Anesthesia physician workload monitoring and early warning system based on multi-source physiological and behavioral data

Abstract

The application discloses anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data, and particularly relates to the technical field of medical data processing and surgical anesthesia monitoring and early warning, which comprises collecting heart rate data, skin electricity data, eye movement data and body movement data of an anesthesiologist, collecting operation records and alarm records of an anesthesia machine, a monitor and an infusion device, collecting interactive records of a surgeon terminal, a nurse terminal and an anesthesia workstation, and outputting original data streams according to the same time scale; performing time correction, field padding, repeated record removal and breakpoint connection on the original data streams, and outputting continuous data streams according to an anesthesiologist identifier, a patient identifier and an operating room identifier; performing unified access, fragmentation reconstruction and digital twin load mapping on anesthesiologist physiological data, device operation and alarm records and collaborative interactive records, and performing backlog identification and early warning determination on the continuous load evolution process.

Description

Technical Field

[0001] This invention relates to the field of medical data processing and surgical anesthesia monitoring and early warning technology, and more specifically, to an anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data. Background Technology

[0002] In the practice of monitoring the workload of anesthesiologists, existing technologies generally rely on the real-time acquisition and analysis of physiological data, such as heart rate, respiratory rate and blood pressure, to assess the workload of anesthesiologists. Typically, these systems identify whether the physician is under high workload by setting thresholds for physiological parameters, and then issue warning signals. However, this method fails to fully consider the complex working environment and multi-factor intertwined situations in anesthesia and surgery. In practice, anesthesiologists not only have to deal with their own physiological reactions, but also need to work closely with other team members during the operation to coordinate drug dosages, monitor the patient's condition, and execute the surgeon's temporary instructions. Taking cardiac surgery as an example, anesthesiologists must quickly adjust strategies and handle changes in patients' vital signs in a high-pressure working environment. At the same time, interactions with other surgical personnel greatly increase their workload. In this situation, traditional systems fail to effectively capture the impact of team collaboration on the anesthesiologist's workload and cannot provide targeted support. In particular, when there are frequent interactions between team members or when unexpected situations occur, existing systems often fail to identify the additional burden brought about by these collaborations in real time, resulting in a significant reduction in the timeliness and accuracy of early warnings. Therefore, the technical problem to be solved by this application is: how to construct a workload monitoring system that can comprehensively consider the collaborative pressure of anesthesiologists and other members of the surgical team, so as to ensure that the workload status of anesthesiologists can be assessed in real time and accurately in complex working environments. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data. This system achieves unified access, fragmented reconstruction, and digital twin workload mapping of anesthesiologist physiological data, equipment operation and alarm records, and collaborative interaction records. It also performs backlog identification and early warning judgment on the continuous workload evolution process, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data, comprising: The data acquisition device is used to collect heart rate data, skin conductance data, eye movement data, and body movement data of anesthesiologists; to collect operation records and alarm records of anesthesia machines, monitors, and infusion equipment; to collect interaction records of surgeon terminals, nurse terminals, and anesthesia workstations; and to output raw data streams according to the same time scale. The access device is used to perform time correction, field completion, duplicate record removal and breakpoint continuation on the raw data stream, and output a continuous data stream according to the anesthesiologist identifier, patient identifier and operating room identifier; The segment generation module is used to read continuous data streams, take the time period between two adjacent device operation records, interaction records or alarm records as working segments, extract physiological change sequences, gaze switching sequences, limb operation sequences, device response sequences and collaborative round-trip sequences within each working segment, and output the segment result set. The twin mapping module is used to read the segment result set, match the physiological change sequence in each work segment with the corresponding gaze switching sequence, limb operation sequence, equipment response sequence and collaborative round-trip sequence, construct the digital twin load segment corresponding to the current work segment, solve the individual occupancy result, equipment traction result and collaborative traction result, and continue to write the unreleased occupancy of the previous work segment into the digital twin load segment corresponding to the next work segment, and output the continuous load result set.

[0005] In a preferred embodiment, it further includes: The early warning module is used to read the continuous load result set, compare the individual occupancy results, equipment traction results, collaborative traction results and unreleased occupancy in the digital twin load segments corresponding to adjacent work segments in chronological order, identify load backlog segments that continue to increase and remain connected in the next work segment, and generate early warning results and handling instructions based on the equipment source and collaborative source corresponding to the load backlog segment, and output early warning information.

[0006] In a preferred embodiment, the data acquisition device includes: It receives heart rate data, skin conductance data, eye movement data, and body movement data output by the anesthesiologist's wearable device, performs sampling time registration, duplicate sampling point removal, and missing sampling point removal within the same sampling cycle, and outputs physiological data with sampling time and anesthesiologist identification. Receive operation records and alarm records output from anesthesia machines, monitors and infusion devices; receive interaction records output from surgeon terminals, nurse terminals and anesthesia workstations; perform event time extraction, event type marking, initiator marking, receiver marking and round-trip sequence marking; and output event data with event timestamps. Using the event time in the event data as the alignment reference, the data in the physiological data located between two adjacent event times is written into the corresponding event interval, and the physiological data and event data in the same event interval are merged according to the anesthesiologist identifier, patient identifier, and operating room identifier, and the original data stream is output.

[0007] In a preferred embodiment, the access device includes: Receive the raw data stream, perform time correction, eliminate data time deviation by synchronizing physiological data, device operation data and interaction data, and output the corrected data stream; The process involves performing field completion, which fills in missing fields in the original data stream. This is done by associating anesthesiologist identifiers, patient identifiers, and operating room identifiers to ensure the integrity and accuracy of the data stream, and then outputting the completed data stream.

[0008] In a preferred embodiment, the access device further includes: Remove duplicate records from the original data stream, detect redundant information in the data stream, remove duplicate content based on timestamps and data type identifiers, and output the deduplicated data stream; Execute breakpoint resume to connect the missing data caused by the interruption of acquisition with the data of the next acquisition, so as to ensure the continuity of the data stream and output a continuous data stream after breakpoint resume processing.

[0009] In a preferred embodiment, the fragment generation module includes: Read the continuous data stream, take the time period between two adjacent device operation records, interaction records or alarm records as the working segment, determine the segment boundary by comparing timestamps, and output the working segment after the time period is divided. Within each work segment, physiological change sequences, gaze switching sequences, limb operation sequences, equipment response sequences, and collaborative round-trip sequences are extracted. By segmenting and calibrating each sequence according to the event time and integrating the data, a corresponding sequence set is generated. The output fragment result set contains each working fragment arranged in chronological order, ensuring that the sequence data in each fragment is complete and without duplication. By performing weighted calculations on each sequence within a fragment, the load performance and temporal characteristics of each sequence within the working fragment are output.

[0010] In a preferred embodiment, the twin mapping module includes: Read the fragment result set, receive multi-source information from physiological data, behavioral data, equipment operation data and collaborative event data, and generate a fragment dataset by allocating work fragments according to anesthesiologist identifier, patient identifier and operating room identifier; The physiological change sequence in each work segment is matched with the gaze switching sequence, limb operation sequence, equipment response sequence, and collaborative round-trip sequence. Through time alignment and sequence calibration, the temporal consistency between each pair of data is ensured, and the corresponding individual occupancy result, equipment traction result, and collaborative traction result are extracted.

[0011] In a preferred embodiment, the twin mapping module further includes: Construct digital twin load segments, generate corresponding digital twin load state bodies within each working segment, ensure the continuity of load state by inheriting the unreleased occupancy of the previous working segment across segments, and write the inherited data into the digital twin load segment of the next working segment. Output a continuous load result set, combining individual occupancy results, equipment traction results, and collaborative traction results from the work segment. By fusing all load data and calculating confidence levels, a comprehensive load assessment result is generated, ensuring that the output result reflects the load evolution process throughout the anesthesia workflow.

[0012] In a preferred embodiment, the early warning module includes: Read the continuous load result set, receive the individual occupancy result, equipment traction result, collaborative traction result and unreleased occupancy in the digital twin load segment corresponding to each work segment, identify the load change trend by comparing the load data in adjacent work segments in time sequence, and output the load change data. By setting rules and threshold packages, load backlog segments in adjacent work segments are identified, it is determined whether the load continues to increase in the next work segment without being released, and a corresponding early warning signal is generated based on the judgment result, outputting load backlog information.

[0013] In a preferred embodiment, the early warning module further includes: Based on the equipment and collaboration sources corresponding to the load backlog section, a multi-objective judgment chain is constructed. Through confidence update and time-series window judgment, the overall trend of load status is calculated, and accurate early warning results and handling directions are generated, outputting early warning information with execution guidance. By performing cross-perspective consistency checks and combining multi-source evidence from individual occupancy results, equipment traction results, and collaborative traction results, it is determined whether there are conflicts in the load data. If there are conflicts, they are corrected through a rollback and re-check mechanism, and data consistency is ensured through a gating token. Finally, optimized early warning information is output.

[0014] The technical effects and advantages of this invention are as follows: 1. This solution collects physiological data, equipment operation and alarm records, and collaborative interaction records simultaneously, and constructs digital twin load segments to jointly calculate individual occupancy, equipment traction, and collaborative traction. This can more completely reflect the actual load status of anesthesiologists in complex collaborative environments than simply relying on physiological thresholds, thereby relatively improving the pertinence and real-time nature of early warning. 2. Perform time correction, field completion, duplicate record removal and breakpoint continuation on the raw data stream to organize multi-source heterogeneous records continuously under a unified time scale axis and unified attribution key. This can relatively reduce the interference of time misalignment, missing attribution and redundant writing on subsequent load determination, thereby improving the availability of continuous monitoring results. 3. Divide the work segments into time periods between adjacent device operation records, interaction records, or alarm records, and extract physiological change sequences, line-of-sight switching sequences, limb operation sequences, device response sequences, and collaborative round-trip sequences within the segments. This can anchor the load identification to the specific work progress interval, thereby relatively improving the refinement of load source location. 4. Align the various sequences in each work segment one by one and solve for individual occupancy results, equipment traction results and collaborative traction results. Then, write the unreleased occupancy amount of the previous work segment into the next work segment. This can characterize the cross-segment continuity relationship of the load, thereby relatively improving the ability to identify continuously backlogged loads. 5. By comparing individual occupancy results, equipment traction results, collaborative traction results, and unreleased occupancy in adjacent work segments in chronological order, the system can identify continuously increasing and unreleased load backlog segments, distinguishing short-term fluctuations from continuous load propagation, thus helping to relatively suppress false alarms caused by instantaneous anomalies. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the system module structure of the present invention. Detailed Implementation

[0016] 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.

[0017] Refer to the instruction manual appendix Figure 1 The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data of the present invention includes: The data acquisition device is used to collect heart rate data, skin conductance data, eye movement data, and body movement data of anesthesiologists; to collect operation records and alarm records of anesthesia machines, monitors, and infusion equipment; to collect interaction records of surgeon terminals, nurse terminals, and anesthesia workstations; and to output raw data streams according to the same time scale. In this embodiment, the data acquisition device is responsible for organizing physiological signals related to the anesthesiologist's workload and equipment and collaborative events during the surgical process into a raw data stream with the same object scope and time caliber. This raw data stream is then used by subsequent access devices to perform time correction, field completion, duplicate record removal, and breakpoint continuation. The processing order is not simply summarizing various records, but rather first forming physiological data with a unified sampling time and attribution identifier, then forming event data with a unified event time and event attributes. Subsequently, event intervals are constructed based on event times, and physiological data is written into the corresponding event intervals. Merging is completed under the condition that the anesthesiologist identifier, patient identifier, and operating room identifier are consistent, thereby avoiding problems such as object mixing, time misalignment, and unclear interval boundaries in subsequent continuous data streams. This implementation process includes the following: First, the heart rate data, skin conductance data, eye movement data, and body movement data output from the anesthesiologist's wearable device are uniformly sampled and processed to form physiological data that can fall within the subsequent event interval. Among them, the heart rate data uses the heart rate value output according to the sampling period, the skin conductance data uses the skin conductance value output according to the sampling period, the eye movement data uses the fixation area code value output according to the sampling period, and the body movement data uses the movement amplitude value output according to the sampling period. The sampling period is taken from the preset sampling configuration in the acquisition device, which is written according to the wearable device specifications during deployment and remains unchanged during the current operation. After receiving various physiological data, the acquisition device generates the sampling time with the start time and cycle number corresponding to the uniform sampling period, and writes it into the physiological record table according to the anesthesiologist's identification, data type, and sampling time. For two physiological records with the same sampling time, data type, and data source, only the record that enters the physiological record table first is retained to remove duplicate sampling points. If no corresponding physiological data is received within a certain sampling period, a missing marker is written at that sampling time, and the anesthesiologist's identifier, data type, and sampling time are retained, but not the numerical field, to ensure that the subsequent time series length is consistent. After processing, physiological data with sampling time and anesthesiologist's identifier is output and written to the original acquisition buffer for subsequent event interval merging. If the anesthesiologist's identifier currently bound to the wearable device is inconsistent with the anesthesia workstation login identifier, the physiological record is written to the binding error table instead of the original acquisition buffer, and reception continues after the binding relationship is restored to consistency. Subsequently, the operation and alarm records output by the anesthesia machine, monitor, and infusion equipment, as well as the interaction records output by the surgeon's terminal, nurse's terminal, and anesthesia workstation, are subjected to event standardization processing to form event data that can serve as interval boundaries and collaboration basis. The acquisition device extracts the record generation time from each record as the event time and performs event type marking according to the event code table. The event code table is pre-written according to the equipment type and terminal type during system deployment. Among them, equipment parameter adjustment, mode switching, and manual confirmation operations are marked as equipment operation events; over-limit alarms, fault alarms, and infusion abnormality alarms are marked as equipment alarm events; and message sending, voice transcription, workstation confirmation, and collaborative response are marked as interaction events. For interaction records, the acquisition device continues to write initiator and receiver tags based on the record source and destination, and arranges them in ascending order of event time under the same interaction topic, writing round-trip order tags, where the first initiation is written as the first round-trip order, the first response is written as the second round-trip order, and subsequent supplementary requests and responses are written in ascending order; after processing, the event data with event time is output and written to the event buffer for reading when subsequent event intervals are generated; if a device record is missing the original event time, the receiving time of the record received by the acquisition device is used as a substitute and a substitute time tag is written; if an interaction record is missing a receiver tag, a tag to be supplemented is written first and the record is retained, and the missing tags are filled in reverse order according to the topic number when a response record of the same topic appears; After completing the physiological data processing and event data processing, the acquisition device generates event intervals based on the event time as the alignment benchmark, writes the physiological data into the corresponding event intervals, and then merges them with the event data according to a unified attribution key to output the raw data stream. Specifically, the acquisition device first arranges the event data in the event buffer in ascending order of event time, and forms event intervals based on the time range between two adjacent event times. The event intervals adopt a front-closed and back-open attribution rule, that is, physiological data with a sampling time greater than or equal to the previous event time and less than the next event time is written into the current event interval, and physiological data with a sampling time equal to the next event time is written into the next event interval. For physiological data before the first event time, it is written into the first preceding event interval, and for physiological data after the last event time and before a new event has been received, it is written into the last holding interval. After the interval is assigned, the physiological data and event data within the same event interval are merged according to the anesthesiologist identifier, patient identifier, and operating room identifier. The patient identifier and operating room identifier are prioritized based on the binding results in the current surgical registration table. If the event data already contains a patient identifier or operating room identifier, the identifier in the event data is used as the priority value. After merging, an original data stream is generated. The original data stream includes at least the interval start time, interval end time, anesthesiologist identifier, patient identifier, operating room identifier, physiological data set, and event data set. The original data stream is written to the access device's read queue. If multiple patient identifiers or multiple operating room identifiers appear within the same event interval, only the set of identifiers that are consistent with the current binding relationship of the anesthesia workstation is retained and written into the original data stream. At the same time, the remaining records are written into the conflict record table for subsequent review and reading. Through the above processing, the acquisition device first completes the time specification and attribution specification of physiological data, then completes the time specification and attribute specification of event data, and finally completes the event interval placement and unified merging, thereby outputting a raw data stream with clear objects, continuous time and clear interval boundaries, providing a stable input for subsequent continuous data stream generation and digital twin load segment construction; In practical applications: When an anesthesiologist maintains anesthesia for the current patient in the current operating room, the wearable device continuously outputs heart rate, skin conductance, gaze area code, and motion amplitude values. The monitor outputs a blood pressure alarm record at a certain moment. The nurse terminal then sends a fluid resuscitation confirmation record to the anesthesia workstation. The data acquisition device first writes various physiological data into the sampling time according to a uniform sampling cycle and removes duplicate sampling points. Then, it extracts the event time, writes the event type, initiator, receiver, and round-trip order from the blood pressure alarm record and the fluid resuscitation confirmation record. Subsequently, it forms an event interval with the blood pressure alarm time and the fluid resuscitation confirmation time, writes the physiological data between the two times into the corresponding event interval, and merges it with the event data in the interval according to the anesthesiologist identifier, patient identifier, and operating room identifier. Finally, it outputs the raw data stream for the access device to continue reading.

[0018] The access device is used to perform time correction, field completion, duplicate record removal and breakpoint continuation on the raw data stream, and output a continuous data stream according to the anesthesiologist identifier, patient identifier and operating room identifier; In this embodiment, the access device is responsible for organizing the raw data stream output by the acquisition device into a continuous data stream that can be read continuously and directly enter the work segment segmentation process. Its processing focus is not on re-acquiring data, but on correcting time deviations, supplementing attribution fields, removing redundant records, and restoring interrupted links, thereby ensuring that data corresponding to the same anesthesiologist, the same patient, and the same operating room are continuously arranged on a unified timescale axis. To avoid the impact of time drift, missing fields, duplicate writing, and interruptions on subsequent work segment segmentation, digital twin workload segment construction, and workload backlog segment identification, the access device sequentially performs time correction, field supplementation, duplicate record removal, and breakpoint continuation, and writes the results to the next processing area after each processing step. This implementation process includes the following steps: First, time correction is performed on the raw data stream to eliminate time discrepancies between physiological data, device operation data, and interaction data, ensuring that subsequent work segmentation is based on the same time axis. The input quantities are the interval start time, interval end time, physiological data set, and event data set in the raw data stream, where the event data set includes device operation data and interaction data. The access device first reads the time reference configuration written during deployment, uses the access device's local clock as the reference clock for the unified time axis, and then calculates the time deviation value according to the data source. Specifically, for data carrying both the source time and the receiving time, the difference between the receiving time and the source time is taken as the deviation value for a single record. For several consecutive records from the same source, the median value of the deviation values ​​of the most recent consecutive records is taken as the current correction amount for that source according to a preset statistical estimation rule, where the number of values ​​for the consecutive records is written into the time reference configuration. The access device then uses the current correction amount to perform a translation correction on the source time of the corresponding source to obtain the corrected time under a unified time scale axis. The corrected time is then used to rewrite the interval start time, interval end time, and time fields of each physiological data, device operation data, and interaction data in the original data stream. The corrected data stream is output and written to the field completion buffer for subsequent field completion reading. If a record is missing a source time, the receiving time is directly retained as the correction time and written as a substitute correction mark. If the number of consecutive records from the same source is insufficient to form the current correction amount, the most recent correction amount retained in the previous round of time correction is used. Subsequently, field completion is performed on the corrected data stream to eliminate object breaks caused by missing identifiers and incomplete attribution, enabling subsequent continuous data streams to be stably merged according to anesthesiologist identifiers, patient identifiers, and operating room identifiers. The input quantities are the anesthesiologist identifier, patient identifier, operating room identifier, physiological data set, and event data set in the corrected data stream. The access device first reads the pre-established attribution mapping table and the current surgical registration table. The attribution mapping table includes at least the correspondence between device number and operating room identifier, the binding relationship between wearable device and anesthesiologist identifier, and the binding relationship between current patient and operating room identifier. When there are missing fields in the original data stream, filling is performed according to a fixed completion order. The fixed completion order is as follows: first, read the field values ​​of the same interval that are not missing in the current original data stream; second, read the corresponding field values ​​of the same source in the previous continuous valid data stream; third, read the binding field values ​​in the attribution mapping table; and finally, read the registration field values ​​in the current surgical registration table. Among them, the anesthesiologist identifier can only be completed from the binding relationship of the wearer or the current login relationship of the anesthesia workstation, the patient identifier can only be completed from the current surgical registration form, and the operating room identifier can only be completed from the attribution mapping table corresponding to the device number or workstation number. After the completion is completed, the completed data stream is output and written to the deduplication buffer for subsequent duplicate record removal and reading. If the completion results given by multiple sources are inconsistent, the field value of the priority source is retained according to the above completion order, and the remaining field values ​​are written to the field conflict table for subsequent review, and are not directly written to the completed data stream. Further, duplicate record removal is performed on the supplemented data stream. The purpose is to eliminate redundant content caused by multiple uploads, delayed retransmissions, or cache write-backs, so as to avoid the same event or the same sampling point being counted repeatedly when segmenting and extracting subsequent work segments. The inputs are the time field, data type identifier, source identifier, and data value field in the supplemented data stream. The access device first expands the supplemented data stream according to the time order under a unified time scale axis, and then constructs a deduplication key for each physiological data, device operation data, and interaction data. The deduplication key includes the correction time, data type identifier, source identifier, and data value summary. The data value summary is taken as the combination of sampled value and data type for physiological data, the combination of event type and device number for device operation data, and the combination of event type, initiator marker, receiver marker, and round-trip sequence marker for interaction data. When two records have identical deduplication keys, they are considered duplicate records. Only the record written to the deduplication buffer is retained, and the remaining duplicate records are deleted. When the time field is the same but the source identifier or the data value digest is different, they are considered concurrent valid records and are not deleted. After deduplication is completed, the deduplicated data stream is output and written to the continuation buffer for subsequent breakpoints to resume reading. If a record is missing a field required for the data value digest, the missing digest marker is added first, and the record is retained without entering the duplicate deletion logic to avoid accidentally deleting valid records. Finally, the deduplicated data stream is reconnected to restore the data link before and after the acquisition interruption, ensuring that the continuous data stream remains readable under a unified time scale, while clearly recording the start and end positions of the interrupted segment. The input parameters are the time sequence records, source identifier, and interval boundary fields in the deduplicated data stream. The access device first calculates the time interval between two adjacent valid records according to the source identifier and compares this time interval with the hold duration in the reconnection configuration. The hold duration is taken from the reconnection configuration written during deployment. When the time interval between two adjacent valid records does not exceed the hold duration, it is determined to be a short-term interruption. The access device writes an interruption marker segment between the two valid records and reconnects the next record to the continuous segment where the previous record is located, forming a continuous data stream that maintains continuity. When the time interval between two adjacent valid records exceeds the hold duration, it is determined to be a long-term interruption. The access device writes an independent interruption segment marker between the interruption start point and the interruption end point, and uses the first valid record after the interruption end point as the start point of the new continuous segment to regenerate the continuous segment number. For all original data stream records within the same continuous segment, the access device writes the continuous segment number, the continuous segment start time, and the continuous segment end time, outputs the continuous data stream after the breakpoint continuation processing, and writes it to the segment generation module's read queue. If there is a supplementary transmission record within the interruption segment, the access device first inserts the supplementary transmission record back into the corresponding interruption segment according to the correction time, and then recalculates the time interval and continuous segment number to avoid erroneous breaks in the subsequent continuous data stream. Through the above processing, the access device eliminates time deviations, missing fields, duplicate records, and interruptions in the original data stream in sequence, and forms a continuous data stream with a unified time scale axis, complete belonging fields, no duplicate content, and continuous segment numbers, so that subsequent work segmentation can be based on a stable and executable data foundation. In practical applications: When anesthesiologists use a wearable device, physiological monitoring equipment, and anesthesia workstation simultaneously during surgery, the physiological data uploaded by the wearable device is delayed by two seconds due to local caching, the infusion device records lack patient identification, and nurse terminal messages are repeatedly uploaded due to network fluctuations. Furthermore, the monitor resumes uploading after a short network outage. The access device first corrects the time fields of the wearable device, physiological monitoring equipment, and workstation records based on the difference between the receiving time and the source time. Then, it completes the patient identification and operating room identification according to the wearable device binding relationship, the current surgical registration table, and the device number attribution table. Subsequently, it removes the repeatedly uploaded nurse terminal messages using the corrected time, data type identification, source identification, and data value summary. Finally, it determines that the monitor record belongs to a short-term interruption based on the time interval between the records before and after the network outage and writes it into the interruption marker segment. The recovered record is then continued into the original continuous segment, and finally, a continuous data stream is output for the segment generation module to read.

[0019] The segment generation module is used to read continuous data streams, take the time period between two adjacent device operation records, interaction records or alarm records as working segments, extract physiological change sequences, gaze switching sequences, limb operation sequences, device response sequences and collaborative round-trip sequences within each working segment, and output the segment result set. In this embodiment, the segment generation module is responsible for dividing the continuous data stream output by the access device into work segments that reflect the local workload status of the anesthesiologist, and forming a set of multiple sequences required for the subsequent construction of digital twin workload segments within each work segment. Its processing logic does not directly segment by fixed duration, but first constructs event boundaries using device operation records, interaction records, and alarm records. Then, within the time range defined by the boundaries, it extracts physiological change sequences, gaze switching sequences, limb operation sequences, device response sequences, and collaborative round-trip sequences. All types of sequences are written into the segment result set according to a unified time scale axis and a unified attribution key, thereby ensuring that subsequent item-by-item alignment, individual occupancy result solving, device traction result solving, and collaborative traction result solving are based on data with clear boundaries, complete sequences, and consistent timing. This implementation process includes the following: First, the continuous data stream is segmented into work segments. The purpose is to replace arbitrary time cuts with event boundaries that reflect the load triggering location, so that each work segment corresponds to a clear work progress process. The input quantities are the continuous segment number, continuous segment start time, continuous segment end time, physiological data set, and event data set in the continuous data stream. The event data set includes equipment operation records, interaction records, and alarm records. The segment generation module first performs a unified sorting of equipment operation records, interaction records, and alarm records within the same continuous segment number according to the correction time. Then, the time period between two adjacent event records is determined as a work segment. The correction time of the previous event record is used as the start time of the work segment, and the correction time of the next event record is used as the end time of the work segment. The work segment boundary adopts the rule of front-closed and back-open. When there are multiple event records at the same correction time, the multiple event records are merged into the same boundary point, and the merged boundary point is used as the start or end position of the work segment. When the start time of a continuous segment is earlier than the correction time of the first event record, the first working segment is formed from the start time of the continuous segment to the correction time of the first event record; when the end time of a continuous segment is later than the correction time of the last event record, the last working segment is formed from the correction time of the last event record to the end time of the continuous segment; when there are no device operation records, interaction records, or alarm records in a continuous segment, the segment generation module reads the silence duration written in the silence segment configuration, and divides the silent working segments according to the silence duration, starting from the start time of the continuous segment. The silence duration is taken from the preset configuration written during deployment based on the operating room data upload frequency; after processing, the working segments after time period division are output, and the segment number, continuous segment number, segment start time, segment end time, anesthesiologist identifier, patient identifier, and operating room identifier of each working segment are written into the segment boundary table for subsequent sequence extraction and reading; if there are event records with reversed time in a continuous segment, the reversed records are first written into the time anomaly table, and then rearranged according to the output order of the access device before continuing to divide, so as to avoid segment boundary interruption; After the work segment is divided, sequence extraction is performed on each work segment. The purpose is to convert the raw records in the continuous data stream into structured sequences that can directly participate in load alignment and load solution. The inputs are the segment boundary table, the physiological data set and the event data set in the continuous data stream. The segment generation module first extracts all physiological data and event data falling within the corresponding time range according to the segment start time and segment end time, and then generates five types of sequences. Among them, the physiological change sequence is formed by performing adjacent difference calculation, direction marking and duration statistics on the heart rate value, skin conductance value, fixation area code value and motion amplitude value arranged in ascending order of sampling time within the same work segment. The difference is the difference between the next sample value and the previous sample value. The direction marking is divided into rising, falling and holding. The duration is the number of sampling points that appear consecutively in the same direction. The gaze switching sequence is formed by performing adjacent comparisons on the gaze region code values ​​in the eye movement data. A gaze switching record is written once when the previous and next gaze region code values ​​are not simultaneously recorded, and the region before switching, the region after switching, and the switching time are recorded. The limb operation sequence is formed by matching the motion amplitude values ​​in the body movement data with the motion recognition rule table to identify operation categories such as touch screen, knob adjustment, infusion operation, and body position adjustment, and writes them into the limb operation record according to the operation start time, operation end time, and operation category. The motion recognition rule table is taken from the rule constraints written during deployment based on the installation method of the wearable sensor. The device response sequence is formed by performing consecutive matching on the device operation record, alarm record, and device status change record. That is, taking the previous device operation event or alarm event as the starting point, the status change record or alarm cancellation record with the same device number is searched in the preset response window. If found, a response completion record is written; if not found, a non-response record is written. The response window is taken from the communication refresh cycle in the device manual and written by the deployment configuration. Collaborative round-trip sequences are formed by recombining interaction records according to interaction topic number, initiator marker, receiver marker, and round-trip order marker. Sequences containing both an initiator record and a corresponding response record are written into a closed collaborative round-trip record; sequences containing only an initiator record without a response record are written into an unclosed collaborative round-trip record. After the five types of sequences are generated, segmentation and data integration are performed according to their respective event times, sampling times, or operation start and end times. These sequences are then uniformly written into the sequence buffer of the current working segment, and the corresponding sequence set is output for reading during segment result set generation. If a working segment contains a missing sampling point, that missing sampling point is only marked as missing in the physiological change sequence and does not participate in adjacent difference calculations. Difference calculations are resumed only after the next valid sampling point appears to avoid artificially amplifying the amount of change. After obtaining the sequence set of each work segment, the segment features of each type of sequence are processed and a segment result set is generated. The purpose is to unify and solidify the five types of sequences into a segmented input that can be directly read by the subsequent twin mapping module. The input consists of the physiological change sequence, gaze switching sequence, limb operation sequence, equipment response sequence, and collaborative round-trip sequence corresponding to each work segment. The segment generation module first performs time sequence verification on the five types of sequences within the same work segment, deletes duplicate sequence items caused by re-transmission and rewriting, and then calculates the temporal features within the segment for each of the five types of sequences. Specifically, the physiological change sequence calculates the number of rises, falls, holds, and longest duration of each type of physiological data; the gaze switching sequence calculates the number of switches and the number of switched coverage areas; the limb operation sequence calculates the number of operation categories, the number of operations, and the total duration; the equipment response sequence calculates the number of response completions, the number of non-responses, and the average response interval; and the collaborative round-trip sequence calculates the number of closed round-trips, the number of unclosed round-trips, and the longest round-trip duration. Based on this, the segment generation module performs weighted calculations according to the segment feature weight table to form the load performance field and temporal feature field within each working segment. The segment feature weight table is obtained by statistical estimation of historical surgical records before system deployment, and fixed weights are written for physiological change sequences, gaze switching sequences, limb manipulation sequences, equipment response sequences, and collaborative round-trip sequences, respectively. These weights are not adjusted during the current surgical procedure. The weighted calculation does not directly generate the final warning result, but is used to unify the representation scale of the five types of sequences within the same segment result set. Specifically, the temporal feature value of each type of sequence is multiplied by its corresponding fixed weight and then written into the load performance field within the segment. The unweighted original time-series feature fields are retained for subsequent twin mapping modules to read simultaneously. After processing, a fragment result set arranged in chronological order is output. The fragment result set includes at least the fragment number, fragment start time, fragment end time, anesthesiologist identifier, patient identifier, operating room identifier, five types of sequence sets, the time-series feature fields corresponding to the five types of sequences, and the load performance field within the fragment, and is written to the twin mapping module's read queue. If a certain type of sequence is empty in a certain working fragment, the null value mark of that type of sequence and the corresponding time-series feature null value mark are retained, and are not replaced with zero values, so that the subsequent twin mapping module can distinguish between the two cases of the absence of this type of behavior and the missing data of this type. Through the above processing, the fragment generation module first divides the continuous data stream into working fragments with clear boundaries, then transforms the original records within the working fragments into five types of sequences that can be aligned, statistically analyzed, and inherited, and finally forms a fragment result set with complete fields, clear order, and can be directly entered into the twin mapping process, thereby providing a stable input foundation for the subsequent construction of digital twin workload fragments. In practical applications: When an anesthesiologist first performs an infusion device parameter adjustment within the same continuous segment, then receives a fluid refill confirmation interaction record from the nurse's terminal, and then a blood pressure alarm record appears on the monitor, the segment generation module first forms two consecutive working segments based on the correction times of the parameter adjustment record, the fluid refill confirmation interaction record, and the blood pressure alarm record. Then, within each working segment, it extracts adjacent differences in heart rate and skin conductance values ​​to form a physiological change sequence, extracts records of the gaze area switching from the monitor interface to the infusion device interface to form a gaze switching sequence, extracts hand movements for adjusting the infusion device and touching the workstation interface to form a limb operation sequence, extracts records of state changes after infusion device parameter adjustment and records of alarm cancellation after blood pressure alarm to form a device response sequence, and extracts fluid refill confirmation requests and responses to form a collaborative round-trip sequence. Subsequently, it calculates the time-series characteristics such as the number of switching times, duration, number of response completions, and number of closed round-trips for the five types of sequences, and writes them into the load performance field within the segment according to the segment feature weight table. Finally, it outputs a segment result set for the twin mapping module to read.

[0020] The twin mapping module is used to read the segment result set, match the physiological change sequence in each work segment with the corresponding gaze switching sequence, limb operation sequence, equipment response sequence and collaborative round-trip sequence, construct the digital twin load segment corresponding to the current work segment, solve the individual occupancy result, equipment traction result and collaborative traction result, and continue to write the unreleased occupancy of the previous work segment into the digital twin load segment corresponding to the next work segment, and output the continuous load result set; In this embodiment, the twin mapping module is responsible for organizing the multiple sequences and intra-segment load performance fields in the segment result set into digital twin load segments that can be continuously inherited and compared segment by segment. It further outputs a continuous load result set for direct reading in subsequent load backlog segment identification and early warning determination. The processing order is not to summarize the five types of sequences in parallel, but rather to first lock the working segments under the same attribution link according to the anesthesiologist identifier, patient identifier, and operating room identifier. Then, within a single working segment, it completes the item-by-item alignment of multiple sequences and the solution of the three types of results. Subsequently, the occupancy status that has not yet been released in the previous working segment is continued into the next working segment. Finally, the comprehensive load assessment result is updated with consistency and conflict results, so that the continuous load result set can truly reflect the load evolution relationship of the anesthesiologist in the same surgical procedure. This implementation process includes the following steps: First, the fragment result set is assigned affiliation. The purpose of this is to lock fragmented data from different sources into the same load evolution link, so as to avoid data from different anesthesiologists, patients or operating rooms from crossing into the same digital twin load fragment. The input includes the fragment number, fragment start time, fragment end time, anesthesiologist identifier, patient identifier, operating room identifier, five types of sequence sets, time series feature field and load performance field within the fragment. The twin mapping module first constructs a ternary affiliation key according to the anesthesiologist identifier, patient identifier and operating room identifier, and then arranges each working fragment in ascending order according to the fragment start time under the same ternary affiliation key to generate the corresponding fragment dataset. When two adjacent work segments have the same ternary attribution key and the segment interval does not exceed the retention time in the continuation configuration, the two work segments are kept in the same segment dataset. The retention time is taken from the preset configuration written according to the regular recording interval of the operating room during system deployment. When the patient identifier or operating room identifier of two adjacent work segments changes, it is determined that the load attribution link has switched, and the work segment after the change is used as the starting segment of the new segment dataset. After processing, the segment dataset is output, which is grouped by the ternary attribution key and arranged in chronological order. The segment number, the sequence number in the group, and the ternary attribution key are written to the twin mapping buffer for subsequent item-by-item reading. If a work segment is missing a patient identifier or operating room identifier, the continuous data stream output by the access device and the current surgical registration table are checked to perform the kernel supplementation. If the kernel supplementation is successful, it is written to the corresponding segment dataset. If the kernel supplementation fails, an attribution pending mark is written and it is temporarily not involved in the subsequent solution. Next, multi-sequence alignment is performed within a single work segment. The aim is to decompose the temporal correlations between physiological changes, gaze shifts, limb manipulations, equipment responses, and collaborative reciprocal movements into three categories of results that directly characterize the anesthesiologist's workload. The input consists of the physiological change sequence, gaze shift sequence, limb manipulation sequence, equipment response sequence, and collaborative reciprocal movement sequence corresponding to the current work segment. The twin mapping module first reads the time alignment window from the alignment configuration, which is taken from the preset configuration written during deployment based on the equipment refresh cycle and terminal interaction delay. Then, it aligns the five types of sequences... Arranged in ascending order according to a unified time scale axis, and using each change record in the physiological change sequence as the primary alignment item, the nearest line-of-sight switching record, limb operation record, equipment response record, and collaborative round-trip record are searched within the time alignment window respectively; when a certain type of record falls within the time alignment window of the current primary alignment item, the record of this type is written as the same alignment group as the primary alignment item; when there are multiple records of the same type within the same time alignment window, the one with the smallest absolute value of the time difference with the primary alignment item is retained as the primary matching record, and the remaining records are retained as parallel associated records in the current alignment group; After forming the alignment group, the twin mapping module extracts the individual occupancy result, device traction result, and collaborative traction result according to the sequence calibration rules. The individual occupancy result is obtained by solving the same-direction change relationship and duration of the physiological change sequence, gaze switching sequence, and limb operation sequence in the same alignment group. If the heart rate value or skin conductance value increases continuously and is accompanied by an increase in the number of gaze switching and an increase in the duration of limb operation, then the alignment group is recorded as an enhanced individual occupancy record. The device traction result is obtained by solving the sequential relationship between the device response sequence, limb operation sequence, and gaze switching sequence. If a gaze switching towards the corresponding device and a limb operation targeting the device occur in the alignment window after the device alarms, then the alignment group is recorded as a device traction record. The collaborative traction result is derived from the switching relationship between the collaborative round-trip sequence, the line-of-sight switching sequence, and the limb operation sequence. If a workstation interactive operation or a line of sight shifting from the device interface to the collaborative terminal interface occurs simultaneously near the initiation or response time corresponding to the closed collaborative round-trip record, then the pairing group is recorded as a collaborative traction record. After processing, the individual occupancy result, device traction result, and collaborative traction result corresponding to the current work segment are output and written to the result temporary storage table of the current work segment for subsequent digital twin load segment construction. If a certain type of sequence is empty in the current work segment, then that type of sequence does not participate in the current round-trip pairing, but retains the null value mark and does not replace it with a zero value to avoid misjudging missing data as load release. Subsequently, a digital twin workload segment is constructed. The purpose is to merge the three types of results solved within a single work segment with the previous legacy state into a workload state body that can be inherited across segments. The input quantities are the segment number, segment start time, segment end time, individual occupancy result, equipment traction result, collaborative traction result, and the digital twin workload segment of the previous work segment. The twin mapping module first generates a digital twin workload segment for the current work segment. The digital twin workload segment includes at least the segment number, segment start time, segment end time, anesthesiologist identifier, patient identifier, operating room identifier, individual occupancy field, equipment traction field, collaborative traction field, unreleased occupancy amount field, and status field. For the individual occupancy results, equipment traction results, and collaborative traction results in the current segment, after writing them into the corresponding fields, it is then determined whether the unreleased occupancy of the previous working segment meets the continuation conditions. The continuation conditions include three items: the ternary attribution key of the previous working segment and the current working segment are consistent, the time interval between the two working segments does not exceed the hold duration in the continuation configuration, and there is no released flag in the status field of the previous working segment. When all three conditions are met, the unreleased occupancy of the previous working segment is written into the inheritance field of the current digital twin workload segment, and together with the newly solved individual occupancy results, equipment traction results, and collaborative traction results in the current segment, it forms the unreleased occupancy of the current segment. If the same type of result in the current segment continues to be maintained or enhanced, the inheritance status is retained. If the same type of result in the current segment no longer appears and continues to be missing for more than the release window in the release configuration, the inheritance status of this type is rewritten as a released flag. The release window is taken from the rule constraints written according to the anesthesia operation rhythm during deployment. After processing, the digital twin load segment corresponding to the current working segment is output and written to the continuous load splicing area for subsequent continuous load result set generation and reading; if the previous working segment does not exist or the current working segment is the first working segment of the new segment dataset, the digital twin load segment is directly initialized with the three newly solved results in the current segment, and cross-segment inheritance is not performed; Finally, a continuous load result set is generated. The purpose is to organize the segmented digital twin load segments into a continuous load chain that can directly reflect the overall evolution relationship and provide a comprehensive load assessment result. The input is the digital twin load segments arranged in chronological order within the same segment dataset. The twin mapping module first reads the individual occupancy field, equipment traction field, collaborative traction field, and unreleased occupancy field of each digital twin load segment in sequence according to the segment start time. Then, it performs a consistency check between the current digital twin load segment and the previous digital twin load segment. The consistency check includes at least source consistency check and time sequence continuity consistency check. The source consistency check is used to determine whether the equipment number and event topic pointed to by the equipment traction field and collaborative traction field in the current segment maintain the same propagation link as the inheritance field of the previous segment. The time sequence continuity consistency check is used to determine whether the load change of the current segment is consistent with the direction of the unreleased occupancy of the previous segment. After consistency verification, the twin mapping module calculates the comprehensive load assessment result of the current digital twin load segment. Specifically, it reads the load performance field within the segment from the segment result set as the basic load value within the segment. Then, it converts the number of records in the individual occupancy field, equipment traction field, and collaborative traction field of the current segment into corresponding incremental items. It converts the unreleased occupancy field inherited from the previous segment into inherited incremental items. Finally, it solves the comprehensive load assessment result of the current segment by adding the basic load value to each incremental item and subtracting the released items. Simultaneously, the twin mapping module updates the confidence result of the current segment. The confidence result is not an abstract score, but a confidence status field formed by the number of consistent records, conflicting records, and missing records. When the number of consistent records increases, a high confidence status is written; when the number of conflicting records increases, a conflict pending inspection status is written; and when the proportion of missing records exceeds the allowable ratio in the missing record configuration, a low confidence status is written. The allowable ratio is estimated based on historical data integrity statistics during system deployment. After processing, each digital twin load segment and its corresponding comprehensive load assessment result and confidence status field are concatenated into a continuous load result set in chronological order and written to the read queue of the early warning module. If the current segment has a conflict pending inspection status, the source of the conflict is written to the conflict verification table and the current continuous load result set is retained without interrupting subsequent output, so that the early warning module can continue to perform conflict resolution in conjunction with subsequent segments. Through the above processing, the twin mapping module first reorganizes the fragment result set according to the unified attribution link, then completes the multi-sequence alignment and three-class result solution within a single working fragment, then constructs an inheritable digital twin load fragment, and finally forms a continuous load result set unfolded in time order, so that the anesthesiologist's personal occupation, equipment traction, collaborative traction and their unreleased state in the same surgical process can be continuously presented, providing a direct basis for the identification of subsequent load backlog segments. In practical applications: When an anesthesiologist performs infusion equipment adjustment, monitor alarm handling, and fluid refill confirmation with the nurse terminal for the same patient in the same operating room, the twin mapping module first categorizes the corresponding work segments into the same segment dataset according to the anesthesiologist's identifier, patient's identifier, and operating room identifier. Then, within each work segment, it uses the heart rate rise record as the primary alignment item and matches the monitor interface line-of-sight switching records, infusion equipment adjustment operation records, alarm response records, and fluid refill confirmation round-trip records that appear at adjacent times to solve the individual occupancy result, equipment traction result, and collaborative traction result corresponding to the current work segment. Subsequently, the equipment traction status that has not yet been released in the previous work segment is continuously written into the digital twin load segment of the next work segment, and the unreleased occupancy field is updated in combination with the newly appearing collaborative traction record in the current segment. Finally, a continuous load result set containing digital twin load segments, comprehensive load assessment results, and reliable status fields is output in chronological order for the early warning module to continue to identify load backlog segments.

[0021] The early warning module is used to read the continuous load result set, compare the individual occupancy results, equipment traction results, collaborative traction results and unreleased occupancy in the digital twin load segments corresponding to adjacent work segments in chronological order, identify load backlog segments that continue to increase and remain connected in the next work segment, and generate early warning results and handling instructions based on the equipment source and collaborative source corresponding to the load backlog segment, and output early warning information. In this implementation, the early warning module is responsible for converting digital twin load segments arranged chronologically from continuous load results into executable early warning information. Its processing focus is not on making a one-time high / low judgment on individual work segments, but rather on first identifying the evolutionary relationship between individual occupancy results, equipment traction results, collaborative traction results, and unreleased occupancy amounts between adjacent work segments. Then, it extracts load backlog segments from the continuous evolutionary relationship, subsequently constructing a multi-objective decision chain around the equipment and collaborative sources corresponding to the load backlog segments. Finally, it eliminates conflicts through cross-perspective consistency verification, rollback re-checking, and gating token writing, outputting optimized early warning information with action guidelines. This ensures that the early warning results correspond to the actual load propagation process and can directly support human intervention. This implementation process includes the following: First, adjacent segments are compared in the continuous load result set. The purpose is to convert the discrete results in the digital twin load segments into load change data that can be identified by subsequent load backlog segments. The inputs are the segment number, segment start time, segment end time, individual occupancy result, equipment traction result, collaborative traction result, unreleased occupancy, comprehensive load assessment result, and confidence status field in the continuous load result set. The early warning module first sorts the execution time of the digital twin load segments according to the same continuous load link corresponding to the anesthesiologist identifier, patient identifier, and operating room identifier. Then, it takes the previous working segment and the current working segment as a set of comparison objects and calculates the difference in individual occupancy result, equipment traction result, collaborative traction result, and unreleased occupancy, respectively. The difference is calculated by subtracting the corresponding field value of the previous working segment from the corresponding field value of the current working segment. After obtaining the differences, the early warning module writes a change direction marker for each type of difference: a difference greater than zero is recorded as an increase, a difference equal to zero is recorded as maintaining, and a difference less than zero is recorded as a decrease. Then, the four types of change direction markers and the difference of the comprehensive load assessment result are written together into the load change table to form the load change data corresponding to each group of adjacent working segments. If the current working segment is the first working segment in a certain continuous load link, the difference field is initialized with a previous null value and an initial segment marker is written, but it does not directly participate in the identification of load backlog segments. If the confidence status field of a certain working segment is a low confidence status, its difference is still included in the calculation, but a low confidence marker is added to the load change table for subsequent multi-objective judgment chains to adjust its usage priority. After obtaining load change data, load backlog segment identification is performed on the working segments in the continuous load link. The purpose is to extract continuously propagating and unreleased load segments from local changes, avoiding misjudging short-term fluctuations as load anomalies requiring intervention. The inputs are the differences in individual occupancy results, equipment traction results, collaborative traction results, unreleased occupancy differences, change direction markers, and segment duration fields in the load change table. The early warning module first reads the backlog judgment table and continuous window configuration in the load identification configuration. The backlog judgment table is written before system deployment based on statistical estimation results from historical surgical records, and at least includes continuously increasing individual occupancy. The conditions for continuous increase of equipment traction, continuous increase of collaborative traction, and maintenance of unreleased occupancy are defined. The continuous window configuration includes at least the minimum number of consecutive segments and the shortest cumulative duration. Then, multiple adjacent working segments in the current continuous load link are used as candidate segments, and it is judged in turn whether the candidate segment meets the following conditions: First, at least one of the individual occupancy result, equipment traction result, and collaborative traction result has continuously increased or remained after increasing within the continuous window; second, the unreleased occupancy does not return to zero within the same continuous window; and third, the comprehensive load assessment result corresponding to the candidate segment is not lower than the comprehensive load assessment result of the previous candidate segment. When the above conditions are met simultaneously, the candidate segment is written as a load backlog segment, and the backlog start segment number, backlog end segment number, dominant change item, and cumulative duration are written in; when the unreleased occupancy in a candidate segment reaches zero or the comprehensive load assessment result continues to fall below the release condition in the backlog judgment table, the candidate segment is terminated and a released mark is written in it; after processing, the load backlog information is output and written to the early warning judgment buffer for subsequent multi-objective judgment chain reading; if there are working segments continuously occupied by low-confidence state fields in a continuous load link, these working segments are only allowed to be written as load backlog segments after the cumulative duration reaches the tightened duration in the continuous window configuration, so as to avoid misjudgment caused by missing data; After identifying the load backlog segment, a multi-objective judgment is performed on the load backlog segment. The purpose is to further expand the simple backlog identification into an early warning result with a clear source, clear trend and disposal direction. The inputs are load backlog information, equipment traction results in the continuous load result set, collaborative traction results, trusted status field, equipment source field, and collaborative source field. The early warning module first extracts the equipment source and collaborative source from each work segment covered by the load backlog segment. The equipment source is taken from the equipment number, equipment type and event type corresponding to the equipment traction result, and the collaborative source is taken from the interaction topic number, initiator mark and receiver mark corresponding to the collaborative traction result. Then, a multi-objective judgment chain is constructed according to the preset judgment chain configuration. The judgment chain configuration is written according to the rules during system deployment and includes at least the source concentration judgment, propagation duration judgment, occupancy superposition judgment and disposal priority judgment. Among them, the source concentration determination is formed by statistically analyzing the proportion of the number of times the same equipment source or the same collaborative source appears in the same load backlog segment to the total number of sources to form the dominant source result; the propagation duration determination is formed by comparing the cumulative duration of the load backlog segment with the graded duration in the duration window configuration to form the trend level; the occupancy superposition determination is formed by the number of work segments with the cumulative individual occupancy result, equipment traction result, and collaborative traction result simultaneously to form the superposition level; and the disposal priority determination is based on the dominant source result, trend level, and superposition level to find the corresponding disposal direction in the disposal mapping table. At the same time, the early warning module updates the confidence result. The update method is not to regenerate the abstract score, but to count the number of high confidence state segments, low confidence state segments, and conflict pending inspection state segments in the current load backlog segment. When the proportion of the number of high confidence state segments exceeds the high confidence ratio in the confidence configuration, a high confidence early warning state is written. When the proportion of the number of conflict pending inspection state segments exceeds the re-inspection ratio in the conflict configuration, a pending review early warning state is written. In other cases, a general confidence early warning state is written. The high confidence ratio and re-inspection ratio are both taken from the preset configuration written based on the historical data consistency statistical estimation results before system deployment. After processing is complete, an early warning message with execution guidelines is output. The early warning message includes at least the early warning time, anesthesiologist identifier, patient identifier, operating room identifier, load backlog range, dominant equipment source, dominant collaborating source, trend level, superposition level, early warning status, and treatment direction. It is written to the consistency verification cache for subsequent optimization processing. If no dominant source result is found for either the equipment source or the collaborating source in the same load backlog, the treatment direction is first written as manual review, and then corrected after the consistency verification result is returned. After generating the early warning information, cross-perspective consistency verification and conflict correction are performed on the early warning information. The purpose is to eliminate inconsistencies between source determination, trend determination, and handling direction, so that the final output early warning information remains consistent under multi-source evidence. The inputs are early warning information, load backlog information, individual occupancy results, equipment traction results, collaborative traction results, trusted status fields, and conflict verification tables. The early warning module first constructs a cross-perspective consistency verifier, which is pre-written according to rule constraints and includes at least source consistency verification, trend consistency verification, and handling consistency verification. The source consistency verification is used to determine whether the dominant equipment source is consistent with the high-frequency equipment number in the equipment traction result, and whether the dominant collaborative source is consistent with the high-frequency interaction topic in the collaborative traction result. The trend consistency verification is used to determine whether the trend level is consistent with the cumulative duration of the load backlog segment and the direction of change of the unreleased occupancy. The handling consistency verification is used to determine whether the handling direction is consistent with the handling mapping table corresponding to the dominant source result and the superimposed level. When all three types of consistency verification pass, the early warning module writes the pass gate token and directly writes the current early warning information into the final early warning result table. When any type of consistency check fails, the early warning module writes the failed gate token and triggers the rollback and re-check mechanism. The rollback and re-check mechanism is executed in a fixed order: first, it rolls back to the dominant source result to re-compare the source concentration; then it rolls back to the trend level to re-compare the propagation persistence; and finally, it rolls back to the handling mapping table to re-match the handling pointer. Each time a rollback and re-check is completed, the corresponding field is rewritten and the cross-perspective consistency check is re-executed. If the consistency check still fails after two consecutive rounds of rollback and re-check, the set of fields with the fewest conflicts is retained, a conflict retention flag and a manual review flag are written, and this is output as the optimized early warning information. After processing, the optimized early warning information is written to the final early warning result table and sent to the early warning display terminal and the management scheduling terminal for subsequent manual intervention and handling linkage reading. If a certain load backlog segment is entirely composed of low-confidence state segments, a low-confidence gate token and a manual review handling pointer are directly written, and the automatic handling mapping process is not entered. Through the above processing, the early warning module first extracts the load change trend between segments from the continuous load result set, then identifies the load backlog segment with continuous propagation characteristics, and then completes multi-target judgment around the equipment source and the collaborative source. It eliminates conflicts through cross-perspective consistency verification, rollback re-check and gating token writing, thereby outputting optimized early warning information with clear source, clear trend and actionable direction. In practical applications: When an anesthesiologist experiences an increase in equipment traction results due to monitor alarms, an increase in collaborative traction results due to nurse terminal fluid refill confirmation, and a continuous increase in individual occupancy results simultaneously across multiple work segments during the same patient's surgery, the early warning module first compares adjacent work segments to determine that the differences in equipment traction results, collaborative traction results, and unreleased occupancy are all increasing. Then, based on the continuous window configuration, this continuous segment is designated as a load backlog segment. Subsequently, the monitor number and fluid refill confirmation interaction topic are extracted from this load backlog segment as the equipment source and collaboration source, respectively. The dominant source result, trend level, and superposition level are calculated according to the multi-objective decision chain to generate the corresponding treatment direction. Then, cross-perspective consistency verification confirms that the dominant equipment source is consistent with the high-frequency source of equipment traction results, the dominant collaboration source is consistent with the high-frequency topic of collaborative traction results, and the trend level is consistent with the cumulative duration. Finally, the optimized early warning information is written through a gating token and output with manual support suggestions and operation load reduction prompts.

[0022] Working Principle: This solution first integrates the heart rate, skin conductance, eye movement, and body movement data output from the anesthesiologist's wearable device with the operation, alarm, and interaction records generated by the anesthesia machine, monitor, infusion equipment, surgeon's terminal, nurse's terminal, and anesthesia workstation. Then, through time correction, field completion, deduplication, and breakpoint continuation, the originally scattered and asynchronous data is organized into a continuous data stream. Subsequently, it segments the data according to the time range between adjacent device operations, interactions, or alarms, and extracts physiological changes, gaze shifts, limb movements, device responses, and collaborative back-and-forth sequences within each segment. Based on this, the system... The system systematically aligns these sequences one by one to construct a digital twin load segment corresponding to the current work segment. It determines whether the anesthesiologist is primarily experiencing increased personal workload, being driven by equipment events, or being driven by collaborative back-and-forth movements. It also extends the unreleased workload status from the previous work segment to the next work segment, forming a continuous load result set. Finally, the system compares the load changes of adjacent work segments in chronological order, identifies continuously increasing and unreleased load backlog segments, and generates early warning information by combining equipment source, collaboration source, consistency verification, and rollback re-check. This achieves a complete process from raw multi-source data to load identification and early warning response. For example, during a complex surgery, the anesthesiologist is simultaneously monitoring vital signs on the monitor, adjusting infusion device parameters, and responding to fluid resuscitation confirmation messages from the nurse. During this time, the wearable device continuously records the anesthesiologist's heart rate and body movement changes, the monitor uploads alarm records, the infusion device uploads parameter adjustment records, and the nurse's terminal and anesthesia workstation leave interaction records. The system first aligns these records according to a uniform time, then segments the period from monitor alarm to fluid resuscitation confirmation into a single work segment. Within this segment, it identifies frequent shifts in the anesthesiologist's gaze to the monitor and workstation, continuous hand manipulation of equipment, increased heart rate and skin conductance, and increased equipment responses and collaborative back-and-forth movements. Therefore, it determines that this workload is not a random fluctuation but a gradual buildup. If this state persists in subsequent work segments, the system identifies it as a workload buildup segment requiring intervention and provides clear prompts, such as that the current workload is primarily driven by monitoring device events and collaboration requests, necessitating prioritizing the reduction of unnecessary disturbances or the arrangement of auxiliary collaborations.

[0023] The above description is merely 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 anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data, characterized in that, include: The data acquisition device is used to collect heart rate data, skin conductance data, eye movement data, and body movement data of anesthesiologists; to collect operation records and alarm records of anesthesia machines, monitors, and infusion equipment; to collect interaction records of surgeon terminals, nurse terminals, and anesthesia workstations; and to output raw data streams according to the same time scale. The access device is used to perform time correction, field completion, duplicate record removal and breakpoint continuation on the raw data stream, and output a continuous data stream according to the anesthesiologist identifier, patient identifier and operating room identifier; The segment generation module is used to read continuous data streams, take the time period between two adjacent device operation records, interaction records or alarm records as working segments, extract physiological change sequences, gaze switching sequences, limb operation sequences, device response sequences and collaborative round-trip sequences within each working segment, and output the segment result set. The twin mapping module is used to read the segment result set, match the physiological change sequence in each work segment with the corresponding gaze switching sequence, limb operation sequence, equipment response sequence and collaborative round-trip sequence, construct the digital twin load segment corresponding to the current work segment, solve the individual occupancy result, equipment traction result and collaborative traction result, and continue to write the unreleased occupancy of the previous work segment into the digital twin load segment corresponding to the next work segment, and output the continuous load result set.

2. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 1, characterized in that: Also includes: The early warning module is used to read the continuous load result set, compare the individual occupancy results, equipment traction results, collaborative traction results and unreleased occupancy in the digital twin load segments corresponding to adjacent work segments in chronological order, identify load backlog segments that continue to increase and remain connected in the next work segment, and generate early warning results and handling instructions based on the equipment source and collaborative source corresponding to the load backlog segment, and output early warning information.

3. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 2, characterized in that: The data acquisition device includes: It receives heart rate data, skin conductance data, eye movement data, and body movement data output by the anesthesiologist's wearable device, performs sampling time registration, duplicate sampling point removal, and missing sampling point removal within the same sampling cycle, and outputs physiological data with sampling time and anesthesiologist identification. Receive operation records and alarm records output from anesthesia machines, monitors and infusion devices; receive interaction records output from surgeon terminals, nurse terminals and anesthesia workstations; perform event time extraction, event type marking, initiator marking, receiver marking and round-trip sequence marking; and output event data with event timestamps. Using the event time in the event data as the alignment reference, the data in the physiological data located between two adjacent event times is written into the corresponding event interval, and the physiological data and event data in the same event interval are merged according to the anesthesiologist identifier, patient identifier, and operating room identifier, and the original data stream is output.

4. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 3, characterized in that: The access device includes: Receive the raw data stream, perform time correction, eliminate data time deviation by synchronizing physiological data, device operation data and interaction data, and output the corrected data stream; The process involves performing field completion, which fills in missing fields in the original data stream. This is done by associating anesthesiologist identifiers, patient identifiers, and operating room identifiers to ensure the integrity and accuracy of the data stream, and then outputting the completed data stream.

5. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 4, characterized in that: The access device also includes: Remove duplicate records from the original data stream, detect redundant information in the data stream, remove duplicate content based on timestamps and data type identifiers, and output the deduplicated data stream; Execute breakpoint resume to connect the missing data caused by the interruption of acquisition with the data of the next acquisition, so as to ensure the continuity of the data stream and output a continuous data stream after breakpoint resume processing.

6. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 5, characterized in that: The fragment generation module includes: Read the continuous data stream, take the time period between two adjacent device operation records, interaction records or alarm records as the working segment, determine the segment boundary by comparing timestamps, and output the working segment after the time period is divided. Within each work segment, physiological change sequences, gaze switching sequences, limb operation sequences, equipment response sequences, and collaborative round-trip sequences are extracted. By segmenting and calibrating each sequence according to the event time and integrating the data, a corresponding sequence set is generated. The output fragment result set contains each working fragment arranged in chronological order, ensuring that the sequence data in each fragment is complete and without duplication. By performing weighted calculations on each sequence within a fragment, the load performance and temporal characteristics of each sequence within the working fragment are output.

7. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 6, characterized in that: The twin mapping module includes: Read the fragment result set, receive multi-source information from physiological data, behavioral data, equipment operation data and collaborative event data, and generate a fragment dataset by allocating work fragments according to anesthesiologist identifier, patient identifier and operating room identifier; The physiological change sequence in each work segment is matched with the gaze switching sequence, limb operation sequence, equipment response sequence, and collaborative round-trip sequence. Through time alignment and sequence calibration, the temporal consistency between each pair of data is ensured, and the corresponding individual occupancy result, equipment traction result, and collaborative traction result are extracted.

8. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 7, characterized in that: The twin mapping module also includes: Construct digital twin load segments, generate corresponding digital twin load state bodies within each working segment, ensure the continuity of load state by inheriting the unreleased occupancy of the previous working segment across segments, and write the inherited data into the digital twin load segment of the next working segment. Output a continuous load result set, combining individual occupancy results, equipment traction results, and collaborative traction results from the work segment. By fusing all load data and calculating confidence levels, a comprehensive load assessment result is generated, ensuring that the output result reflects the load evolution process throughout the anesthesia workflow.

9. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 8, characterized in that: The early warning module includes: Read the continuous load result set, receive the individual occupancy result, equipment traction result, collaborative traction result and unreleased occupancy in the digital twin load segment corresponding to each work segment, identify the load change trend by comparing the load data in adjacent work segments in time sequence, and output the load change data. By setting rules and threshold packages, load backlog segments in adjacent work segments are identified, it is determined whether the load continues to increase in the next work segment without being released, and a corresponding early warning signal is generated based on the judgment result, outputting load backlog information.

10. The anesthesiologist workload monitoring and early warning system based on multi-source physiological and behavioral data according to claim 9, characterized in that: The early warning module also includes: Based on the equipment and collaboration sources corresponding to the load backlog section, a multi-objective judgment chain is constructed. Through confidence update and time-series window judgment, the overall trend of load status is calculated, and accurate early warning results and handling directions are generated, outputting early warning information with execution guidance. By performing cross-perspective consistency checks and combining multi-source evidence from individual occupancy results, equipment traction results, and collaborative traction results, it is determined whether there are conflicts in the load data. If there are conflicts, they are corrected through a rollback and re-check mechanism, and data consistency is ensured through a gating token. Finally, optimized early warning information is output.

Images