning data to the decision control layer. S103: Establishing a decision control layer. The decision control layer is established based on a dual-track collaborative threshold mechanism. The dual-track collaborative threshold mechanism includes a static threshold and a dynamic threshold. The static threshold is generated based on four-dimensional heterogeneous cube data, and the dynamic threshold is generated based on the PPO algorithm. The final dynamic threshold is synthesized based on the static and dynamic thresholds. At the same time, the final dynamic threshold is output to the early warning response layer. The data source for the static threshold is the peak CPU / memory / bandwidth utilization in the physical dimension and the peak TPS data in the chain layer dimension in the four-dimensional heterogeneous cube data. Before obtaining the static threshold, an attenuation compensation factor is generated based on the attenuation of hardware performance. The static threshold is calculated based on the attenuation compensation factor and the static threshold data source: ； in, Expressed as a static threshold, 、 、 and Represented as weight factors of CPU / memory / bandwidth utilization and TPS data respectively, Expressed as the peak CPU utilization, Expressed as peak memory utilization, Expressed as the peak bandwidth utilization, Expressed as TPS data peak, Expressed as a constant, Expressed as attenuation compensation factor; The data source for dynamic thresholds is perception optimization data, forecasting and early warning data, and the historical maximum TPS data of the chain layer dimension in the four-dimensional heterogeneous cube data, as well as the market data and market public opinion sentiment index of the chain layer dimension; Before obtaining the dynamic threshold, obtain the market correction factor based on market data and market public opinion sentiment index; Dynamically obtain dynamic thresholds based on market correction factors, PPO algorithm and dynamic threshold data source; Based on the static threshold and dynamic threshold, a dual-track collaborative synthesis is performed to generate the final dynamic threshold, which is then output to the early warning response layer; S104: Establish an early warning response layer. The early warning response layer executes a hierarchical strategy based on the final dynamic threshold, taking actions such as preventive resource allocation, digital asset transaction process optimization, digital asset disaster recovery, asset preservation, or no action, and sends relevant early warning information to the administrator. At the same time, the early warning information is output to the evolution feedback layer. S105: Establish an evolutionary feedback layer. The evolutionary feedback layer generates analysis optimization data and decision optimization data based on the past five minutes of slices, warning information, and final dynamic thresholds of the four-dimensional heterogeneous cube data. The analysis optimization data is output to the intelligent analysis layer, and the decision optimization data is sent to the decision control layer. S106: During the execution of S101 to S105, relevant operating data and warning information are recorded to facilitate maintenance and optimization by the administrator.

2. The digital asset load early warning management method based on machine learning according to claim 1 is characterized in that: S101 specifically includes: The intelligent perception layer includes three links when it is established: native link, market link and physical link; Multi-dimensional heterogeneous data refers to diverse data obtained from different sources, forms, and dimensions. Specifically, multi-dimensional heterogeneous data is obtained by acquiring TPS data, smart contract call frequency, gas price volatility, and cross-chain transaction latency based on native links. This data is used to reflect the real-time status of digital asset transactions. Market data and market sentiment index are obtained based on market links. This data is used to reflect the market environment and the potential impact of market sentiment. Physical links are used to obtain server CPU / memory / bandwidth utilization and server environment parameters. This data is used to reflect the physical carrying capacity boundaries of servers used to process digital asset transactions. After the multivariate heterogeneous data is acquired, data cleaning operations are performed on the multivariate heterogeneous data to ensure the normality of the multivariate heterogeneous data. The multivariate heterogeneous data that has completed data cleaning is then denoised based on the sliding window filtering algorithm to remove short-term noise in the multivariate heterogeneous data. When performing the denoising operation, the window size is dynamically adjusted and the sensitivity coefficient is set to 0.85; When acquiring multivariate heterogeneous data, the time information and regional location information of the multivariate heterogeneous data are acquired simultaneously; Based on the structured integration of multi-dimensional heterogeneous data, time information and regional location information, four-dimensional heterogeneous cube data is established. The four-dimensional heterogeneous cube data includes time dimension, space dimension, chain layer dimension and physical dimension. Based on the four-dimensional heterogeneous cube data, multi-dimensional data fusion is achieved; The intelligent analysis layer outputs the four-dimensional heterogeneous cube data to the intelligent analysis layer, decision control layer and evolution feedback layer respectively.

3. The digital asset load early warning management method based on machine learning according to claim 1 is characterized in that: The intelligent perception layer performs optimization based on perception optimization data, including: When the perception optimization data is greater than the optimization threshold, the intelligent perception layer performs hierarchical control. When the perception optimization data exceeds the optimization threshold by 15%, the crawling rate increases to 1.2 times, and the intelligent perception layer dispatches a small amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 30%, the crawling rate increases to 1.5 times, and the intelligent perception layer dispatches an appropriate amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 50%, the crawling rate increases to 2 times, and the intelligent perception layer dispatches a large amount of excess resources for data crawling operations; when the perception optimization data exceeds the optimization threshold by 80%, the expansion contract and review contract are triggered.

4. The digital asset load early warning management method based on machine learning according to claim 1 is characterized in that: Expansion contracts and review contracts, including: The expansion contract automatically connects the hot backup server, and the intelligent perception layer and intelligent analysis layer schedule the resources of the hot backup server for expansion operation, allocating a maximum of 50% of the hot backup server's resources. At the same time, the operation of the evolutionary feedback layer is stopped, and the resources of the evolutionary feedback layer are dispatched to the intelligent perception layer and intelligent analysis layer to maximize the intelligent perception layer's crawling of multi-heterogeneous data and the intelligent analysis layer's data generation operations. Until the perception optimization data drops to within 50% of the optimization threshold, the expansion contract is stopped, the hot backup server is disconnected, and the operation of the evolutionary feedback layer is resumed. The review contract specifically sends an abnormal alert to the administrator, who will manually review the relevant data in the intelligent analysis layer and the intelligent perception layer. The administrator can manually stop the expansion contract, thereby disconnecting the connected hot backup server and restoring the operation of the evolution feedback layer.

5. The digital asset load early warning management method based on machine learning according to claim 1 is characterized in that: S103 specifically includes: The decision control layer is connected to the intelligent analysis layer and the intelligent perception layer respectively. The decision control layer receives the perception optimization data and prediction and warning data from the intelligent analysis layer, and receives the four-dimensional heterogeneous cube data from the intelligent perception layer. The market correction factor is calculated as: ； in, Expressed as the market correction factor, Represented as market data, Expressed as the market sentiment index, the market correction factor changes dynamically with the market data captured by the intelligent perception layer and the market sentiment index.

6. The digital asset load early warning management method based on machine learning according to claim 1 is characterized in that: S104 specifically includes: The early warning response layer connects with the decision control layer and receives the final dynamic threshold from the decision control layer. After the early warning response layer receives the final dynamic threshold from the decision control layer, it executes the classification strategy based on the final dynamic threshold. The classification strategy includes blue warning, yellow warning, orange warning and red warning. Compare the final dynamic threshold to 1: If the final dynamic threshold is between 40% and 60%, a blue warning message will be sent to the administrator, and preventive resource allocation will be performed at the same time. The hot backup server will be automatically connected and resource allocation will be performed based on the hot backup server. If the final dynamic threshold is between 60% and 75%, a yellow warning message will be sent to the administrator, and the digital asset transaction process will be optimized. The free transaction links will be allocated based on the spatial dimension to carry out the digital asset transaction process. If the final dynamic threshold is between 75% and 90%, an orange warning message will be sent to the administrator, and digital asset disaster recovery operations will be implemented. High-risk operations in the transaction link will be stopped based on the spatial dimension, transaction links will be scheduled based on the spatial dimension, and backup transaction links will be activated for digital asset transactions. If the final dynamic threshold is greater than 90% and digital asset preservation operations are being executed at the same time, a red alert message will be sent to the administrator and asset preservation operations will be executed at the same time, immediately freezing digital assets that are not being traded based on the spatial dimension; If the final dynamic threshold is not within the above range, no relevant action will be performed; When the early warning response layer is running, a visual progress bar is generated. Administrators can directly watch the progress bar to understand the range of the final dynamic threshold. The visual progress bar is rectangular, and each quarter of the progress bar is separated and filled with blue, yellow, orange and red respectively. The final dynamic threshold is displayed at one end of the progress bar.

7. The digital asset load early warning management method based on machine learning according to claim 1 is characterized in that: S105 specifically includes: The evolutionary feedback layer is connected to the decision control layer, the execution response layer, and the intelligent perception layer respectively. The evolutionary feedback layer receives the final dynamic threshold from the decision control layer, the past five-minute slices of the four-dimensional heterogeneous cube data of the intelligent perception layer, and the warning information of the warning response layer, and generates analysis optimization data and decision optimization data based on the above information. After generation, the evolutionary feedback layer outputs and imports the analysis optimization data and decision optimization data into the intelligent analysis layer and the decision control layer respectively, thereby performing feedback optimization on the dual-model dynamic coupling architecture of the intelligent analysis layer and the dual-track collaborative threshold mechanism of the decision control layer.

8. A digital asset load early warning management system based on machine learning, applying any of the methods described in claims 1 to 7, characterized in that: include: The intelligent perception layer includes three links: native link, market link, and physical link. It is used to capture multi-dimensional heterogeneous data and build four-dimensional heterogeneous cube data based on multi-dimensional heterogeneous data. Native link, used to obtain TPS data, smart contract call frequency, gas price volatility and cross-chain transaction latency; Market link, used to obtain market data and market sentiment index; Physical link, used to obtain server CPU / memory / bandwidth utilization and server environment parameters; The intelligent analysis layer is built based on a dual-model dynamic coupling architecture. The dual-model dynamic coupling architecture is coupled by a TCN-LSTM hybrid model and an association analysis model. The intelligent analysis layer generates and exports perception optimization data based on the TCN-LSTM hybrid model and generates and exports warning prediction data based on the association analysis model. The decision-making control layer is established based on the dual-track collaborative threshold mechanism. The dual-track collaborative threshold mechanism is used to generate static thresholds and dynamic thresholds. The decision-making control layer generates the final dynamic threshold based on the dual-track collaborative synthesis; The early warning response layer is used to execute the hierarchical strategy and send relevant early warning information to the administrator based on the hierarchical strategy; The evolutionary feedback layer is used to generate analysis and decision optimization data, thereby providing feedback optimization for the dual-model dynamic coupling architecture of the intelligent analysis layer and the dual-track collaborative threshold mechanism of the decision control layer; Hot backup server, used as a backup server for resource scheduling; Output execution modules for preventive resource allocation, digital asset transaction process optimization, digital asset disaster recovery, or asset preservation operations; The data recording module is used to record relevant operating data and warning information during the execution of steps S101 to S105 to facilitate maintenance and optimization by the administrator.

9. The digital asset load early warning management system based on machine learning according to claim 8 is characterized in that: The output end of the intelligent perception layer is unidirectionally connected to the input ends of the intelligent analysis layer, decision control layer and evolutionary feedback layer based on the communication link. The output end of the intelligent analysis layer is unidirectionally connected to the input end of the decision control layer and the intelligent perception layer based on the communication link. The intelligent analysis layer and the intelligent perception layer form a closed loop. The output end of the decision control layer is unidirectionally connected to the early warning response layer and the input end of the evolutionary feedback layer based on the communication link. The output end of the early warning response layer is unidirectionally connected to the evolutionary feedback layer and the input end of the output execution module based on the communication link. The output end of the evolutionary feedback layer is unidirectionally connected to the input end of the decision control layer and the intelligent analysis layer based on the communication link. The hot backup server is bidirectionally connected to the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer based on the communication link. The input end of the data recording module is unidirectionally connected to the output end of the intelligent perception layer, intelligent analysis layer, decision control layer, early warning response layer and evolutionary feedback layer based on the communication link.

Images