Safety production integrated service platform

Abstract

This invention discloses an integrated service platform for safe production, relating to the field of information technology in safe production. It addresses the problem of inconsistent organization and continuous correction of multi-source business records during aggregation and processing, leading to inaccurate risk status identification, large fluctuations in early warning results, and a lack of unified quantitative basis for subsequent closed-loop handling. The platform achieves continuous identification, dynamic correction, and closed-loop control of risk status by unifying and archiving data, constructing status chains, implementing closed-loop risk assessment, and performing playback correction and updates on enterprise-side attendance records, inspection task records, hazard reporting records, handling result records, and review confirmation records. This includes unified transcription, risk object coding, record quality assessment, status credibility calculation, risk quantification and grading, early warning task binding, and playback recalculation based on review results. This improves the consistency of risk assessment, the traceability of results, and the stability of business closed-loop processing.

Description

Technical Field

[0001] This invention relates to the field of information technology for safe production, and more specifically, to an integrated service platform for safe production. Background Technology

[0002] Existing safety production information platforms typically need to access information from enterprise mobile applications, inspection terminals, hazard reporting terminals, regulatory review terminals, and external business platforms. Based on the access records, they display, remind, and track the enterprise's performance of its responsibilities, inspection completion status, hazard handling progress, and review results. Most existing platforms can complete basic data aggregation, task flow, and message push. However, in actual operation, there are significant differences in the format, arrival order, evidence completeness, and business semantics of records from different sources. The same on-site action may also be repeatedly transmitted or supplemented by different systems at different times, which makes it easy for the platform to mix and use conflicting records when identifying the current risk status.

[0003] The existing technology has the following shortcomings: Existing technologies generally adopt a processing method that directly maps a single record to the current state. This lacks a complete mechanism for unified coding, temporal correction, record quality assessment, state chain construction, and subsequent playback correction around the same risk object. When there are late entries, duplicate write-backs, cross-platform transmissions, or re-examinations for counter-evidence, the platform is prone to the following problems: judging based on the order of entry into the database rather than the order of business occurrence, resulting in distortion of the current state; inability to distinguish between high-confidence and low-confidence records, resulting in abnormal records having an excessive impact on the overall judgment; lack of boundary-based correction capabilities for new evidence after the warning is generated, resulting in the inability to upgrade, downgrade, or recover erroneous warnings in a timely manner; and lack of unified and implementable risk quantification rules between different task types, resulting in poor consistency of platform results.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide an integrated service platform for safe production to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The integrated safety production service platform includes: a data standardization and filing module, a status chain construction module, a risk assessment closed-loop module, and a playback correction and update module, with signal connections between each module; Data normalization and archiving module: acquires raw data from multiple sources and uniformly transcribs it into raw business records, generates risk object codes, uses the difference between the entry time and the business occurrence time to obtain the time series offset, uses weight coefficients and normalized time constants to calculate the initial record quality value, and determines candidate merging groups based on the proximity of the business occurrence time and the overlap ratio of the evidence summary keyword set; State chain construction module: Collect all corresponding records according to risk object code, sort them in order of priority by business occurrence time, initial record quality value, and entry time, divide the records into four state intervals, select the record with the highest initial record quality value in each interval as the main record, and the rest as supporting records, take the quality value of the main record as the state credibility of the interval, determine the time boundary of each state interval, and judge and temporarily store the candidate records for playback according to the rules. Risk assessment closed-loop module: Extracts the current unfinished state interval of each risk state chain, generates temporally consistent judgment segments, calculates the timeout duration, calculates the risk assessment value in combination with the state confidence, inputs the specified structured features into the random forest model to complete the scene classification, and falls back to rule classification when the classification confidence is insufficient. It divides the warning level according to the risk assessment value threshold, generates the corresponding warning task and binds the disposal task, and executes the corresponding closed-loop release logic according to the task type. Replay Correction Update Module: When the triggering conditions are met, replay correction is initiated. Based on the review results, the record quality value of the corresponding original business record is updated according to the rules. Starting from the start time of the replay, the state chain is reconstructed, the state credibility and risk judgment value are recalculated, and the corresponding early warning task is updated according to the recalculation results. Each replay generates a state chain version record.

[0007] In a preferred embodiment, the data normalization and archiving module includes the following steps: Acquire raw data from multiple sources, including enterprise check-in records, inspection task records, hazard reporting records, handling result records, review and confirmation records, reported clue records, and task information returned by external platforms; They are uniformly transcribed into original business records that include enterprise identifier, location identifier, task type identifier, business occurrence time, record entry time, record source type, evidence integrity status, operating entity identifier, and original status category; The risk object code is generated based on a combination of region identifier, enterprise identifier, location identifier, task type identifier, and business date identifier; The initial record quality value is assigned based on the record source type and the state of evidence integrity; The initial record quality value is calculated by combining the time offset and the evidence integrity value. The evidence integrity value is determined by the evidence integrity status of the record. The more complete the evidence and the smaller the time offset, the higher the initial record quality value. When the difference in the time of occurrence of the two original business records is not greater than the preset merging time difference tolerance, and the ratio of the number of elements in the intersection to the number of elements in the keyword set formed after the original evidence summary is segmented is not less than the preset summary similarity threshold, the two original business records are grouped into the same candidate merging group.

[0008] In a preferred embodiment, the state chain construction module includes the following steps: All original business records corresponding to the risk object are collected according to the risk object code, and sorted by the time of business occurrence as the first sorting key, the initial record quality value as the second sorting key, and the record entry time as the third sorting key. And based on the original status category of the original business records, the sorted records are divided into status intervals of triggered state, pending disposal state, pending review state, and deactivated state; Within each state interval, the original business record with the highest initial record quality value is selected as the master record, and the remaining records are used as supporting records. The initial record quality value of the master record is used as the state confidence level for this state interval; The start time of the business occurrence in the master record is taken as the start time of the state interval, and the end time of the business occurrence in the master record of the next state interval is taken as the end time of the state interval.

[0009] In a preferred embodiment, the risk assessment closed-loop module includes the following steps: Extract the currently unfinished state interval from the risk state chain and generate a temporally consistent judgment fragment; Calculate the ratio of the actual duration of the temporal consistency determination segment from the start time to the current time to the allowed duration corresponding to the task type, and obtain the timeout duration. The risk assessment value is calculated by combining the timeout duration with the state reliability of the state interval. The higher the timeout duration or the lower the state reliability, the higher the risk assessment value. Structured features, including task type identifier, current state category, state confidence, and timeout duration, are input into a pre-defined machine learning model for scene classification. When the model output confidence is lower than the pre-defined classification confidence threshold, it falls back to rule classification. The system classifies warning levels based on the comparison between the risk assessment value and the preset warning threshold, generates warning tasks, and binds them to the handling tasks.

[0010] In a preferred embodiment, the replay correction update module includes the following steps: When a candidate record is found where the business occurred earlier than the end of the risk state chain, or when the review results are inconsistent with the state chain conclusions, a replay correction is triggered. Update the record quality value of the original business records based on the review results; When the review results support the status represented by the original business record, its record quality value is increased; when the review results negate the status represented by the original business record, its record quality value is decreased. The earliest business occurrence that triggers the replay is taken as the replay start time. From that time, the state chain is reconstructed and the state credibility is recalculated, and then the risk judgment value is recalculated. Based on the recalculated results, the warning task is upgraded, downgraded, restored, or corrected and closed, while a state chain version record is generated for each replay.

[0011] The technical effects and advantages of the integrated safety production service platform of this invention are as follows: This invention enables the continuous organization of records from different sources, with different arrival times, and varying degrees of evidence completeness around the same risk object by performing unified transcription, unified encoding, and quality assessment on multi-source business records. Through a state chain construction mechanism, it transforms static judgments based on individual records into dynamic judgments based on state intervals. Through a risk assessment closed-loop mechanism, it achieves unified quantification and tiered handling of risk levels under different task types. Through a replay correction and update mechanism, the platform can perform bounded recalculation and correction of existing judgment results after late records and re-examination of counter-evidence arrive, thereby improving the consistency, traceability, and stability of the platform's output results and the business closed loop. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the integrated safety production service platform of the present invention.

[0013] Figure 2 A flowchart for unifying and archiving business records from multiple sources. Detailed Implementation

[0014] 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.

[0015] Example Please see Figures 1-2 As shown, the present invention discloses an integrated service platform for safe production, including: a data normalization and filing module, a status chain construction module, a risk assessment closed-loop module, and a playback correction and update module, with signal connections between the modules; Data normalization and archiving module: acquires raw data from multiple sources and uniformly transcribs it into raw business records, generates risk object codes, uses the difference between the entry time and the business occurrence time to obtain the time series offset, uses weight coefficients and normalized time constants to calculate the initial record quality value, and determines candidate merging groups based on the proximity of the business occurrence time and the overlap ratio of the evidence summary keyword set; State chain construction module: Collect all corresponding records according to risk object code, sort them in order of priority by business occurrence time, initial record quality value, and entry time, divide the records into four state intervals, select the record with the highest initial record quality value in each interval as the main record, and the rest as supporting records, take the quality value of the main record as the state credibility of the interval, determine the time boundary of each state interval, and judge and temporarily store the candidate records for playback according to the rules. Risk assessment closed-loop module: Extracts the current unfinished state interval of each risk state chain, generates temporally consistent judgment segments, calculates the timeout duration, calculates the risk assessment value in combination with the state confidence, inputs the specified structured features into the random forest model to complete the scene classification, and falls back to rule classification when the classification confidence is insufficient. It divides the warning level according to the risk assessment value threshold, generates the corresponding warning task and binds the disposal task, and executes the corresponding closed-loop release logic according to the task type. Replay Correction Update Module: When the triggering conditions are met, replay correction is initiated. Based on the review results, the record quality value of the corresponding original business record is updated according to the rules. Starting from the start time of the replay, the state chain is reconstructed, the state credibility and risk judgment value are recalculated, and the corresponding early warning task is updated according to the recalculation results. Each replay generates a state chain version record.

[0016] In the data normalization and archiving module, the following steps are taken: acquire raw data from multiple sources and uniformly transcribe it into raw business records, generate risk object codes, use the difference between the entry time and the business occurrence time to obtain the time series offset, use weight coefficients and normalized time constants to calculate the initial record quality value, and determine candidate merging groups based on the proximity of the business occurrence time and the overlap ratio of the evidence summary keyword set. State chain construction module: Collect all corresponding records according to risk object code, sort them in order of priority by business occurrence time, initial record quality value, and entry time, divide the records into four state intervals, select the record with the highest initial record quality value in each interval as the main record, and the rest as supporting records, take the quality value of the main record as the state credibility of the interval, determine the time boundary of each state interval, and judge and temporarily store the candidate records for playback according to the rules. Risk assessment closed-loop module: Extracts the current unfinished state interval of each risk state chain, generates temporally consistent judgment segments, calculates the timeout duration, calculates the risk assessment value in combination with the state confidence, inputs the specified structured features into the random forest model to complete the scene classification, and falls back to rule classification when the classification confidence is insufficient. It divides the warning level according to the risk assessment value threshold, generates the corresponding warning task and binds the disposal task, and executes the corresponding closed-loop release logic according to the task type. Replay Correction and Update Module: When the triggering conditions are met, replay correction is initiated. Based on the review results, the record quality value of the corresponding original business record is updated according to the rules. Starting from the replay start time, the state chain is reconstructed, the state credibility and risk judgment value are recalculated, and the corresponding early warning task is updated according to the recalculation results. Each replay generates a state chain version record, the specific content of which includes: The raw data obtained from several business sources includes: enterprise attendance records, inspection task records, hazard reporting records, handling result records, review and confirmation records, reported clue records, and task information returned by external platforms, and is uniformly transcribed into raw business records; For example, each original business record includes: enterprise identifier, location identifier, task type identifier, business occurrence time, record entry time, record source type, evidence integrity status, operating entity identifier, and original status category; The raw data from several business sources are uniformly transcribed into raw business records. Each raw business record includes: enterprise identifier, location identifier, task type identifier, business occurrence time, record entry time, record source type, evidence integrity status, operating entity identifier, original status category, and original evidence summary. The enterprise identifier is used to uniquely identify the responsible enterprise; the location identifier is used to identify the location where the business occurs; the task type identifier is used to distinguish between check-in tasks, patrol tasks, hazard tasks, handling tasks, and review tasks; the business occurrence time is used to identify the actual time when the on-site action occurred; the record entry time is used to identify the time when the platform received the record; the record source type is used to identify which terminal or external platform generated the record; the evidence completeness status is used to identify whether the record contains location, images, text descriptions, and task-related information; and the original status category is used to identify the business semantics corresponding to the record when it was accessed. To enable the merging and continuous tracking of records of the same business object from multiple sources, a unified coding process is performed on the original business records to generate risk object codes; Preferably, the risk object code is formed by a combination of regional identifier, enterprise identifier, location identifier, task type identifier, and business date identifier; For example, when the task type is a hazard-related task, the risk object code is further linked to the hazard category identifier; when the task type is a fixed patrol point task, the risk object code is further linked to the point identifier. Through the risk object code, all subsequent records can be organized around the same object. After obtaining the risk object code, in order to identify the deviation between the arrival order of records and the order in which business occurs, the platform further calculates the time offset for each original business record, which can be used as the... The entry time of the original business record and the first The difference between the timestamps of the original business records is used to represent the timestamps of the transactions. Time offset of the original business record This time offset can be used to distinguish between real-time reporting records, normal delayed records, and obvious supplementary records, providing a time-dimensional basis for the selection of subsequent status master records. The time offset alone is not enough to support the construction of the subsequent state chain. Since two records with similar times may still have different levels of credibility, for example, a review pass record uploaded on-site by regulatory review personnel is usually more worthy of priority than a supplementary record containing only a single line of text description. Therefore, an initial record quality value is further generated for each original business record. The initial record quality value is directly assigned based on the record source type and the completeness of the evidence: the highest quality value is assigned when the record comes from regulatory review and is a complete record; the second highest quality value is assigned when the record comes from the enterprise's on-site mobile terminal and is a complete record; the lower quality value is assigned when the record comes from back-end supplementation but the evidence is incomplete; and the lowest quality value is assigned when the record is only supplementary to clues or lacks key fields. The initial record quality value is calculated by combining the evidence integrity value and the time offset. The formula can be expressed as follows: ;in, For the first The initial record quality value of each original business record. For the first The evidence completeness value for each original business record; a higher value corresponds to a complete record, and a lower value corresponds to an incomplete record. For the first The time-series offset of the original business record. The offset normalized time constant, The weighting coefficient for the completeness of evidence; Therefore, the more complete and timely the evidence is, the higher the initial record quality score; the less complete or obviously late the evidence is, the lower the initial record quality score. Based solely on risk object coding, proximity of business occurrence time, consistency of task type, and similarity of evidence summaries, multiple records that may represent the same on-site action are grouped into the same candidate merging group. The proximity of business occurrence time is preferably determined using a time difference threshold method; when the difference between the business occurrence times of two original business records is not greater than a preset merging time difference tolerance, they are considered to have similar business occurrence times. Evidence summaries are preferably calculated using the keyword set overlap ratio. Specifically, the original evidence summaries are first segmented to form keyword sets, and then the ratio of the number of elements in the intersection of the two sets to the number of elements in the union is used as the evidence summaries similarity. When this similarity is not less than a preset summary similarity threshold, the two original business records are considered to correspond to the same on-site action. For example, the preset merging time difference tolerance can be ten minutes, and the preset summary similarity threshold can be 0.6. Preferably, the candidate merging group is only used to identify these records that need to be compared together in subsequent modules, and does not directly determine the overlay relationship in the current module. In safe production operations, late supplementary recording, duplicate write-back and external platform back-transmission are common phenomena. If irreversible deletion is performed in the initial access stage, it will be difficult to explain the reason for the status change later, and it will also be difficult to correct the previous wrong judgment through the review results. This module establishes a unified semantic foundation at the risk object level for subsequent end-to-end processing. By performing unified transcription, unified encoding, time offset calculation, and initial record quality value generation on business records from different sources, in different formats, and with different arrival times, it transforms previously fragmented and difficult-to-compare records into a structured set of records that can be continuously organized around the same risk object and prioritized according to a unified standard. Through a candidate merging group mechanism, it preserves space for subsequent comparison and error correction, enabling the platform to handle real-time reporting scenarios as well as supplementary recording, write-back, and cross-platform transmission scenarios. This fundamentally enhances the platform's adaptability to complex business timelines and provides a sortable, comparable, and traceable underlying record foundation for the state chain construction module to build risk state chains. This ensures that subsequent state division is no longer based on isolated records but on a set of business facts that have undergone unified identification and quality assessment.

[0017] In the state chain construction module, all corresponding records are collected according to the risk object code and sorted in order of priority by the time of business occurrence, initial record quality value, and time of entry into the database. The records are divided into four state intervals. Within each interval, the record with the highest initial record quality value is selected as the master record, and the rest are used as supporting records. The quality value of the master record is taken as the state credibility of that interval. The time boundaries of each state interval are determined, and candidate records for playback are judged and temporarily stored according to rules. Specific content includes: For each risk object, all corresponding original business records are extracted and used to construct a risk state chain. The risk state chain is used to reflect the continuous evolution process of the same risk object from triggering, pending handling, pending review and finally to resolution within an observation period. Unlike the traditional solution that directly regards a record as the current state, safety production risk is essentially a time-series evolution process. Therefore, the state chain rather than isolated records should be used as the basis for judgment. For the actual on-site business sequence, it should be determined as much as possible by the time the business occurs; for multiple records with highly similar times of occurrence that may be the same action, the record with higher quality should be used as the status-dominant record; only when the first two are difficult to distinguish should the entry time be used to assist in decision-making. By adopting this multi-level sorting rule, it can be ensured that the platform organizes records according to the actual business sequence as much as possible, rather than according to the order in which the platform receives the records. All original business records under the same risk object code are sorted. The sorting rule is not simply sorting by the time of record entry into the database. Instead, the time of business occurrence is used as the first sorting key, the initial record quality value is used as the second sorting key, and the time of record entry into the database is used as the third sorting key. When two records have the same time of business occurrence or the difference between them is less than the preset time difference tolerance, the initial record quality value and the time of entry into the database are compared in turn. After sorting, the records under the risk object code are divided into four categories based on their status: triggered, pending action, pending review, and deactivated. The triggered state indicates that the current risk object has met the warning and attention conditions but no effective disposal task or disposal result has been formed; the pending disposal state indicates that the risk object has entered the disposal process but has not yet submitted a qualified disposal result; the pending review state indicates that the responsible party has submitted the disposal result but the regulator has not yet confirmed its effectiveness; the deactivated state indicates that the regulator has confirmed that the risk object has completed the risk elimination in the current round. For example, for check-in and inspection tasks, the risk status chain is built only around the anomaly handling process. That is, when there is no valid completion record after the planned completion time, it enters the trigger state; when the completion record is subsequently submitted and verified by the system, or when the regulatory side confirms that the anomaly has been eliminated, it directly enters the release state from the trigger state; if the platform further generates rectification tasks for the anomaly, it is allowed to enter the pending state from the trigger state, then the pending review state, and finally the release state; for hidden danger tasks, it enters the pending state after the hidden danger is discovered and registered; when the responsible party submits the handling result, it enters the pending review state; when the regulatory side reviews and approves it, it enters the release state; when the regulatory side reviews and approves it, it returns to the pending state. Through the above task type mapping relationship, check-in, inspection and hidden danger business can be expressed in the same terminology on the same platform, while retaining the differences in status flow paths under different task types; Within a certain state interval, there may be multiple original business records with similar semantics. By introducing a master record and supporting record mechanism, the record that best represents the current state is selected. All original business records within a certain state interval are compared according to their initial record quality values. The record with the highest initial record quality value is selected as the master record of the state interval, and the remaining records are attached to the state interval as supporting records. In this way, the highly representative record of the state interval can be fixed, while other records are retained for subsequent interpretation, playback and error correction. After the master record is determined, this embodiment further defines the initial record quality value of the master record as the state reliability of the state interval, and the formula can be expressed as: ;in, For the first The reliability of the state in each state interval For belonging to the first The first state interval The initial record quality value of each original business record. For the first The set of all original business records within each state interval; This definition method directly transitions from the quality value of the original record to the credibility expression at the status level. The higher the credibility of the status, the more accurately the status interval can reflect the on-site business status; the lower the credibility of the status, the greater the possibility that the status interval is affected by late supplementary recording, missing evidence, or unstable sources. After obtaining the credibility of the state, the start and end times of the state interval are further determined. The start time of a state interval is taken as the time when the business occurs in the main record of the state interval. The end time of the state interval is preferably taken as the time when the business occurs in the main record of the next state interval. If the current state interval is the last unfinished state interval, its end time is taken as the current platform time, so that each state interval has a clear time boundary. Furthermore, to address the potential impact of late arrival records on the existing state chain, the following processing is performed: When a new arrival record meets any of the following conditions, it is marked as a replay candidate record and temporarily stored: First, the business occurrence time of the new arrival record is earlier than the start time of any confirmed state interval in the current risk state chain; Second, the business occurrence time of the new arrival record falls between the start and end times of a confirmed state interval, but its initial record quality value is higher than the initial record quality value of the main record of that state interval; Third, the original state category represented by the new arrival record conflicts with the conclusion of the end state of the current risk state chain, and its initial record quality value is not lower than the preset replay threshold, which can be 0.6. For example, if the reviewer uploads a record of a failed review taken on-site the previous day the next day, the record may change the end state of the current state chain. However, the platform does not immediately rewrite the entire chain at this time. Instead, it marks it as a replay candidate record and stores it temporarily, handing it over to the subsequent replay correction and update module for unified processing. This prevents the platform from frequently fluctuating due to a single late record and improves operational stability. This module further organizes the discretized original business records in the data normalization and archiving module into a risk state chain that reflects the continuous evolution of risk objects. This transforms the traditional static processing method of judging the current state based on a single record into a dynamic processing method of identifying the real business process based on state intervals. By introducing a multi-layered sorting rule based on the time of business occurrence as the primary factor and the initial record quality value and the time of entry into the database as secondary factors, as well as a mechanism for master records, supporting records, and state credibility, the most representative records in each state interval are further fixed, giving the state expression clear and credible basis. Late but high-quality records are identified as replay candidate records and temporarily stored, rather than immediately overturning the existing link. While maintaining operational stability, this module reserves an entry point for subsequent corrections and provides a direct basis for the risk judgment closed-loop module to extract the current unfinished state interval and form temporally consistent judgment fragments. It also pre-sets clear trigger objects and processing boundaries for the replay correction and update module to implement replay correction and link reconstruction.

[0018] In the risk assessment closed-loop module, the currently unfinished state intervals of each risk state chain are extracted, a temporally consistent assessment segment is generated, the timeout duration is calculated, and the risk assessment value is calculated based on the state confidence level. The specified structured features are input into a random forest model to complete scene classification. If the classification confidence is insufficient, it reverts to rule-based classification. Warning levels are divided according to the risk assessment value threshold, corresponding warning tasks are generated and bound to handling tasks, and the corresponding closed-loop release logic is executed according to the task type. Specific content includes: The currently unfinished state interval is extracted from each risk state chain to form a temporally consistent judgment segment. This temporally consistent judgment segment refers to a judgment unit formed within the same risk object, state category, and continuous time interval. Each judgment unit corresponds to only one master record, one state confidence level, one start time, and one duration for subsequent calculations. The purpose of this structure is to avoid mixing and splicing original business records from different time periods or different state stages, thereby ensuring that risk judgments are based on a unified temporal semantics. Another core adjustment result alongside state credibility is timeout duration. Timeout duration characterizes the extent to which the current state interval has persisted since its inception, and is compared with the normalized duration for this task type. The formula can be expressed as: ;in, For the first The duration of timeout for each temporally consistent segment. For the first The actual duration of each temporally consistent segment from the start time to the current time. The allowed duration corresponding to the current task type; When a state has just entered the triggered state or the pending state, the actual duration is less than the allowed duration, and the timeout duration value is low. As the state continues to approach or exceed the allowed duration, the timeout duration value gradually approaches one. Although the allowed durations are different for different task types, they can all fall into the same value range after normalization, which is convenient for subsequent risk assessment in conjunction with state credibility. After obtaining the state credibility and timeout duration, the risk assessment value of the current temporal consistency judgment segment is calculated. The risk assessment value is used for unified control of early warning classification and closed-loop escalation. The formula can be expressed as: ;in, For the first Risk assessment value for each temporal consistency segment. For the first The duration of timeout for each temporally consistent segment. For the first The consistency of each temporal segment determines the reliability of the state interval corresponding to that segment. For the duration weighting coefficient; The longer a certain status interval lasts, the less time should be delayed in handling it. Therefore, the higher the duration of the timeout, the higher the risk assessment value. On the other hand, if the credibility of the current status is low, it means that the status may be affected by supplementary recording, lateness, or unstable source. The platform should not easily regard it as having been safely resolved, and the risk assessment value will also increase. Therefore, the severity of the warning is not only determined by the duration of the timeout, but also depends on whether the current status is credible enough. Preferably, a machine learning algorithm is used to classify the current temporal consistency judgment segment into scenes, and a random forest model can be used for classification and recognition. During the model training phase, time-consistency judgment segments that have been confirmed by regulatory results in historical business are selected as training samples, and each sample is labeled with a scenario category based on the final closed-loop result. Scenario categories include risks of not clocking in, risks of not inspecting, risks of not handling hidden dangers, risks of handling exceeding time limits, risks of review exceeding time limits, and risks of suspected supplementary recording conflicts. On the model input side, structured features are used as input instead of directly inputting original business records spliced ​​across time. The structured features include task type identifier, current state category, state credibility, timeout duration, credibility of the current master record source, completeness of evidence for the current master record, and number of supporting records within the current state interval. During the model inference phase, the structured features of the current temporal consistency judgment segment are input into the trained random forest model, which outputs the corresponding scene category. When the confidence level of the model output is lower than the preset classification confidence threshold, the platform reverts to the rule-based classification method based on task type identifier and current state category. After the scenario classification is completed, the risk grading trigger is further carried out. That is, under the premise of knowing the scenario category, the risk judgment value is used as a unified grading quantity to map different scenarios into Level 1 warning, Level 2 warning or Level 3 warning. Preferably, a first warning threshold, a second warning threshold, and a third warning threshold are set, wherein the first warning threshold is less than the second warning threshold, and the second warning threshold is less than the third warning threshold. When the risk assessment value is greater than or equal to the first warning threshold and less than the second warning threshold, a level-one warning task is generated; when the risk assessment value is greater than or equal to the second warning threshold and less than the third warning threshold, a level-two warning task is generated; when the risk assessment value is greater than or equal to the third warning threshold, a level-three warning task is generated. For example, a Level 1 warning can be pushed to the person in charge of the area, a Level 2 warning can be pushed to the person in charge of the supervisor, and a Level 3 warning can be pushed to the person in charge of higher-level supervision. The platform can reach these persons through multiple channels, such as message reminders, SMS reminders, or voice calls. While generating early warning tasks, an action task is automatically created, and the action task is bound to the current risk object code, current status range, current risk assessment value, and responsible entity. Preferably, for hazard handling tasks, after the responsible party submits the handling results, the risk object is only allowed to change from the pending handling state to the pending review state, and not directly enter the deregulation state; the platform only allows the corresponding warning task to be deregulated when the regulatory review is passed and the state chain actually enters the deregulation state under the logic of the state chain construction module. For tasks that are not checked in or not inspected, once the completion record is submitted and verified by the system, or the supervisor confirms that the anomaly is no longer valid, the system can directly transition from the trigger state to the release state. If the platform has already generated a rectification task for the anomaly, the system will complete the closed loop according to the path of trigger state, pending handling state, pending review state, and release state. This will prevent the upload of any handling result from being mistakenly considered as the risk being eliminated, while ensuring that the release logic for different task types is consistent with their actual business processes. Furthermore, to ensure consistent results across multiple platforms, the multi-terminal synchronous display unit synchronizes the risk object code, warning level, risk assessment value, handling progress, and current status interval to the business management terminal, mobile terminal, and display terminal. This module further converges the risk state chain obtained by the state chain construction module into a temporally consistent judgment segment oriented towards the current moment. Under a unified temporal semantic, it completes risk identification, risk classification, and closed-loop task binding. By extracting the current unfinished state interval into a judgment unit corresponding to a single master record, a single state credibility, and a single timeout duration, it avoids judgment distortion caused by mixing and splicing information from different time periods and different state stages. At the same time, by mapping the state credibility and timeout duration together into a risk judgment value, and combining the scenario classification results for graded early warning, the platform can achieve unified quantification, unified escalation, and unified handling control across different risk types. Furthermore, it binds the early warning result with the handling task, responsible entity, risk object code, and state interval, and requires that the handling result must be reviewed before the early warning can be lifted. This integrates "risk identification" and "driving the closed loop" into a continuous process, forming an executable business response result. It also provides a clear correction object and comparison benchmark for the playback correction update module to determine whether the original early warning needs to be upgraded, downgraded, restored, or corrected and closed.

[0019] In the replay correction and update module, replay correction is initiated when the triggering conditions are met. Based on the review results, the record quality value of the corresponding original business record is updated according to the rules. Starting from the replay start time, the state chain is reconstructed, the state credibility and risk judgment value are recalculated, and the corresponding early warning task is updated based on the recalculation results. Each replay generates a state chain version record, the specific content of which includes: Although more stable warning results are generated based on temporal consistency judgment segments, two types of events that affect the stability of the results will still occur, including late arrival records and re-examination of evidence. The re-examination results and late arrival records are used to perform replay correction and small step iteration on the results of the previous three steps. Preferably, replay correction is triggered when any of the following conditions are met: First, a replay candidate record appears, and the business occurrence time of the record is earlier than the start time of the end state interval of the current risk state chain; Second, the review result is inconsistent with the state conclusion expressed by the current state chain; Third, the same risk object undergoes multiple state reversals within a preset period. After replay is triggered, the platform does not directly rewrite the upper-level state manually, but starts to correct from the bottom-level record quality value. Preferably, the platform updates the initial record quality value generated by the data normalization and archiving module in small steps based on the review results, and corrects the quality value of original business records that are supported or rejected by the review. The formula can be expressed as: ;in, For the first The updated record quality value of the original business record. For the first The record quality value of the original business record before it was updated. For single update step size, This serves as a consistency marker; when the review results support the status represented by the original business record. When the review result negates the status represented by the original business record This simple and stable update method gradually learns the reliability of records from different sources in actual operation without the need to retrain complex models. After updating the record quality value, starting from the earliest business occurrence time that triggered the replay, the state chain construction module is re-executed only for the corresponding risk object code. That is, the state intervals are reordered, the master record is reselected, and the state credibility is recalculated. Since the state credibility of the state chain construction module is determined by the record quality value through the maximum value rule, the state credibility will automatically change as long as the underlying quality value changes. For the affected current state range, the risk assessment closed-loop module is re-executed, which involves recalculating the timeout duration, recalculating the risk assessment value, and reassessing the warning level. If the new calculation result is inconsistent with the original warning task, the platform simultaneously performs warning task upgrades, downgrades, restorations, or corrections / closures. Preferably, if the original warning level is too low, it will be automatically upgraded and pushed to the newly added responsible entities; if the original warning conclusion is overturned, the original warning will be marked as corrected and closed, but its historical version will be retained; if the original lifting conclusion is overturned, the handling task or review task will be resumed and the closed-loop process will be re-entered. Furthermore, to ensure traceability throughout the entire process and generate a state chain version record for each replay correction, preferably, the state chain version record includes at least: risk object code, replay reason, replay start time, set of state intervals before replay, set of state intervals after replay, risk judgment value before replay, and risk judgment value after replay. The purpose of this module is to establish a self-correcting mechanism for late arrival records and re-examination of evidence for the aforementioned three modules. This mechanism enables the platform to not only generate risk assessment results but also to correct and update them in a bounded manner when new evidence arrives. Starting from the quality value of the underlying records, the original business records are corrected in small steps based on the re-examination results. Then, starting from the earliest business occurrence time that triggers the playback, the state chain construction, state credibility calculation, and risk assessment value calculation are re-executed. This achieves a bottom-up chain-like playback correction path, which preserves the historical version and traceability of the original judgment process while avoiding a crude overwrite of the platform by a single abnormal record. This allows the platform to gradually learn the actual reliability of records from different sources during continuous operation.

[0020] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0021] The above embodiments can be implemented, in whole or in part, by software, hardware, 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.

[0022] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0023] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0024] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0025] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An integrated service platform for safe production, characterized in that: Signal connections between modules; Data normalization and archiving module: acquires raw data from multiple sources and uniformly transcribs it into raw business records, generates risk object codes, uses the difference between the entry time and the business occurrence time to obtain the time series offset, uses weight coefficients and normalized time constants to calculate the initial record quality value, and determines candidate merging groups based on the proximity of the business occurrence time and the overlap ratio of the evidence summary keyword set; State chain construction module: Collect all corresponding records according to risk object code, sort them in order of priority by business occurrence time, initial record quality value, and entry time, divide the records into four state intervals, select the record with the highest initial record quality value in each interval as the main record, and the rest as supporting records, take the quality value of the main record as the state credibility of the interval, determine the time boundary of each state interval, and judge and temporarily store the candidate records for playback according to the rules. Risk assessment closed-loop module: Extracts the current unfinished state interval of each risk state chain, generates temporally consistent judgment segments, calculates the timeout duration, calculates the risk assessment value in combination with the state confidence, inputs the specified structured features into the random forest model to complete the scene classification, and falls back to rule classification when the classification confidence is insufficient. It divides the warning level according to the risk assessment value threshold, generates the corresponding warning task and binds the disposal task, and executes the corresponding closed-loop release logic according to the task type. Replay Correction Update Module: When the triggering conditions are met, replay correction is initiated. Based on the review results, the record quality value of the corresponding original business record is updated according to the rules. Starting from the start time of the replay, the state chain is reconstructed, the state credibility and risk judgment value are recalculated, and the corresponding early warning task is updated according to the recalculation results. Each replay generates a state chain version record.

2. The integrated safety production service platform according to claim 1, characterized in that, The data standardization and filing module is specifically used to: acquire raw data from multiple sources, including enterprise check-in records, inspection task records, hazard reporting records, handling result records, review and confirmation records, reported clue records, and task information returned by external platforms; They are uniformly transcribed into original business records that include enterprise identifier, location identifier, task type identifier, business occurrence time, record entry time, record source type, evidence integrity status, operating entity identifier, and original status category; The risk object code is generated by combining the region identifier, enterprise identifier, location identifier, task type identifier, and business date identifier.

3. The integrated safety production service platform according to claim 2, characterized in that, The initial record quality value is assigned based on the record source type and the state of evidence integrity; The initial record quality value is calculated by combining the time offset and the evidence integrity value. The evidence integrity value is determined by the evidence integrity status of the record. The more complete the evidence and the smaller the time offset, the higher the initial record quality value.

4. The integrated safety production service platform according to claim 3, characterized in that, When the difference in the time of occurrence of the two original business records is not greater than the preset merging time difference tolerance, and the ratio of the number of elements in the intersection to the number of elements in the keyword set formed after the original evidence summary is segmented is not less than the preset summary similarity threshold, the two original business records are grouped into the same candidate merging group.

5. The integrated safety production service platform according to claim 1, characterized in that, The state chain construction module is specifically used to: collect all the original business records corresponding to the risk object code, and sort them by the time of business occurrence as the first sorting key, the initial record quality value as the second sorting key, and the record entry time as the third sorting key; And based on the original status category of the original business records, the sorted records are divided into status intervals of triggered state, pending disposal state, pending review state, and deactivated state.

6. The integrated safety production service platform according to claim 5, characterized in that, Within each state interval, the original business record with the highest initial record quality value is selected as the master record, and the remaining records are used as supporting records. The initial record quality value of the master record is used as the state confidence level for this state interval; The start time of the business occurrence in the master record is taken as the start time of the state interval, and the end time of the business occurrence in the master record of the next state interval is taken as the end time of the state interval.

7. The integrated safety production service platform according to claim 1, characterized in that, The risk assessment closed-loop module is specifically used to: extract the currently unfinished state interval from the risk state chain and generate a temporally consistent assessment fragment; Calculate the ratio of the actual duration of the temporal consistency determination segment from the start time to the current time to the allowed duration corresponding to the task type, and obtain the timeout duration. The risk assessment value is calculated by combining the timeout duration with the state reliability of the state interval. The higher the timeout duration or the lower the state reliability, the higher the risk assessment value.

8. The integrated safety production service platform according to claim 7, characterized in that, Structured features, including task type identifier, current state category, state confidence, and timeout duration, are input into a pre-defined machine learning model for scene classification. When the model output confidence is lower than the pre-defined classification confidence threshold, it falls back to rule classification. The system classifies warning levels based on the comparison between the risk assessment value and the preset warning threshold, generates warning tasks, and binds them to the handling tasks.

9. The integrated safety production service platform according to claim 1, characterized in that, The replay correction update module is specifically used to: trigger replay correction when there are candidate records where the business occurred earlier than the end of the risk state chain, or when the review results are inconsistent with the state chain conclusions. Update the record quality value of the original business records based on the review results; When the review results support the status represented by the original business record, its record quality value is increased; when the review results negate the status represented by the original business record, its record quality value is decreased.

10. The integrated safety production service platform according to claim 9, characterized in that, The earliest business occurrence that triggers the replay is taken as the replay start time. From that time, the state chain is reconstructed and the state credibility is recalculated, and then the risk judgment value is recalculated. Based on the recalculated results, the warning task is upgraded, downgraded, restored, or corrected and closed, while a state chain version record is generated for each replay.

Images