A risk-driven intelligent method for credible traceability of industrial events

Abstract

This invention provides a risk-driven intelligent method for reliable traceability of industrial events, relating to the field of intelligent event traceability technology. The method includes: initial screening of events through risk perception to obtain the risk level of each event; obtaining event identifiers for each event; establishing a deterministic index for the event using a hierarchical weighted Bloom filter, wherein the hierarchical weighted Bloom filter includes M sub-weighted Bloom filters, where M is the total number of risk levels, and the j-th sub-weighted Bloom filter is assigned a target false positive rate; obtaining a hash digest of the event's original data snapshot; obtaining the final root hash and the hash value of the index state snapshot corresponding to the event using a Merkel-Patricia tree; and periodically optimizing the hierarchical weighted Bloom filter. This invention achieves real-time indexing, risk perception, and reliable traceability of massive industrial events.

Description

Technical Field

[0001] This invention relates to the field of intelligent event tracing technology, and more specifically, to a risk-driven intelligent method for reliable tracing of industrial events. Background Technology

[0002] In the field of industrial production and critical infrastructure management, the core challenge of ensuring safe production, achieving efficient operation and maintenance, and compliance auditing is how to quickly identify, accurately index, and establish an immutable chain of event evidence in the face of massive and high-speed sensor data streams.

[0003] Traditional monitoring systems typically rely on centralized databases and static rule-based alerts. These systems are prone to performance bottlenecks during data queries and lack the ability to adapt to fluctuations in system operating conditions. Furthermore, their evidence chains are susceptible to tampering due to centralized storage. These systems often struggle to distinguish truly "immediate hazard" events in real time amidst massive amounts of background noise, leading to response delays or resource misallocation.

[0004] Therefore, it is necessary to optimize the traceability of industrial data to achieve real-time indexing, risk perception, and reliable traceability of massive industrial events. Summary of the Invention

[0005] The purpose of this invention is to provide a risk-driven intelligent method for credible traceability of industrial events, which enables real-time indexing, risk perception, and credible traceability of massive industrial events.

[0006] This invention is achieved through the following technical solution:

[0007] A risk-driven intelligent method for trusted traceability of industrial incidents, comprising:

[0008] Risk perception is used to conduct an initial screening of events and determine their risk level.

[0009] Get the event identifier from the event A deterministic index for events is established using a hierarchical weighted Bloom filter, which comprises M sub-weighted Bloom filters, where M is the total number of risk levels. The j-th sub-weighted Bloom filter is assigned a target false positive rate. The higher the risk level corresponding to risk level j, the higher the risk level. The lower the value;

[0010] Get the hash digest of the raw data snapshot of the event. Obtain the final root hash corresponding to the event through the Merkel-Patricia tree. and index status snapshot The hash value is used to anchor trusted data for events;

[0011] The hierarchical weighted Bloom filter is periodically optimized.

[0012] Preferably, the method for initial event screening through risk perception is as follows:

[0013] At the point of time Data was collected multiple times from N sensors deployed in various locations up to time point t, forming multiple sets of time-series data vectors. , , For time points The Data from one sensor, , Multiple sets of time-series data vectors Forming a timing window ;

[0014] Through pre-trained deep learning models The timing window is processed to map the risk level of the event.

[0015] Preferably, the method uses a pre-trained deep learning model. The method for handling timing windows is as follows:

[0016] Time window This is mapped to an information structure, which includes exception events. and its metadata;

[0017] Based on the risk rule base, each identified abnormal event is... Classified into a discrete and ordered set of severity levels Medium, Severity Level Set This includes Level 1 risk, Level 2 risk, and Level 3 risk, with the risk level decreasing progressively.

[0018] Preferably, the step of obtaining the event identifier for the event... The method is as follows:

[0019] use cryptographic hash algorithm The key metadata of the event is processed to obtain the event identifier. :

[0020] ;

[0021] in, The precise timestamp at which the event was first identified. This is the ordered set of identifiers for the relevant sensors that contribute the most to the event determination. This refers to the stated risk level of the event;

[0022] The method for creating a deterministic index for an event is as follows:

[0023] event identifier Injected into a hierarchical weighted Bloom filter.

[0024] Preferably, the operation method of the layered weighted Bloom filter is as follows:

[0025] Different The values ​​differ exponentially from each other;

[0026] According to the event Risk level , its event identifier Add to the corresponding sub-weighted Bloom filter, the j-th sub-weighted Bloom filter The corresponding risk level is j, where j = 1, 2, ..., M.

[0027] Preferably, the hash digest of the original data snapshot of the event is obtained. The method is as follows:

[0028] Solidification event Original data snapshot and through algorithm Calculate the original data snapshot hash digest :

[0029] ;

[0030] The event event identifier As a key, hash digest As values, they form key-value pairs.

[0031] Preferably, the step of obtaining the final root hash corresponding to the event through a Merkel-Patricia tree is... and index status snapshot The method for obtaining hash values ​​is as follows:

[0032] Insert the key-value pairs into the Merkel-Patricia tree to generate cryptographic proofs for any of the key-value pairs;

[0033] At the end of each preset archiving period, perform the following state anchoring operation:

[0034] Extract the current final root hash of the Merkel-Patricia tree ;

[0035] The bit arrays of all the sub-weighted Bloom filters in the hierarchical weighted Bloom filter are serialized and concatenated to obtain a complete index state snapshot. ,pass algorithm Calculate index status snapshot hash value :

[0036] ;

[0037] Final root hash and index status snapshot hash value Package and release it to a consortium blockchain network.

[0038] Preferably, the method for periodically optimizing the hierarchical weighted Bloom filter is as follows:

[0039] Through the model Continuous monitoring in the past time window Inner Risk Level actual frequency of occurrence The time series forecasting algorithm is used to predict the next time window. Risk Level Number of events , ;

[0040] pass The parameters of the hierarchical weighted Bloom filter are dynamically adjusted.

[0041] Preferably, the passage The method for dynamically adjusting the parameters of the hierarchical weighted Bloom filter is as follows:

[0042] The bit array required to calculate the sub-weighted Bloom filter corresponding to the i-th risk level in the next time window. :

[0043] ;

[0044] Calculate the number of hash functions in the sub-weighted Bloom filter corresponding to the i-th risk level that optimizes query efficiency in the next time window. :

[0045] ;

[0046] Here, round(.) is the rounding function.

[0047] Preferably, when querying or verifying an event, a dual-mode query and cryptographic verification are performed, wherein the method for the dual-mode query and cryptographic verification is as follows:

[0048] Accept requests for event queries or verification;

[0049] Transform the request into a corresponding event identifier. ;

[0050] A pre-check query is performed on the sub-weighted Bloom filters corresponding to the risk level in the hierarchical weighted Bloom filter. If the result is that the event does not exist, the event is excluded and the query is stopped. If the result is that the event has a probability of existing, the re-check process is initiated. The re-check process is as follows:

[0051] With event identifier Using the key, retrieve the original data hash corresponding to the event from the Merkel-Patricia tree. and its corresponding Merkel proof path and with the final root hash Perform a comparison to complete the verification.

[0052] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0053] This invention introduces a prediction-driven adaptive parameter optimization mechanism, which uses the model to predict future event loads and dynamically adjusts the index structure parameters, enabling the system to proactively cope with event storms and always remain in the optimal performance range.

[0054] This invention achieves instant filtering of the vast majority of invalid queries and on-chain finality verification of key events through an efficient dual-mode query and cryptographic verification process.

[0055] This invention eliminates the need for complex backend database reconstruction and enables real-time indexing, risk perception, and reliable traceability of massive industrial events with extremely high resource efficiency.

[0056] This invention is rationally designed and applicable to key fields with stringent requirements for safety, real-time performance, and compliance, such as intelligent manufacturing, power grid dispatching, petrochemicals, and transportation systems, providing solid technical support for achieving intelligent operation and maintenance and digital supervision. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating a risk-driven intelligent method for credible traceability of industrial events, as provided in Embodiment 1 of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0059] Example 1

[0060] This embodiment presents a risk-driven intelligent method for credible traceability of industrial events. (See attached document.) Figure 1 ,include:

[0061] Step S1: Conduct initial screening of events through risk perception to obtain the risk level of the events.

[0062] In this embodiment, the method for initial event screening through risk perception is as follows:

[0063] At the point of time Data was collected multiple times from N sensors deployed in various locations up to time point t, forming multiple sets of time-series data vectors. , , For time points The Data from one sensor, , Multiple sets of time-series data vectors Forming a timing window . This represents the shortest time scale required for pattern recognition. The choice depends on the physical characteristics of the abnormal mode: for slow degradation that requires a cumulative effect to manifest (such as equipment wear). It might take hours or even days; while for transient failures such as sudden pressure changes, a shorter time... (Several seconds or minutes) is enough.

[0064] Through pre-trained deep learning models Process time-series windows to map the risk level of events. Deep learning model. This is an adaptive improvement to existing Long Short-Term Memory (LSTM) networks, which directly outputs a risk level judgment through a two-stage architecture: In the first stage, parallel LSTM units independently extract the temporal features of each sensor; in the second stage, these features are dynamically weighted and fused through an attention mechanism, and then mapped to a preset risk level by the final classifier. This model is designed to identify potential and complex spatiotemporal anomaly patterns in data vectors.

[0065] Therefore, deep learning models In practice, it works continuously over a sliding time window to capture dynamic behavior at different time scales. Its function can be formally described as a mapping function, which is achieved through a pre-trained deep learning model. The method for handling timing windows is as follows:

[0066] Time window This is mapped to an information structure, which includes exception events. and its metadata;

[0067] Based on a risk rule base built using domain expert knowledge, each identified abnormal event... Classified into a discrete and ordered set of severity levels Medium, Severity Level Set This includes Level 1 risk, Level 2 risk, and Level 3 risk, with risk levels decreasing progressively. An example of setting a risk level is as follows:

[0068] Level 1 Risk (L1): Corresponds to "Immediate Hazard" events. These events have the highest priority and may directly lead to catastrophic consequences for safety, the environment, or equipment, requiring a response time in milliseconds or seconds.

[0069] Level 2 Risk (L2): Corresponds to "Performance Decline Level" events. These events significantly impact production efficiency, product quality, or cause economic losses, requiring handling within minutes or hours.

[0070] Level 3 Risk (L3): Corresponds to "Early Warning Level" events. These events are early signals of problems such as equipment degradation and slight energy drift. They do not pose a direct threat in themselves, but have predictive maintenance value and are usually monitored and planned for at the daily or weekly level.

[0071] In other words, the event Risk level satisfy .

[0072] Step S2: Obtain a deterministic, reproducible event identifier for the event. A deterministic index for events is established using a hierarchical weighted Bloom filter, which comprises M sub-weighted Bloom filters, where M is the total number of risk levels. The j-th sub-weighted Bloom filter is assigned a target false positive rate. The higher the risk level corresponding to risk level j, the higher the risk level. The lower the value, the better. A hierarchical weighted Bloom filter is a composite data structure. Here, we assume, as with the three risk levels mentioned earlier, that each level is represented by a set of sub-Bloom filters corresponding to the three risk levels. Composition. Target false positive rate This directly translates business risk into quantifiable query accuracy, which in turn determines the resources (such as memory usage) allocated to each sub-filter. In other words, the higher the accuracy requirement of a given level, the higher the physical specifications of its corresponding sub-filters. An example configuration is as follows:

[0073] ;

[0074] The parameters are set according to the principle that the higher the risk, the stricter the accuracy requirement, and the exponential difference reflects the different treatment of different levels of events in terms of query accuracy.

[0075] Specifically, the step of obtaining the event identifier for the event The method is as follows:

[0076] use cryptographic hash algorithm The key metadata of the event is processed to obtain the event identifier. :

[0077] ;

[0078] in, The precise timestamp at which the event was first identified. This is the ordered set of identifiers for the relevant sensors that contribute the most to the event determination. It refers to the risk level of the event, such as the one mentioned earlier. ;

[0079] The method for creating a deterministic index for an event is as follows:

[0080] event identifier Injected into a hierarchical weighted Bloom filter.

[0081] The working method of the hierarchical weighted Bloom filter is as follows:

[0082] Different The values ​​differ exponentially from each other;

[0083] According to the event Risk level , its event identifier Add to the corresponding sub-weighted Bloom filter, the j-th sub-weighted Bloom filter For risk level j, j=1,2,…,M. This operation stores the "digital fingerprint" of the event in an index subspace whose query precision is strictly matched with the importance of the event itself, thereby achieving differentiated service quality with extremely high resource efficiency.

[0084] Step S3: Obtain the hash digest of the original data snapshot of the event. Obtain the final root hash corresponding to the event through the Merkel-Patricia tree. and index status snapshot The hash value is used to anchor trusted data for events.

[0085] To ensure that every indexed anomalous event has irrefutable original evidence, this step executes a parallel trusted data anchoring process, using Merkel-Patricia trees to provide the cryptographic foundation for subsequent verification phases.

[0086] In this step, firstly, a hash digest of the raw data snapshot of the event is obtained. The method is as follows:

[0087] Solidification event Original data snapshot and through algorithm Calculate the original data snapshot hash digest :

[0088] ;

[0089] The event event identifier As a key, hash digest As a value, it forms a key-value pair. .

[0090] Based on this, the final root hash corresponding to the event is obtained through the Merkel-Patricia tree. and index status snapshot The method for obtaining hash values ​​is as follows:

[0091] Insert the key-value pairs into the Merkel-Patricia tree to generate cryptographic proofs for any of the key-value pairs;

[0092] At the end of each preset archiving period, perform the following state anchoring operation:

[0093] Extract the current final root hash of the Merkel-Patricia tree ;

[0094] The bit arrays of all the sub-weighted Bloom filters in the hierarchical weighted Bloom filter are serialized and concatenated to obtain a complete index state snapshot. ,pass algorithm Calculate index status snapshot hash value :

[0095] ;

[0096] Final root hash and index status snapshot hash value Package and release it to a consortium blockchain network.

[0097] The above steps link the event index and its original evidence to an immutable anchor with a globally consensus timestamp, providing the highest level of integrity and non-repudiation proof for the system state.

[0098] Furthermore, to address the challenge of non-stationarity in industrial system operation, the hierarchical weighted Bloom filter can be periodically optimized. The method for periodically optimizing the hierarchical weighted Bloom filter can be as follows:

[0099] Through the model Continuous monitoring in the past time window Inner Risk Level actual frequency of occurrence Based on this historical data, a time series forecasting algorithm is used to predict the next time window. Risk Level Number of events , This enables a forward-looking prediction of the number of events in the next time window. It was then used to solve a constrained optimization problem, the core objective of which is to strictly satisfy the predetermined false positive rate at each level. Under the premise of minimizing the consumption of system resources.

[0100] pass Dynamically adjusting the parameters of the hierarchical weighted Bloom filter specifically includes:

[0101] The bit array required to calculate the sub-weighted Bloom filter corresponding to the i-th risk level in the next time window. :

[0102] ;

[0103] Calculate the number of hash functions in the sub-weighted Bloom filter corresponding to the i-th risk level that optimizes query efficiency in the next time window. :

[0104] ;

[0105] Here, round(.) is the rounding function.

[0106] The above optimized prediction output is directly converted into the physical space required for the data structure, ensuring that the system can proactively expand to maintain service quality when the number of events surges. At the same time, through this prediction-based periodic parameter update, the hierarchical weighted Bloom filter can actively adapt to the dynamic changes in system operating conditions and always operate on the theoretically optimal performance curve.

[0107] Based on the above scheme, when querying or verifying events, a dual-mode query and cryptographic verification are performed, using the following method:

[0108] Accept requests for event queries or verification;

[0109] Transform the request into a corresponding event identifier. ;

[0110] A pre-check query is performed on the sub-weighted Bloom filters corresponding to the risk level in the hierarchical weighted Bloom filter. Thanks to the mathematical properties of Bloom filters, if the result is that the event does not exist, the system can exclude the event with 100% certainty and immediately terminate the query, thereby greatly filtering the impact of invalid requests on the backend system. If the result is that the event has a probability of existing, a re-check process is initiated. The re-check process is as follows:

[0111] With event identifier Using the key, retrieve the original data hash corresponding to the event from the Merkel-Patricia tree. and its corresponding Merkel proof path and with the final root hash The comparison is performed to complete the final cryptographic verification of the event's authenticity, content integrity, and timestamp.

[0112] It should be noted that, due to the low false positive rate set in step S2 and the continuous optimization of parameters in step four, the probability of entering this re-query stage for critical events such as L1 is controlled at an extremely low level.

[0113] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A risk-driven intelligent method for credible traceability of industrial events, characterized in that, include: Risk perception is used to conduct an initial screening of events and determine their risk level. Get the event identifier from the event A deterministic index for events is established using a hierarchical weighted Bloom filter, which comprises M sub-weighted Bloom filters, where M is the total number of risk levels. The j-th sub-weighted Bloom filter is assigned a target false positive rate. The higher the risk level corresponding to risk level j, the higher the risk level. The lower the value; Get the hash digest of the raw data snapshot of the event. Obtain the final root hash corresponding to the event through the Merkel-Patricia tree. and index status snapshot The hash value is used to anchor trusted data for events; The hierarchical weighted Bloom filter is periodically optimized. The method for periodically optimizing the hierarchical weighted Bloom filter is as follows: Through the model Continuous monitoring in the past time window Inner Risk Level actual frequency of occurrence The time series forecasting algorithm is used to predict the next time window. Risk Level Number of events , ; pass The parameters of the hierarchical weighted Bloom filter are dynamically adjusted; The passage The method for dynamically adjusting the parameters of the hierarchical weighted Bloom filter is as follows: The bit array required to calculate the sub-weighted Bloom filter corresponding to the i-th risk level in the next time window. : ； Calculate the number of hash functions in the sub-weighted Bloom filter corresponding to the i-th risk level that optimizes query efficiency in the next time window. : ； Here, round(.) is the rounding function.

2. The risk-driven intelligent method for credible traceability of industrial events according to claim 1, characterized in that, The method for initial event screening through risk perception is as follows: At the point of time Data was collected multiple times from N sensors deployed in various locations up to time point t, forming multiple sets of time-series data vectors. , , For time points The Data from one sensor, , Multiple sets of time-series data vectors Forming a timing window ; Through pre-trained deep learning models The timing window is processed to map the risk level of the event.

3. The risk-driven intelligent method for credible traceability of industrial events according to claim 2, characterized in that, The pre-trained deep learning model The method for handling timing windows is as follows: Time window This is mapped to an information structure, which includes exception events. and its metadata; Based on the risk rule base, each identified abnormal event is... Classified into a discrete and ordered set of severity levels Medium, Severity Level Set This includes Level 1 risk, Level 2 risk, and Level 3 risk, with the risk level decreasing progressively.

4. The risk-driven intelligent method for credible traceability of industrial events according to claim 1, characterized in that, The event identifier is obtained for the event. The method is as follows: use cryptographic hash algorithm The key metadata of the event is processed to obtain the event identifier. : ； in, The precise timestamp at which the event was first identified. This is the ordered set of identifiers for the relevant sensors that contribute the most to the event determination. This refers to the stated risk level of the event; The method for creating a deterministic index for an event is as follows: event identifier Injected into a hierarchical weighted Bloom filter.

5. The risk-driven intelligent method for credible traceability of industrial events according to claim 4, characterized in that, The working method of the hierarchical weighted Bloom filter is as follows: Different The values ​​differ exponentially from each other; According to the event Risk level , its event identifier Add to the corresponding sub-weighted Bloom filter, the j-th sub-weighted Bloom filter The corresponding risk level is j, where j = 1, 2, ..., M.

6. The risk-driven intelligent method for credible traceability of industrial events according to claim 1, characterized in that, The hash digest of the original data snapshot of the event is obtained. The method is as follows: Solidification event Original data snapshot and through algorithm Calculate the original data snapshot hash digest : ； The event event identifier As a key, hash digest As values, they form key-value pairs.

7. The risk-driven intelligent method for credible traceability of industrial events according to claim 6, characterized in that, The final root hash corresponding to the event is obtained through the Merkel-Patricia tree. and index status snapshot The method for obtaining hash values ​​is as follows: Insert the key-value pairs into the Merkel-Patricia tree to generate cryptographic proofs for any of the key-value pairs; At the end of each preset archiving period, perform the following state anchoring operation: Extract the current final root hash of the Merkel-Patricia tree ; The bit arrays of all the sub-weighted Bloom filters in the hierarchical weighted Bloom filter are serialized and concatenated to obtain a complete index state snapshot. ,pass algorithm Calculate index status snapshot hash value : IndexHash = SM - 3(HWBFstate); Final root hash and index status snapshot hash value Package and release it to a consortium blockchain network.

8. The risk-driven intelligent method for credible traceability of industrial events according to claim 1, characterized in that, When querying or verifying an event, a dual-mode query and cryptographic verification are performed. The method for the dual-mode query and cryptographic verification is as follows: Accept requests for event queries or verification; Transform the request into a corresponding event identifier. ; A pre-check query is performed on the sub-weighted Bloom filters corresponding to the risk level in the hierarchical weighted Bloom filter. If the result is that the event does not exist, the event is excluded and the query is stopped. If the result is that the event has a probability of existing, the re-check process is initiated. The re-check process is as follows: With event identifier Using the key, retrieve the original data hash corresponding to the event from the Merkel-Patricia tree. and its corresponding Merkel proof path and with the final root hash Perform a comparison to complete the verification.

Images