A Safety Risk Assessment and Control Method and System Based on Feedback Correction

Abstract

This invention discloses a safety risk assessment and control method and system based on feedback correction, belonging to the field of safety risk assessment technology. It addresses the problem that historical correction experience cannot adapt to the current risk state during safety risk assessment, leading to increased risk assessment bias and unstable intervention effects. By constructing a matching mechanism for similar events, historical handling cycles are reused and screened. Fusion correction is performed by combining historical correction trajectories with the current assessment model weight state. The correction direction is verified based on post-intervention effect feedback data, achieving dynamic adaptive updating of the risk assessment model. This enhances the matching ability between control intervention and actual risk state, reduces the interference of ineffective historical experience on subsequent risk assessments, and improves the stability and continuous optimization capability of the control system during long-term operation.

Description

Technical Field

[0001] This invention relates to the field of safety risk assessment technology, and more specifically, to a safety risk assessment and control method and system based on feedback correction. Background Technology

[0002] With the widespread application of intelligent monitoring and control systems, safety risk assessment is gradually evolving from fixed-rule judgment to dynamic feedback correction. Existing safety monitoring and control systems typically use sensors to continuously collect equipment operating status data and combine this with risk assessment models to identify and control abnormal events. During long-term system operation, similar risk events will repeatedly occur. Therefore, how to utilize the correction trajectories and feedback results from historical handling processes to dynamically optimize the current risk assessment model has become an important research direction in the field of safety risk assessment.

[0003] The existing technology has the following shortcomings: Currently, existing safety risk assessment systems typically rely solely on current monitoring data to perform risk analysis. They lack a dynamic correlation mechanism between historical correction experience and the current risk status, making it impossible to continuously verify the effectiveness of historical correction trajectories. This can easily lead to the repeated use of ineffective correction experience, causing the assessment model's correction direction to deviate from the actual risk evolution trend. Consequently, this results in a decrease in the accuracy of risk assessment and a reduction in the stability of control interventions. Therefore, a safety risk assessment and control method based on feedback correction is proposed.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a safety risk assessment and control method based on feedback correction. By combining the methods of matching and identifying similar events, reusing and screening historical correction trajectories, fusion and correction of assessment model weights, and feedback verification of intervention effects, a dynamic correlation mechanism between historical handling experience and the current risk status is established to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a safety risk assessment and control method based on feedback correction, comprising the following steps: Step S1: Collect the sensor feature sequence of the current event and extract the feature fingerprint, retrieve the starting features of each historical processing cycle in the historical database, calculate the homology matching degree between the feature fingerprint and each starting feature, and determine whether to enter the homology reuse path. Step S2: Retrieve the correction trajectory file of the same source historical processing cycle in the same source reuse path to obtain the historical correction trajectory and processing result mark, sort the historical correction trajectory according to the processing result mark, and filter the current reuse trajectory; Step S3: Read the weight distribution of the current evaluation model, detect the historical bias direction of the reuse trajectory, generate the weight fusion correction amount and calculate the weight offset by combining the weight distribution of the current evaluation model, and add the weight offset to the current evaluation model to complete the weight update. Step S4: After completing the weight update, perform control intervention and collect effect feedback data. Based on the effect feedback data, verify whether the adjustment direction of the weight fusion correction amount is effective. Based on the verification results, select whether to reverse the weight fusion correction amount. Mark the correction trajectory and treatment results of this treatment cycle and write them back to the historical database.

[0007] In a preferred embodiment, in step S1, the sensor feature sequence is multi-channel time-series data continuously output by each sensor within a preset acquisition time window after the current event is triggered by the measurement and control system; The sensor feature sequences are subjected to multidimensional compression processing. Specifically, the mean, variance, and extreme value ratio are extracted from each channel sequence. The statistical features of all channels are concatenated into a fixed-length vector in channel order to obtain the feature fingerprint.

[0008] In a preferred embodiment, in step S1, the starting feature of the historical processing cycle is the starting feature vector of the corresponding historical processing cycle, which is read and called one by one by accessing the historical database. Calculate the cosine similarity between the feature fingerprint and the starting feature vector of each historical processing cycle to obtain the homology matching degree corresponding to each historical record; Take the maximum value among all homology matching scores and compare it with the preset homology matching confidence threshold: When the maximum homogeneity matching degree is greater than or equal to the preset homogeneity matching confidence threshold, the historical record index of the corresponding historical processing period is recorded, and it is determined to enter the homogeneity reuse path. The historical record index is then passed to step S2. When the maximum homology matching degree is less than the preset homology matching confidence threshold, it is determined that the homology reuse path will not be entered. Among them, the historical record index is a unique identifier for a specific historical processing period in the historical database.

[0009] In a preferred embodiment, in step S2, the historical correction trajectories of the same origin are retrieved according to the historical record index and sorted and filtered according to the treatment result label; Based on the historical record index, retrieve all correction trajectories from the correction trajectory archive database of the historical database for all historical processing cycles of the same origin; The correction trajectory is the complete adjustment path of the evaluation model weights from the initial state to the final convergence state during a historical treatment cycle, including the adjustment direction and magnitude of each dimension in each correction step.

[0010] In a preferred embodiment, in step S2, the disposal result mark corresponding to the same historical disposal cycle is retrieved from the disposal result mark database of the historical database. The disposal result mark refers to the disposal effect evaluation written into the historical database after the historical disposal cycle is completed. Convert the treatment outcome labels into numerical scores: The highest score is marked as successful handling and no recurrence of similar events within the preset review period; Successful treatment but recurrence of the same event within the preset review period is marked as neutral. Those interventions that did not achieve the expected results were marked with low scores; The historical correction trajectories of the same origin are sorted in descending order according to their numerical scores, and the correction trajectory ranked first is selected as the trajectory to be reused this time.

[0011] In a preferred embodiment, in step S3, the historical bias direction vectors of each weight dimension are extracted from the current reuse trajectory. ,in The i-th component Hi represents the cumulative adjustment direction of the i-th dimension weight in the historical processing period. A positive value indicates that the weight of this dimension is adjusted to increase as a whole during the historical correction process, while a negative value indicates that it is adjusted to decrease. right After L2 normalization, the historical bias direction unit vector is obtained. ; Read the current weight distribution vector from the evaluation model device. , This represents the current weight value for the i-th dimension. Calculate the parameters for each dimension sequentially using the following formulas: The formula for calculating the directional difference coefficient is as follows: in, Let be the difference coefficient in the i-th dimension, and sign be the sign function. for The i-th dimension component, Let i be the i-th component of the current weight distribution vector; A value of 0 indicates that the historical bias direction is in the same direction as the current weight, and the correction amount is retained; A value of 1 indicates the reverse direction, and the correction value is set to zero.

[0012] In a preferred embodiment, the formula for calculating the weight fusion correction in step S3 is: ; in, The weight fusion correction for the i-th dimension is expressed as the directional consistency coefficient ( The weights map the historical bias direction to the current correction amount; when the two are in the same direction... equal When the two are in opposite directions Zero; The formula for calculating the weight offset is: ; in, Let i be the weight offset in the i-th dimension. Preset fusion step size; The weight update formula is: ; in, The updated weight value for the i-th dimension; The updated weight values ​​for each dimension are written into the evaluation model device to complete this round of weight update.

[0013] In a preferred embodiment, in step S4, after the weight update is completed, the measurement and control system outputs control commands and performs control intervention based on the updated evaluation model. The effect feedback data is the sensor output data collected by the feedback sensor device within the preset feedback observation time after the control intervention is performed. The effect feedback data is collected from the feedback sensor device within a preset feedback observation period, and the value of the effect feedback data is used as the intervention effect quantity. The sign direction of the intervention effect quantity is compared with the principal direction of the weighted fusion correction quantity: the principal direction of the weighted fusion correction quantity is determined by the sign of the largest absolute value of the fusion correction quantity for each dimension. When the intervention effect is negative and the main direction of the weight fusion correction is to increase the high-risk weight dimension, the verification result is determined to be effective in the direction. Conversely, if the intervention effect is not negative or the main direction of the weighted fusion correction is inconsistent with the direction of the intervention effect, the verification result is determined to be invalid in terms of direction.

[0014] In a preferred embodiment, in step S4, when the verification result is that the direction is valid, the processing result corresponding to the current processing cycle is marked as a high confidence marker, the correction trajectory and processing result marker of the current processing cycle are packaged and written back to the historical database, and the event of the high confidence marker is set as processing successful and without recurrence. When the verification result is invalid in the direction, the weight fusion correction amount is inverted dimension by dimension to generate the reverse correction amount. The reverse correction amount is multiplied by the preset fusion step size to obtain the reverse weight offset amount, which is then superimposed on the current weight distribution to complete the re-update. At the same time, the processing result corresponding to this processing cycle will be marked as a low-confidence marker in the historical database. The low-confidence marker will be downgraded in the subsequent same-source matching and sorting.

[0015] A security risk assessment and control system based on feedback correction includes a homogeneous matching and identification module, a reuse trajectory screening module, a weight fusion correction module, and a feedback archiving and verification module. The homology matching and identification module is used to collect the sensor feature sequence of the current event and extract the feature fingerprint, retrieve the starting features of each processing cycle in the historical database, calculate the homology matching degree between the feature fingerprint and each starting feature, and determine whether to enter the homology reuse path. The reuse trajectory filtering module is used to retrieve the correction trajectory files of the same source historical processing cycle in the same source reuse path, obtain the historical correction trajectory and processing result mark, sort the historical correction trajectory according to the processing result mark, and filter the current reuse trajectory. The weight fusion correction module is used to read the weight distribution of the current evaluation model, detect the historical bias direction of the current reuse trajectory, generate the weight fusion correction amount and calculate the weight offset by combining the weight distribution of the current evaluation model, and then add the weight offset to the current evaluation model to complete the weight update. The feedback archiving and verification module is used to perform control intervention and collect effect feedback data after the weight update is completed. Based on the effect feedback data, it verifies whether the adjustment direction of the weight fusion correction amount is effective. Based on the verification results, it selects whether to reverse the weight fusion correction amount and writes the correction trajectory and the handling results of this handling cycle back to the historical database.

[0016] The technical effects and advantages of this invention are as follows: This invention establishes a dynamic correlation between historical correction trajectories and the current risk status, filters and reuses historical handling experience, and combines the feedback of the intervention effect to conduct closed-loop verification of the correction direction of the assessment model, thereby achieving continuous optimization of the risk assessment model, improving the accuracy of risk assessment, enhancing the stability of control intervention, and reducing the probability of ineffective correction experience repeatedly participating in model correction. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the implementation of a safety risk assessment and control method based on feedback correction according to the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the steps of a safety risk assessment and control method based on feedback correction according to the present invention.

[0019] Figure 3 This is a module framework diagram of a safety risk assessment and control system based on feedback correction according to the present invention. Detailed Implementation

[0020] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention enables the risk assessment model to dynamically adjust its correction direction based on actual intervention results by performing source matching and feedback verification on the correction trajectory in the historical handling process. This improves the matching ability between risk identification and control intervention, and enhances the model's adaptive capability during the long-term operation of the safety measurement and control system. Example 1

[0022] Please see Figures 1 to 2 A safety risk assessment and control method based on feedback correction, the specific operation process is as follows: Step S1: Collect the sensor feature sequence of the current event and extract the feature fingerprint, retrieve the starting features of each historical processing cycle in the historical database, calculate the homology matching degree between the feature fingerprint and each starting feature, and determine whether to enter the homology reuse path. Step S2: Retrieve the correction trajectory file of the same source historical processing cycle in the same source reuse path to obtain the historical correction trajectory and processing result mark, sort the historical correction trajectory according to the processing result mark, and filter the current reuse trajectory; Step S3: Read the weight distribution of the current evaluation model, detect the historical bias direction of the reuse trajectory, generate the weight fusion correction amount and calculate the weight offset by combining the weight distribution of the current evaluation model, and add the weight offset to the current evaluation model to complete the weight update. Step S4: After completing the weight update, perform control intervention and collect effect feedback data. Based on the effect feedback data, verify whether the adjustment direction of the weight fusion correction amount is effective. Based on the verification results, select whether to reverse the weight fusion correction amount. Mark the correction trajectory and treatment results of this treatment cycle and write them back to the historical database.

[0023] The specific implementation is as follows: In step S1, the sensor feature sequence of the current event is collected and matched with the starting features of the historical processing cycle to determine whether to enter the same source reuse path.

[0024] The sensor feature sequence is multi-channel time-series data continuously output by each sensor within a preset acquisition time window after the current event is triggered by the measurement and control system, reflecting the real-time operating status of the controlled equipment when the event occurs; It should be explained that, in this example, the event refers to an abnormal state instance in which the operating parameters of the controlled equipment exceed the preset safety boundary during the continuous monitoring process of the measurement and control system, triggering the assessment model to initiate the current round of handling procedures; the measurement and control system refers to a closed-loop control system that integrates three functional modules: sensor data acquisition, risk assessment model calculation, and control command output. In this example, it is used to continuously monitor the operating status of the controlled equipment, quantify risks, and execute interventions.

[0025] Multidimensional compression processing is performed on the sensor feature sequences. Specifically, three types of statistical features, namely mean, variance, and extreme value ratio, are extracted from each channel sequence. The mean reflects the overall amplitude level of the channel signal, the variance reflects the fluctuation intensity of the signal, and the extreme value ratio is the ratio of the maximum value to the minimum value of the channel, reflecting the dynamic range of the signal. The statistical features of all channels are concatenated into a fixed-length vector in channel order to obtain the feature fingerprint. The feature fingerprint is a compressed representation of the current event in the multi-dimensional sensor space, used for similarity comparison with the starting features of historical processing cycles.

[0026] The starting feature of the historical processing cycle is the starting feature vector of the corresponding historical processing cycle. This is obtained by accessing the historical database one by one. The cosine similarity between the feature fingerprint and the starting feature vector of each historical processing cycle is calculated to obtain the homology matching degree corresponding to each historical record.

[0027] It should be explained that the historical database refers to a structured data set established during the operation of the measurement and control system, which stores the starting feature vector, correction trajectory and handling result mark of each historical handling cycle using the handling cycle identifier as an index; the homogeneity matching degree reflects the degree of proximity between the current event and a certain historical handling cycle in the sensor feature space, and the higher the value, the more likely the two are to belong to the same type of event.

[0028] Take the maximum value among all homology matching scores and compare it with the preset homology matching confidence threshold; When the maximum homogeneity matching degree is greater than or equal to the preset homogeneity matching confidence threshold, the historical record index of the corresponding historical processing period is recorded, and it is determined to enter the homogeneity reuse path. The historical record index is then passed to step S2. When the maximum homology matching degree is less than the preset homology matching confidence threshold, it is determined that the current event and the historical processing cycle do not have a homology relationship, and the normal correction logic is returned without proceeding to the subsequent steps.

[0029] The historical record index is a unique identifier for a specific historical treatment period in the historical database. It is used to accurately retrieve the correction trajectory and treatment result marker corresponding to that treatment period in subsequent steps.

[0030] Through the above feature extraction and homogeneous matching processing, the similarity between the current event and the historical handling cycle is quantitatively judged, providing a historical record index for step S2, and supporting the retrieval of the historical correction trajectory of the subsequent homogeneous reuse path.

[0031] It should be noted that the preset confidence threshold for same-source matching can be set according to the size of the historical processing database and the distribution of event types. Specifically, after the equipment is put into operation, the cosine similarity of each archived historical processing cycle is calculated, and the lower limit of the similarity distribution of known similar event pairs is statistically analyzed. The 90th quantile of this lower limit is taken as the preset confidence threshold for same-source matching to ensure that same-source and non-same-source event pairs can be effectively distinguished when there is sufficient historical data. The methods for extracting statistical features include, but are not limited to, time-domain statistics, frequency-domain statistics, or time-frequency joint analysis, which will not be elaborated here.

[0032] In step S2, within the same-source reuse path, the same-source historical correction trajectory is retrieved based on the historical record index, and sorted and filtered according to the processing result mark to determine the current reuse trajectory.

[0033] The treatment outcome marker refers to the evaluation of treatment effectiveness written into the historical database after the completion of a historical treatment cycle. It is used to record whether the control intervention achieved the expected risk elimination effect during the treatment cycle and whether a relapse occurred subsequently.

[0034] The correction trajectory is the complete adjustment path of the evaluation model weights from the initial state to the final convergence state during a historical processing cycle, including the adjustment direction and magnitude of each dimension in each correction step.

[0035] Based on the historical record index, the correction trajectories for all corresponding historical processing cycles are retrieved from the correction trajectory archive database of the historical database; at the same time, the processing result tags corresponding to the above-mentioned corresponding historical processing cycles are retrieved from the processing result tag database of the historical database.

[0036] Convert the treatment outcome labels into numerical scores: The highest score is marked as successful handling and no recurrence of similar events within the preset review period; Successful treatment but recurrence of the same event within the preset review period is marked as neutral. Those interventions that fail to achieve the expected results are marked with a low score.

[0037] The historical correction trajectories of the same origin are sorted in descending order according to their numerical scores, and the correction trajectory ranked first is selected as the trajectory to be reused this time.

[0038] It should be noted that both the trajectory correction archive database and the handling result marker database are components of the historical database, and are linked and indexed using the handling cycle identifier as the primary key; the preset review cycle can be set according to the typical fault interval of the equipment.

[0039] In step S3, the historical bias direction of the reused trajectory is adversarially fused with the weight distribution of the current evaluation model to generate the weight fusion correction and weight offset, thus completing the weight update for this round.

[0040] Extract the historical bias direction vectors of each weight dimension from the reused trajectory. ,in The i-th component Hi represents the cumulative adjustment direction of the i-th dimension weight over the historical processing period. A positive value indicates that the weight of this dimension is adjusted towards increasing during the historical correction process, while a negative value indicates that it is adjusted towards decreasing. L2 normalization yields the historical bias direction unit vector. The calculation formula is: ; in, for The L2 norm.

[0041] Read the current weight distribution vector from the evaluation model device. , The current weight value for the i-th dimension; the evaluation model device refers to the device that stores the weight parameters of each dimension of the current security risk evaluation model.

[0042] Calculate the parameters for each dimension sequentially using the following formulas: The formula for calculating the directional difference coefficient is as follows: in, Let be the difference coefficient in the i-th dimension, and sign be the sign function. for The i-th dimension component, Let i be the i-th component of the current weight distribution vector; A value of 0 indicates that the historical bias direction is in the same direction as the current weight, and the correction amount is retained; A value of 1 indicates the reverse direction, and the correction value is set to zero; The formula for calculating the weight fusion correction is: ; in, The weight fusion correction for the i-th dimension is expressed as the directional consistency coefficient ( The weights map the historical bias direction to the current correction amount; when the two are in the same direction... equal When the two are in opposite directions The value is zero to avoid forcibly applying historical biases that are opposite to the current weight state; The formula for calculating the weight offset is: ; in, Let i be the weight offset in the i-th dimension. Preset fusion step size; The weight update formula is: ; in, Let be the updated weight value for the i-th dimension; write the updated weight values ​​for each dimension into the evaluation model device to complete this round of weight update.

[0043] The weight fusion correction and weight offset are used in subsequent steps to verify the effectiveness of the adjustment direction and to perform reverse correction.

[0044] Through the aforementioned adversarial fusion calculation, the bias direction of the historical optimal correction trajectory is combined with the current weight state to generate the weight fusion correction amount and weight offset amount, thereby realizing the effective transfer of historical correction experience to the current evaluation model and providing a correction benchmark for the effect feedback verification of subsequent steps.

[0045] It should be noted that the preset fusion step size can be set according to the statistical average of the weight correction magnitude of the same source events in the historical processing data. The larger the value, the stronger the influence of the historical correction trajectory on the current weight update. L2 normalization processing eliminates the interference of the difference in the absolute value of the correction magnitude in different historical processing cycles on the direction comparison, so that the fusion calculation depends only on the direction information.

[0046] In step S4, after the weight update is completed, the measurement and control system outputs control commands and executes control intervention based on the updated evaluation model, collects feedback data on the effect of the control intervention, verifies whether the adjustment direction of the weight fusion correction is effective, and completes file archiving or reverse correction based on the verification results. The effect feedback data is the sensor output data collected by the feedback sensor device within a preset feedback observation period after the control intervention is implemented. The sensor output data is the change in the risk index of the controlled equipment under the intervention.

[0047] It should be explained that control intervention refers to the operational instructions issued by the measurement and control system to the controlled equipment to reduce the current risk, including but not limited to parameter limit adjustment, equipment load reduction, or partial shutdown; feedback sensor devices are sensor modules in the measurement and control system used to collect the effect of intervention, and in this example, they are used to collect feedback data on the effect after the control intervention is executed.

[0048] The effect feedback data is collected from the feedback sensor device within a preset feedback observation period, and the value of the effect feedback data is used as the intervention effect quantity. The intervention effect measure reflects the actual degree to which the control intervention reduces the risk status of the controlled equipment. When the intervention effect measure is negative, it indicates that the risk indicator decreased after the intervention, and the intervention was effective. When the intervention effect is positive or close to zero, it indicates that the risk indicators have not decreased significantly and the intervention effect has not met expectations.

[0049] The sign direction of the intervention effect quantity is compared with the principal direction of the weighted fusion correction quantity: the principal direction of the weighted fusion correction quantity is determined by the sign of the largest absolute value of the fusion correction quantity for each dimension. When the intervention effect is negative and the main direction of the weight fusion correction is to increase the high-risk weight dimension, the verification result is determined to be effective in the direction. Conversely, if the intervention effect is not negative or the main direction of the weighted fusion correction is inconsistent with the direction of the intervention effect, the verification result is determined to be invalid in terms of direction.

[0050] When the verification result is that the direction is valid, the processing result corresponding to this processing cycle is marked as a high-confidence marker, the correction trajectory and processing result marker of this processing cycle are packaged and written back to the historical database, and the events of the high-confidence markers are set as successfully processed and without recurrence, so that step S2 can sort and filter them when similar events recur in the future, and complete the archive archiving. When the verification result is invalid in the direction, the weight fusion correction amount is inverted dimension by dimension to generate the reverse correction amount. The reverse correction amount is multiplied by the preset fusion step size to obtain the reverse weight offset amount, which is then superimposed on the current weight distribution to complete the re-update. At the same time, the processing result corresponding to this processing cycle will be marked as a low-confidence marker in the historical database. The low-confidence marker will be downweighted in the subsequent same-source matching and sorting, reducing the probability that it will be selected as the current reuse trajectory in subsequent similar events.

[0051] Through the above-mentioned feedback verification and archiving process, a closed-loop evaluation of the effectiveness of this source reuse is achieved. Effective handling experience is dynamically written into the historical database, and invalid records are marked with reduced weight to ensure that the historical correction trajectory archive is continuously optimized in long-term operation, providing a higher quality reuse basis for source matching of similar events in the future.

[0052] It should be noted that the preset feedback observation duration can be set according to the dynamic response time constant of the controlled equipment, taking the typical response time from the issuance of the intervention command to the stabilization of the state; the weighting method for low confidence targets is as follows: during the same source matching sorting in step S1, the same source matching degree of the historical records marked as low confidence targets is multiplied by a preset weighting coefficient (the value range is 0 to 1), so that it is moved to the back in the sorting. The preset weighting coefficient can be set according to the misjudgment rate statistics of historical low confidence records.

[0053] Example 2, please refer to Figure 3 A security risk assessment and control system based on feedback correction includes a homogeneous matching identification module, a reuse trajectory screening module, a weight fusion correction module, and a feedback archiving verification module. The homology matching and identification module is used to collect the sensor feature sequence of the current event and extract the feature fingerprint, retrieve the starting features of each processing cycle in the historical database, calculate the homology matching degree between the feature fingerprint and each starting feature, and determine whether to enter the homology reuse path. The reuse trajectory filtering module is used to retrieve the correction trajectory files of the same source historical processing cycle in the same source reuse path, obtain the historical correction trajectory and processing result mark, sort the historical correction trajectory according to the processing result mark, and filter the current reuse trajectory. The weight fusion correction module is used to read the weight distribution of the current evaluation model, detect the historical bias direction of the current reuse trajectory, generate the weight fusion correction amount and calculate the weight offset by combining the weight distribution of the current evaluation model, and then add the weight offset to the current evaluation model to complete the weight update. The feedback archiving and verification module is used to perform control intervention and collect effect feedback data after the weight update is completed. Based on the effect feedback data, it verifies whether the adjustment direction of the weight fusion correction amount is effective. Based on the verification results, it selects whether to reverse the weight fusion correction amount and writes the correction trajectory and the handling results of this handling cycle back to the historical database.

[0054] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0055] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0056] In this document, the singular forms “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that terms such as “comprising / including” or “having” specify the presence of the stated features, integrals, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, integrals, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0057] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0058] The above description of the disclosed embodiments will enable those skilled in the art to make or use various modifications to these embodiments. It will be readily apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A safety risk assessment and control method based on feedback correction, characterized in that: Includes the following steps: Step S1: Collect the sensor feature sequence of the current event and extract the feature fingerprint, retrieve the starting features of each historical processing cycle in the historical database, calculate the homology matching degree between the feature fingerprint and each starting feature, and determine whether to enter the homology reuse path. Step S2: Retrieve the correction trajectory file of the same source historical processing cycle in the same source reuse path to obtain the historical correction trajectory and processing result mark, sort the historical correction trajectory according to the processing result mark, and filter the current reuse trajectory; Step S3: Read the weight distribution of the current evaluation model, detect the historical bias direction of the reuse trajectory, generate the weight fusion correction amount and calculate the weight offset by combining the weight distribution of the current evaluation model, and add the weight offset to the current evaluation model to complete the weight update. Step S4: After completing the weight update, perform control intervention and collect effect feedback data. Based on the effect feedback data, verify whether the adjustment direction of the weight fusion correction amount is effective. Based on the verification results, select whether to reverse the weight fusion correction amount. Mark the correction trajectory and treatment results of this treatment cycle and write them back to the historical database.

2. The safety risk assessment and control method based on feedback correction according to claim 1, characterized in that: In step S1, the sensor feature sequence is the multi-channel time-series data continuously output by each sensor within a preset acquisition time window after the current event is triggered by the measurement and control system; The sensor feature sequences are subjected to multidimensional compression processing. Specifically, the mean, variance, and extreme value ratio are extracted from each channel sequence. The statistical features of all channels are concatenated into a fixed-length vector in channel order to obtain the feature fingerprint.

3. The safety risk assessment and control method based on feedback correction according to claim 1, characterized in that: In step S1, the starting feature of the historical processing cycle is the starting feature vector of the corresponding historical processing cycle, which is read and called one by one by accessing the historical database; Calculate the cosine similarity between the feature fingerprint and the starting feature vector of each historical processing cycle to obtain the homology matching degree corresponding to each historical record; Take the maximum value among all homology matching scores and compare it with the preset homology matching confidence threshold: When the maximum homogeneity matching degree is greater than or equal to the preset homogeneity matching confidence threshold, the historical record index of the corresponding historical processing period is recorded, and it is determined to enter the homogeneity reuse path. The historical record index is then passed to step S2. When the maximum homology matching degree is less than the preset homology matching confidence threshold, it is determined that the homology reuse path will not be entered. Among them, the historical record index is a unique identifier for a specific historical processing period in the historical database.

4. The safety risk assessment and control method based on feedback correction according to claim 3, characterized in that: In step S2, the historical correction trajectories of the same source are retrieved according to the historical record index and sorted and filtered according to the processing result label; Based on the historical record index, retrieve all correction trajectories from the correction trajectory archive database of the historical database for all historical processing cycles of the same origin; The correction trajectory is the complete adjustment path of the evaluation model weights from the initial state to the final convergence state during a historical treatment cycle, including the adjustment direction and magnitude of each dimension in each correction step.

5. The safety risk assessment and control method based on feedback correction according to claim 4, characterized in that: In step S2, the disposal result tags corresponding to the same historical disposal cycle are retrieved from the disposal result tag database of the historical database. The disposal result tag refers to the disposal effect evaluation written into the historical database after the historical disposal cycle is completed. Convert the treatment outcome labels into numerical scores: The highest score is marked as successful handling and no recurrence of similar events within the preset review period; Successful treatment but recurrence of the same event within the preset review period is marked as neutral. Those interventions that did not achieve the expected results were marked with low scores; The historical correction trajectories of the same origin are sorted in descending order according to their numerical scores, and the correction trajectory ranked first is selected as the trajectory to be reused this time.

6. The safety risk assessment and control method based on feedback correction according to claim 5, characterized in that: In step S3, the historical bias direction vectors of each weight dimension are extracted from the current reuse trajectory. ,in The i-th component Hi represents the cumulative adjustment direction of the i-th dimension weight in the historical processing period. A positive value indicates that the weight of this dimension is adjusted to increase as a whole during the historical correction process, while a negative value indicates that it is adjusted to decrease. right After L2 normalization, the historical bias direction unit vector is obtained. ; Read the current weight distribution vector from the evaluation model device. , This represents the current weight value for the i-th dimension. Calculate the parameters for each dimension sequentially using the following formulas: The formula for calculating the directional difference coefficient is as follows: in, Let be the difference coefficient in the i-th dimension, and sign be the sign function. for The i-th dimension component, Let i be the i-th component of the current weight distribution vector; A value of 0 indicates that the historical bias direction is in the same direction as the current weight, and the correction amount is retained; A value of 1 indicates the reverse direction, and the correction value is set to zero.

7. The safety risk assessment and control method based on feedback correction according to claim 6, characterized in that: In step S3, the formula for calculating the weight fusion correction is: ； in, Let i be the weight fusion correction amount for the i-th dimension, and let y be the directional consistency coefficient ( The weights map the historical bias direction to the current correction amount; when the two are in the same direction... equal When the two are in opposite directions Zero; The formula for calculating the weight offset is: ； in, Let i be the weight offset in the i-th dimension. Preset fusion step size; The weight update formula is: ； in, The updated weight value for the i-th dimension; The updated weight values ​​for each dimension are written into the evaluation model device to complete this round of weight update.

8. The safety risk assessment and control method based on feedback correction according to claim 7, characterized in that: In step S4, after the weight update is completed, the measurement and control system outputs control commands and executes control intervention based on the updated evaluation model. The effect feedback data is the sensor output data collected by the feedback sensor device within the preset feedback observation time after the control intervention is executed. The effect feedback data is collected from the feedback sensor device within a preset feedback observation period, and the value of the effect feedback data is used as the intervention effect quantity. The sign direction of the intervention effect quantity is compared with the principal direction of the weighted fusion correction quantity: the principal direction of the weighted fusion correction quantity is determined by the sign of the largest absolute value of the fusion correction quantity for each dimension. When the intervention effect is negative and the main direction of the weight fusion correction is to increase the high-risk weight dimension, the verification result is determined to be effective in the direction. Conversely, if the intervention effect is not negative or the main direction of the weighted fusion correction is inconsistent with the direction of the intervention effect, the verification result is determined to be invalid in terms of direction.

9. The safety risk assessment and control method based on feedback correction according to claim 8, characterized in that: In step S4, when the verification result is that the direction is valid, the processing result corresponding to this processing cycle is marked as a high confidence marker, the correction trajectory of this processing cycle and the processing result marker are packaged and written back to the historical database, and the event of the high confidence marker is set as processing successful and without recurrence; When the verification result is invalid in the direction, the weight fusion correction amount is inverted dimension by dimension to generate the reverse correction amount. The reverse correction amount is multiplied by the preset fusion step size to obtain the reverse weight offset amount, which is then superimposed on the current weight distribution to complete the re-update. At the same time, the processing result corresponding to this processing cycle will be marked as a low-confidence marker in the historical database. The low-confidence marker will be downweighted in the subsequent same-source matching and sorting.

10. A security risk assessment and control system based on feedback correction, used to implement the security risk assessment and control method based on feedback correction as described in any one of claims 1-9, characterized in that: It includes a homogeneous matching and identification module, a reuse trajectory filtering module, a weight fusion and correction module, and a feedback archiving and verification module; The homology matching and identification module is used to collect the sensor feature sequence of the current event and extract the feature fingerprint, retrieve the starting features of each processing cycle in the historical database, calculate the homology matching degree between the feature fingerprint and each starting feature, and determine whether to enter the homology reuse path. The reuse trajectory filtering module is used to retrieve the correction trajectory files of the same source historical processing cycle in the same source reuse path, obtain the historical correction trajectory and processing result mark, sort the historical correction trajectory according to the processing result mark, and filter the current reuse trajectory. The weight fusion correction module is used to read the weight distribution of the current evaluation model, detect the historical bias direction of the current reuse trajectory, generate the weight fusion correction amount and calculate the weight offset by combining the weight distribution of the current evaluation model, and then add the weight offset to the current evaluation model to complete the weight update. The feedback archiving and verification module is used to perform control intervention and collect effect feedback data after the weight update is completed. Based on the effect feedback data, it verifies whether the adjustment direction of the weight fusion correction amount is effective. Based on the verification results, it selects whether to reverse the weight fusion correction amount and writes the correction trajectory and the handling results of this handling cycle back to the historical database.

Images