Perinatal follow-up health record automatic filing method and system

Abstract

The present application belongs to the technical field of industrial data storage, and relates to a perinatal follow-up health record automatic filing method and system, which comprises the following steps: receiving an asynchronous data stream and converting it into a data frame, extracting a feature vector and a time stamp to generate a feature code; in the memory, a time slot serial number and the feature code are spliced by an algebraic addressing operator to generate an addressing token; an offset is calculated according to the token value and the memory address is corrected, and the rearranged data is stored in the high-speed buffer area by using a circular buffer array, wherein the present application translates the alignment demand into an addressing rule through the pre-token, eliminates the lock competition and transaction lock conflict in the writing path, avoids thread deadlock and channel blocking caused by asynchronous out-of-order back flushing, and realizes the lock-free double-track storage of high-concurrency streaming data.

Description

Technical Field

[0001] This invention belongs to the field of industrial data storage technology, and relates to an automated method and system for establishing perinatal follow-up health records. Background Technology

[0002] Currently, using digital models to dynamically monitor the health cycle and medical data of pregnant women has become a core approach to smart healthcare. However, existing technologies still have limitations in handling the continuity of processes in the health management of pregnant women, the reliable fusion of time-series multi-source data streams, and the privacy and security governance of multi-terminal interactions.

[0003] First, there is the health monitoring technology based on barcode interaction and multi-terminal binding. For example, Chinese invention patent CN120183750A discloses a health monitoring system applied to obstetrics and gynecology that covers the entire process of pregnant women's data management. It collects information from pregnant women, their families, and doctors, and generates QR codes based on the diagnosis and treatment data for dynamic updates across multiple terminals. Its shortcomings are: the solution is essentially based on instantaneous status feedback based on barcode scanning, and the data synchronization and multi-terminal interaction lack an automated closed-loop mechanism for high-frequency real-time streaming transmission; at the same time, the system relies heavily on fixed alarm thresholds or established empirical rules when conducting health monitoring, and lacks the ability to adaptively and deeply mine complex temporal physiological characteristics and predict potential disease trends.

[0004] Second, there is the multi-source health data fusion technology based on adaptive dynamic windows. For example, Chinese invention patent CN121256721A discloses a method and system for multi-source health data acquisition and dynamic fusion analysis. It uses an event-time-based adaptive dynamic window mechanism to align multi-source health data streams and uses a credit assessment model to dynamically calculate the credibility of data points and impute missing values. Its drawback is that this technology focuses on the status recognition and coarse data fusion of underlying IoT smart hardware in general chronic disease scenarios. It does not deeply bind and refine the clinical business process of pregnant women throughout the entire cycle in obstetrics and gynecology, and cannot effectively cope with the risk of disruption in the refined business flow when pregnant women move across platforms and institutions.

[0005] Third, electronic health record management technology based on event processing engines and optimized record scheduling; for example, Chinese invention patent application CN117854664A discloses an electronic pregnant woman health record management method and system, which uses a complex event processing engine for appointment, diagnosis update and treatment feedback monitoring, and optimizes the record organization structure through priority queues and genetic algorithms. Its shortcomings are: existing record management systems of this kind usually regard the health cycle as discrete or specific step-length event units, ignore the continuous impact of high-frequency physiological data from IoT devices over long time, and lack a real-time time alignment mechanism for high-frequency physiological indicators and a causal reasoning verification mechanism for conflicts in multi-source measurement readings; moreover, its security strategies are mostly limited to traditional basic encryption, and lack underlying dynamic desensitization control and robust privacy governance mechanisms when sharing and interacting across institutions, resulting in a high risk of data leakage.

[0006] Therefore, how to construct a health monitoring architecture with deep end-to-end business integration, reliable adaptive fusion of multi-source high-frequency physiological data, and underlying distributed privacy governance capabilities, so that health management and risk warning logic converges to the real medical safety boundary throughout the entire cycle, is the technical problem to be solved by this invention. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and solve the technical problems of lock contention, transaction lock conflict and channel blockage caused by high concurrency out-of-order backflush, and to provide an automated method and system for establishing perinatal follow-up health records.

[0008] To achieve the above-mentioned objectives, this invention provides an automated method for establishing perinatal follow-up health records, comprising the following steps: Step S1, extract features and timestamp to generate anonymous feature code: receive multi-source asynchronous data stream, parse the multi-source asynchronous data stream to separate application layer business fields and convert them into unstructured state data frames, extract the original identity feature vector and time-series metadata timestamp carried in the unstructured state data frames, perform feature mapping on the original identity feature vector through a one-way hash algorithm and generate a fixed-length anonymous feature code. Step S2, Concatenate features and time slots to generate address tokens: Establish a data mapping channel in memory and call the built-in algebraic addressing operator to divide the time-series metadata timestamp by the preset time segmentation span constant and round down to separate the discrete time slot sequence number. Concatenate the fixed-length anonymous feature code with the discrete time slot sequence number to generate the first hash value. Perform a bitwise XOR operation between the first hash value and the hardware inherent hash of the acquisition node to generate an integer spatiotemporal address token. Step S3, Correcting the address based on the token and rearranging the data to disk: Allocate physical memory space and build a high-speed buffer through the storage middleware. Perform algebraic modulo operation on the capacity of the circular buffer array based on the value of the integer spacetime addressing token to establish the slot physical index. Calculate the slot offset using the product of the slot physical index and the array's starting address. Correct the physical memory address based on the slot offset. Perform in-situ smooth rearrangement of the unstructured state data frame within the high-speed buffer using the circular buffer array. After calibrating the rearranged unstructured state data frame into a monotonically continuous linear sequence, store it on disk.

[0009] The refinement of step S3 in this invention includes the following sub-steps: Step S31, calculate the image address of the unstructured state data frame in the high-speed buffer using the slot offset, and write the unstructured state data frame into the high-speed buffer; Step S32, within the high-speed buffer, use a circular buffer array to perform in-situ smooth rearrangement of the asynchronously out-of-order unstructured state data frames, calibrating them into a monotonically continuous linear sequence.

[0010] The method described in this invention further includes step S4, which involves a cold / hot tiered storage migration process for data performed by a storage audit unit in a distributed industrial hybrid storage cluster. This process includes the following sub-steps: Step S41, the storage audit unit scans the update frequency vector of the global state bitmap and the block read / write count according to a preset period to obtain the read / write count of the data corresponding to the integer spatiotemporal addressing token within a continuous preset time period; Step S42, the storage audit unit compares the read / write count with a preset low-frequency threshold. When the read / write count is determined to be lower than the low-frequency threshold, the storage data state is switched. Historical fragmented data in the time-series storage area is aggregated using a merging algorithm, and the historical fragmented data is merged with the baseline snapshot data in the relational storage area to rewrite and generate a columnar storage block; Step S43, the storage audit unit migrates the columnar storage block to the magnetic storage cold area and establishes a reset pointer at the original index node to establish a single sequential read path for subsequent data queries.

[0011] The present invention refines step S42, which includes the following sub-steps: Step S421, the storage audit unit extracts the baseline snapshot data in the relational storage area as the reference axis, and arranges the historical fragment data in the time-series storage area according to the integer spatiotemporal address token for time alignment; Step S422, the storage audit unit merges the aligned historical fragment data into the columnar storage structure of the baseline snapshot data in an incremental append form, and rewrites to generate columnar storage blocks.

[0012] The method described in this invention further includes step S5, which involves performing a delay-compensated writing process on historical archived data through a data storage middleware. This process includes the following sub-steps: Step S51, the data storage middleware receives out-of-order unstructured state data frames and calculates the historical address token corresponding to the out-of-order unstructured state data frames; Step S52, the data storage middleware allocates delay-compensated storage slots in the high-speed buffer, introduces the main data block index head pointer and the incremental patch block mounted on the main data block index head pointer in the magnetic storage cold area, and writes the parameter information in the out-of-order unstructured state data frames into the incremental patch block in an incremental append form; Step S53, the data storage middleware updates the checksum in the global state bitmap through lock-free atomic bitmap operations to establish a first-order data check path.

[0013] The present invention refines step S52, including the following sub-steps: Step S521, the data storage middleware locates the main data block index head pointer in the magnetic storage cold zone and reads the mount path of the incremental patch block pointed to by the main data block index head pointer; Step S522, the data storage middleware writes the parameter information in the out-of-order unstructured state data frame into the incremental patch block and updates the file size attribute of the incremental patch block in the delay compensation storage slot.

[0014] The continuous preset time period described in this invention is 30 to 90 days, the low frequency threshold is no more than 5 times within the preset period, and the time segmentation span constant in step S2 is 86,400,000 ms.

[0015] After correcting the physical memory address based on the value of the integer spatiotemporal addressing token in step S3, the present invention further includes the following steps: storing the structured text data separated from the unstructured state data frame into a distributed relational storage engine, and storing the time-series parameter data extracted from the unstructured state data frame into a distributed time-series storage engine, and performing parallel dual-track persistent storage of the structured text data and the time-series parameter data.

[0016] After establishing a reset pointer at the original index node, the method described in this invention further includes the following query step: when a global history query request for an integer spatiotemporal addressing token is received, the query path is reset to the columnar storage block in the magnetic storage cold zone through the reset pointer, and a single continuous read addressing is performed on the columnar storage block in the magnetic storage cold zone.

[0017] This invention also provides an automated perinatal follow-up health record filing system, which is used to implement an automated perinatal follow-up health record filing method, including: The data access module is used to receive multi-source asynchronous data streams, parse the multi-source asynchronous data streams to separate application layer business fields and convert them into unstructured state data frames, extract the original identity feature vector and time-series metadata timestamp carried in the unstructured state data frames, perform feature mapping on the original identity feature vector through a one-way hash algorithm and generate a fixed-length anonymous feature code. The algebraic addressing module, connected to the data access module, is used to establish a data image channel in memory and call the built-in algebraic addressing operator to divide the time-series metadata timestamp by the preset time segmentation span constant and round down to separate the discrete time slot sequence number. The fixed-length anonymous feature code is concatenated with the discrete time slot sequence number to generate the first hash value. The first hash value is XORed with the hardware inherent hash of the acquisition node to generate an integer spatiotemporal addressing token. The in-situ rearrangement and disk-to-grid module, connected to the algebraic addressing module, is used to allocate physical memory space and build a high-speed buffer through the storage middleware. It performs algebraic modulo operation on the capacity of the circular buffer array based on the value of the integer spacetime addressing token to establish the physical index of the slot. It calculates the slot offset by multiplying the physical index of the slot by the first address of the array, corrects the physical memory address based on the slot offset, and performs in-situ smooth rearrangement of the unstructured state data frame in the high-speed buffer using the circular buffer array. After calibrating the rearranged unstructured state data frame into a monotonically continuous linear sequence, it is disk-to-grid-stored.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: 1. In the automated archiving of perinatal follow-up health records, by generating an integer spatiotemporal addressing token before writing data to the storage engine, the write position of perinatal follow-up data in the high-speed buffer and persistent storage area can be directly determined. Structured text data is written to the distributed relational storage engine using this token as the unique primary key, and time-series parameter data is appended to the distributed time-series storage engine using this token as the aggregated physical partition key. This reduces the need for frame-by-frame entity matching, cross-database storage queries, and transaction coordination, thereby reducing lock contention during high-concurrency writes, reducing the overhead of full table scans and random disk reads and writes, and improving the continuous write capability and storage efficiency of perinatal follow-up health records.

[0019] 2. By monitoring the latency between the system's physical master clock and the timestamp of the time-series metadata, when asynchronous out-of-order backflow occurs, unstructured state data frames that have not yet completed time-series calibration are transferred to a high-speed buffer. The circular buffer array is then used for in-situ smooth rearrangement. The rearranged data frames form a monotonically continuous linear sequence in chronological order and are then continuously written to the underlying storage blocks. This method can reduce transaction cross-waiting and write channel blocking caused by out-of-order data directly impacting the storage engine, reduce the risk of cascading deadlocks under high concurrency conditions, and ensure that perinatal follow-up data can still be written to disk in an orderly and stable manner when network latency fluctuates.

[0020] 3. By periodically analyzing the update frequency and block read / write count of the global status bitmap through the storage audit unit, historical archive data that has been in a low-frequency access state for a long time can be identified. The historical fragment data and baseline snapshot data that are scattered in storage can be centrally merged, rewritten to generate columnar storage blocks, and then migrated to the magnetic storage cold zone. This method can reduce the long-term occupation of high-performance storage resources by historical data and the space waste caused by fragmented storage. At the same time, by establishing a reset pointer at the original index node, subsequent historical queries can be directly located to the corresponding columnar storage block and read continuously, reducing cross-block random addressing and disk seek, and improving the archiving and query efficiency of historical archives. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the multi-source asynchronous data stream addressing and rearrangement disk writing process of the present invention; Figure 2 This is a schematic diagram of the interactive architecture of the automated filing system modules of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0023] Example 1: This embodiment discloses an automated method for establishing perinatal follow-up health records, including the following steps: Step S1, extract features and timestamp to generate anonymous feature code: receive multi-source asynchronous data stream, parse the multi-source asynchronous data stream to separate application layer business fields and convert them into unstructured state data frames, extract the original identity feature vector and time-series metadata timestamp carried in the unstructured state data frames, perform feature mapping on the original identity feature vector through a one-way hash algorithm and generate a fixed-length anonymous feature code. Step S2, Concatenate features and time slots to generate address tokens: Establish a data mapping channel in memory and call the built-in algebraic addressing operator to divide the time-series metadata timestamp by the preset time segmentation span constant and round down to separate the discrete time slot sequence number. Concatenate the fixed-length anonymous feature code with the discrete time slot sequence number to generate the first hash value. Perform a bitwise XOR operation between the first hash value and the hardware inherent hash of the acquisition node to generate an integer spatiotemporal address token. Step S3, Correcting the address based on the token and rearranging the data to disk: Allocate physical memory space and build a high-speed buffer through the storage middleware. Perform algebraic modulo operation on the capacity of the circular buffer array based on the value of the integer spacetime addressing token to establish the slot physical index. Calculate the slot offset using the product of the slot physical index and the array's starting address. Correct the physical memory address based on the slot offset. Perform in-situ smooth rearrangement of the unstructured state data frame within the high-speed buffer using the circular buffer array. After calibrating the rearranged unstructured state data frame into a monotonically continuous linear sequence, store it on disk.

[0024] The refinement of step S3 in this embodiment includes the following sub-steps: Step S31, calculate the image address of the unstructured state data frame in the high-speed buffer using the slot offset, and write the unstructured state data frame into the high-speed buffer; Step S32, in the high-speed buffer, use a circular buffer array to perform in-situ smooth rearrangement of the asynchronously out-of-order unstructured state data frames, and calibrate them into a monotonically continuous linear sequence.

[0025] The method described in this embodiment further includes step S4, which involves a cold / hot tiered storage migration process for data performed by a storage audit unit in a distributed industrial hybrid storage cluster. This process includes the following sub-steps: Step S41, the storage audit unit scans the update frequency vector of the global state bitmap and the block read / write count according to a preset period to obtain the read / write count of the data corresponding to the integer spatiotemporal addressing token within a continuous preset time period; Step S42, the storage audit unit compares the read / write count with a preset low-frequency threshold. When the read / write count is determined to be lower than the low-frequency threshold, the storage data state is switched. The historical fragmented data in the time-series storage area is aggregated by a merging algorithm, and the historical fragmented data is merged with the baseline snapshot data in the relational storage area to rewrite and generate a columnar storage block; Step S43, the storage audit unit migrates the columnar storage block to the magnetic storage cold area and establishes a reset pointer at the original index node to establish a single sequential read path for subsequent data queries.

[0026] The refinement of step S42 in this embodiment includes the following sub-steps: Step S421, the storage audit unit extracts the baseline snapshot data in the relational storage area as the reference axis, and arranges the historical fragment data in the time-series storage area according to the integer spatiotemporal address token for time alignment; Step S422, the storage audit unit merges the aligned historical fragment data into the columnar storage structure of the baseline snapshot data in an incremental append form, and rewrites to generate columnar storage blocks.

[0027] The method described in this embodiment further includes step S5, which involves performing a delay-compensated writing process on historical archived data through a data storage middleware. This process includes the following sub-steps: Step S51, the data storage middleware receives out-of-order unstructured state data frames and calculates the historical address token corresponding to the out-of-order unstructured state data frames; Step S52, the data storage middleware allocates delay-compensated storage slots in the high-speed buffer, introduces the main data block index head pointer and the incremental patch block mounted on the main data block index head pointer in the magnetic storage cold area, and writes the parameter information in the out-of-order unstructured state data frames into the incremental patch block in an incremental append form; Step S53, the data storage middleware updates the checksum in the global state bitmap through lock-free atomic bitmap operations and establishes a first-order data check path.

[0028] The refinement of step S52 in this embodiment includes the following sub-steps: Step S521, the data storage middleware locates the main data block index head pointer in the magnetic storage cold zone and reads the mount path of the incremental patch block pointed to by the main data block index head pointer; Step S522, the data storage middleware writes the parameter information in the out-of-order unstructured state data frame into the incremental patch block and updates the file size attribute of the incremental patch block in the delay compensation storage slot.

[0029] The continuous preset time period described in this embodiment is 30 to 90 days, the low frequency threshold is no more than 5 times within the preset period, and the time segmentation span constant in step S2 is 86400000ms.

[0030] In this embodiment, after correcting the physical memory address based on the value of the integer spacetime addressing token in step S3, the following steps are also included: storing the structured text data separated from the unstructured state data frame into a distributed relational storage engine, and storing the time-series parameter data extracted from the unstructured state data frame into a distributed time-series storage engine, and performing parallel dual-track persistent storage of the structured text data and the time-series parameter data.

[0031] After establishing a reset pointer at the original index node as described in this embodiment, the method further includes the following query step: when a global history query request for an integer spatiotemporal addressing token is received, the query path is reset to the columnar storage block in the magnetic storage cold zone through the reset pointer, and a single continuous read addressing is performed on the columnar storage block in the magnetic storage cold zone.

[0032] This embodiment also provides an automated perinatal follow-up health record filing system, including: The data access module is used to receive multi-source asynchronous data streams, parse the multi-source asynchronous data streams to separate application layer business fields and convert them into unstructured state data frames, extract the original identity feature vector and time-series metadata timestamp carried in the unstructured state data frames, perform feature mapping on the original identity feature vector through a one-way hash algorithm and generate a fixed-length anonymous feature code. The algebraic addressing module, connected to the data access module, is used to establish a data image channel in memory and call the built-in algebraic addressing operator to divide the time-series metadata timestamp by the preset time segmentation span constant and round down to separate the discrete time slot sequence number. The fixed-length anonymous feature code is concatenated with the discrete time slot sequence number to generate the first hash value. The first hash value is XORed with the hardware inherent hash of the acquisition node to generate an integer spatiotemporal addressing token. The in-situ rearrangement and disk-to-grid module, connected to the algebraic addressing module, is used to allocate physical memory space and build a high-speed buffer through the storage middleware. It performs algebraic modulo operation on the capacity of the circular buffer array based on the value of the integer spacetime addressing token to establish the physical index of the slot. It calculates the slot offset by multiplying the physical index of the slot by the first address of the array, corrects the physical memory address based on the slot offset, and performs in-situ smooth rearrangement of the unstructured state data frame in the high-speed buffer using the circular buffer array. After calibrating the rearranged unstructured state data frame into a monotonically continuous linear sequence, it is disk-to-grid-stored.

[0033] Example 2: In this embodiment, a distributed industrial hybrid storage cluster continuously receives high-frequency time-series physiological parameters and discrete clinical texts generated by different acquisition nodes. The physical clocks of each acquisition node deviate, and data frames arrive asynchronously via different network paths. It is difficult to directly establish a stable physical storage correspondence between the time-series parameter data and the structured text data. When using a processing method that first queries the entity identity, then requests a cross-database storage lock and writes the data, the increased number of concurrent nodes causes the storage engine to frequently wait for mutex locks, amplifying disk read / write operations and leading to write channel blockage and broken follow-up record data links. This embodiment generates a complete set of data before the data enters the storage engine. A time-space addressing token is used to determine the write position of the data frame in the high-speed buffer and persistent storage area. After the network receiving port receives the multi-source asynchronous data stream, the data bus reads the field identifier, field length and field value according to the access protocol, separates the application layer business fields from the original message, and writes them into memory according to a unified data frame format. The unstructured state data frame includes a frame header and a data area. The frame header records the data source identifier, time-series metadata timestamp and the data area length. The data area saves the field identifier, field length and field value according to the order in which the fields appear in the original message. Different data sources are not required to use the same relational field structure.

[0034] The compute node reads the original identity feature vector and time-series metadata timestamp from the unstructured state data frame. The original identity feature vector is formed by concatenating the ID card number, medical card number, and follow-up sequence number in a fixed field order. The compute node uses a 256-bit secure hash algorithm to perform feature mapping on the original identity feature vector, generating a fixed-length anonymous feature code of 32 bytes. The time-series metadata timestamp uses the millisecond integer value corresponding to the physical sampling time, with a value range of 0 to 20000000000000. The compute node establishes a data mapping channel in memory, which consists of contiguous memory pages and records... The starting address, data length, and corresponding time-series metadata timestamp of each unstructured state data frame are obtained. The fixed-length anonymous feature code, time-series metadata timestamp, and data frame starting address are then passed to the algebraic addressing operator. The algebraic addressing operator first divides the time-series metadata timestamp by the time-slicing span constant and rounds down to obtain the discrete time slot sequence number. The time-slicing span constant is fixed at 86,400,000 ms. After the discrete time slot sequence number is encoded with a fixed byte length, it is concatenated with the 32-byte fixed-length anonymous feature code. Then, a 256-bit secure hash operation is performed on the concatenation result to generate a 32-byte first hash value.

[0035] During initialization, the acquisition node reads its stable hardware identifier and generates a 32-byte hardware-intrinsic hash using the same 256-bit secure hash algorithm. The first hash value and the hardware-intrinsic hash are then XORed bitwise according to their byte positions. The XOR result is interpreted as an unsigned integer value according to the byte order uniformly adopted by the cluster, yielding an integer spacetime address token. The calculation relationship is as follows: ,in, It is an integer spacetime addressing token. For fixed-length anonymous signatures, For time-series metadata timestamps, The time partitioning span constant, This refers to the discrete-time slot sequence number. For string concatenation operations, For bitwise XOR operation, For the inherent hash of the acquisition node's hardware, the above operations use a fixed field order, byte length, and byte order. The same set of original identity feature vectors, time-series metadata timestamps, and acquisition node hardware identifiers correspond to the same integer spatiotemporal addressing token. The storage middleware allocates physical memory space and constructs a high-speed buffer. A circular buffer array is set up in the high-speed buffer. Each array slot stores the address and length of the unstructured state data frame. When the same slot corresponds to multiple data frames, the addresses of each data frame are stored in the slot through a data frame linked list. The storage middleware moduloes the integer spatiotemporal addressing token with the capacity of the circular buffer array, and the remainder is used as the physical index of the slot. Then, based on the first address of the array, the physical index of the slot is multiplied by the number of bytes occupied by a single slot to obtain the slot offset. The first address of the array is added to the slot offset to obtain the image address of the unstructured state data frame in the high-speed buffer.

[0036] After unstructured state data frames are written to the image address, the circular buffer array is smoothly rearranged in situ based on the timing metadata timestamp. Multiple data frames within the same slot are arranged in ascending order of timing metadata timestamp. When timestamps are the same, they are arranged according to the order in which the data frames arrived at the high-speed buffer. When refreshing data, starting from the slot corresponding to the earliest timing metadata timestamp, valid slots are read sequentially along the circular buffer array, and the data frames within each slot are output in order. This ensures that the timing metadata timestamps in the output result remain non-decreasing, and that each data frame is arranged continuously in the output memory area, thus forming a monotonically continuous linear sequence. The system monitoring module synchronously calculates the current system physical master clock and timing metadata... When the absolute latency difference between data timestamps does not exceed 3000ms within the normal network latency disturbance window, the storage middleware, after completing the calculation of the slot physical index and the correction of the physical memory address, directly reads the sorted data frame in that slot and executes disk write. When the absolute latency difference exceeds 3000ms but does not exceed the time limit threshold of 300000ms, the foreground blind insertion write path pauses refresh and continues to write subsequent arriving data frames to the corresponding slot, with the circular buffer array completing the in-situ smooth rearrangement. After the absolute latency difference falls back to within 3000ms, the background thread continuously writes the rearranged data frames in the high-speed buffer to the underlying storage block according to the monotonically continuous linear sequence arrangement.

[0037] After physical memory address correction and in-situ smooth rearrangement are completed, the storage middleware separates structured text data and time-series parameter data from the unstructured state data frame and performs parallel dual-track persistent storage. The structured text data includes identity-based fields and discrete clinical follow-up fields, which are written to the distributed relational storage engine using an integer spatiotemporal addressing token as the unique primary key. The time-series parameter data includes continuously collected dynamic vital signs parameters, which are appended to the physical time-series block of the distributed time-series storage engine using an integer spatiotemporal addressing token as the aggregated physical partition key. The logical physical partition index value corresponding to the physical time-series block is calculated according to the following formula: ,in, For logical physical partition index values, For Cyclic Redundancy Check (CRC) hash operators, This is a constant representing the total number of physical partitions in the cluster, with a fixed value of 1024. The time slot width is a constant, fixed at 10000ms. mod is the modulo operation. When the logical physical partition index value is equal to or greater than the total number of physical partitions, the modulo operation is performed on 1024, and the remainder is used as the actual physical partition number to be written. Data within the same time shard is thus concentrated and written to the physical time block within a limited range. Structured text data and time parameter data are persisted along the relational storage path and the time storage path, respectively.

[0038] At the underlying storage routing layer, a mapping relationship consisting of contiguous memory pages is established between the data mapping channel, the high-speed buffer, and the physical storage blocks. Each memory page records the corresponding physical storage block and its page offset. The storage middleware directly determines the write address of the data frame based on the integer spatiotemporal addressing token and the slot offset. Each physical storage page is pre-allocated with an address range corresponding to the access period. Structured text data is subjected to primary key blind interpolation, and time-series parameter data is subjected to partition append writing. When the data frame is written to disk, it is arranged in the order of physical storage page and time slot. The application layer no longer performs cross-table entity matching for each data frame. Each integer spatiotemporal addressing token corresponds to a bit in the global state bitmap. This bit is used to record the archiving status and solidification status of the corresponding data. Every time the bit is updated, the corresponding 8-bit counter increments by 1. The storage audit unit reads the change value of each counter at a preset period and forms the update frequency vector of the global state bitmap accordingly. The block read and write count is summarized by the storage engine according to the integer spatiotemporal addressing token, which summarizes the number of reads and writes of the corresponding storage block. The sum of the number of reads and writes is used as the read and write count of the data corresponding to the token.

[0039] The storage audit unit scans and updates the frequency vector and block read / write counts over a continuous preset period of 30 to 90 days. When the read / write count of the data corresponding to the integer spatiotemporal address token within the preset period is no more than 5 times, the storage audit unit initiates cold / hot tiered storage migration. During processing, a temporary alignment buffer is first allocated in memory, and historical fragmented data read from the time-series storage area is time-aligned and arranged according to the integer spatiotemporal address token and its time-series metadata timestamp. Then, the corresponding baseline snapshot data is read from the relational storage area, and this baseline snapshot data is used as the reference axis. The storage audit unit uses a dual-pointer scanning method to compare the baselines sequentially. The snapshot data and the aligned historical fragment data are incrementally appended to the columnar storage structure of the baseline snapshot data. After merging, the compression interface of the underlying file system is called to remove redundant null fields in the columnar storage structure and rewrite to generate a columnar storage block. This columnar storage block is migrated to the magnetic storage cold area through a single continuous input / output operation, and a reset pointer is established at the original index node. When a global historical query request for an integer spatiotemporal addressing token is received, the query path is located to the columnar storage block in the magnetic storage cold area through the reset pointer, and a single continuous read addressing is performed on the columnar storage block.

[0040] After adopting the above processing method, the memory location and persistence location of structured text data and time-series parameter data are determined by integer spatiotemporal addressing tokens before writing. Under normal latency, the data frame is directly written to disk after address correction. When asynchronous out-of-order backflush occurs, the data frame is smoothly rearranged in place in the high-speed buffer and then written in order. During the writing process, it is not necessary to perform cross-database queries and transaction lock waits for each data frame. Disk read and write is changed from random addressing to continuous writing within physical storage blocks. After the archived data is migrated to the magnetic storage cold area, continuous reading is completed by resetting the pointer, thereby maintaining the storage continuity of perinatal follow-up health records under high concurrency and network latency fluctuation conditions.

[0041] Example 3: In this embodiment, the experimental platform consists of eight distributed storage hosts, each connected via a network switch with a transmission bandwidth of 10Gbps. The test data is taken from the follow-up vital sign monitoring dataset, including continuously collected time-series parameter data and discrete clinical structured text data. The physiological sampling frequency of the network receiving port is set to 100Hz, and the number of concurrent connections per session is increased to 5000. Combined with the front-end data throughput rate and the physical block allocation cycle of the underlying storage engine, the time-series slot width is kept constant. With a slot width set to 10000ms, as the slot width continues to shrink, the number of index entries in the circular buffer array increases, and the number of memory addressing times increases accordingly. When the slot width increases, the dwell time of the data frame in the high-speed buffer is extended. In the test, Gaussian white noise with a signal-to-noise ratio of 25dB is superimposed on the receiving path of the multi-source asynchronous data stream, and dynamic delays from 0ms to 350000ms are injected through a network blocking simulator to form a multi-source asynchronous out-of-order input sample group.

[0042] The test program calculates the absolute delay difference between the current system physical master clock and the timestamp of the timing metadata carried in the unstructured state data frame. Based on this, three test gradients were set. The absolute latency difference of the first gradient was no more than 3000ms, the second gradient was 50000ms, and the third gradient was 300000ms. The sample group and the control group of this embodiment were run simultaneously under each gradient. The sample group of this embodiment determined the write position in the high-speed buffer according to the integer spatiotemporal address token, performed in-situ smooth rearrangement of the out-of-order unstructured state data frames, and then performed parallel dual-track persistent storage of structured text data and time-series parameter data. The control group canceled the generation of integer spatiotemporal address tokens and the pre-addressing processing, and instead had the storage middleware perform cross-database joint transaction queries in the foreground before writing the data.

[0043] Under the first gradient, the arrival order of multi-source asynchronous data streams changes little. In this embodiment, the sample group calculates the slot physical index and slot offset based on the integer spatiotemporal addressing token, corrects the physical memory address, and completes the data writing, with a disk write throughput of 12450.6 MB / s. The disk write throughput of the control group is 8920.4 MB / s, and no data write failures occur in either group under this gradient. Under the second gradient, the cross-database joint transaction in the control group experiences cross-waiting, the number of mutex locks in the foreground write path increases, and the disk write throughput drops to 3450.2 MB / s. In this embodiment, the sample group still writes structured text data and time-series parameter data to the distributed relational storage engine and the control group respectively according to the integer spatiotemporal addressing token. The distributed time-series storage engine achieved a disk write throughput of 12100.8 MB / s. Under the third gradient, concurrent transactions in the control group resulted in deadlock, storage channel congestion, and disk write throughput dropped to 420.5 MB / s. In this embodiment, the sample group wrote out-of-order unstructured state data frames into a high-speed buffer. The slot physical index was obtained by taking the modulo of the ring buffer array capacity based on the integer spatiotemporal address token, and the physical memory address was corrected using the slot offset. The ring buffer array performed in-situ smooth rearrangement of the data frames according to the time-series metadata timestamps, calibrated them into a monotonically continuous linear sequence, and then wrote them to disk. The disk write throughput remained at 10890.3 MB / s, and the archiving success rate remained above 99.85%.

[0044] In the out-of-range latency test, after increasing the absolute latency difference to 450,000 ms, the unwritten data in the circular buffer array exceeded the available slot capacity. Later arriving data began to overlap with slots that had not yet been refreshed, and memory utilization rose to 98.4%. Due to buffer space occupancy and slot conflicts, the physical addressing latency of the foreground write path increased from 12.3 ms to 465.8 ms. The physical partition boundary test used the total number of physical partitions in the cluster as a constant. Using a configuration of 1024 as a baseline, After reducing the value to 64, the available modulo remainder when mapping integer spacetime addressing tokens to physical partitions decreases, the number of times different tokens fall into the same physical partition increases, local physical partitions continuously accept concurrent writes, and the actual disk throughput drops to 1150.7 MB / s; in the addressing calculation at specific sampling points, fixed-length anonymous signatures are read from the test records. and the hardware-inherent hash of the acquisition node Time-series metadata timestamp Take 1717958400000ms as the time slice span constant. The test program calculates the integer spacetime address token according to a predetermined string concatenation, hashing, and bitwise XOR order, taking a time interval of 86,400,000ms. Combined with the aforementioned timing slot width constant and the total number of physical partitions constant Calculate physical partition index value The resulting index points to the 512th time-series storage block.

[0045] The normal network latency disturbance window is determined based on the 99% confidence interval of packet transmission latency in a typical wireless follow-up network. If the window is set too small, regular latency fluctuations in the network will frequently trigger buffer writes and foreground path switching. The time limit threshold corresponds to the longest latency range that the buffer can accommodate out-of-order data frames within its current capacity. Exceeding this range will cause slot overlap and memory overload in the circular buffer array. The total number of physical partitions is constant and matches the file allocation blocks of the distributed file system. When the number of partitions continues to be lower than the aforementioned reference value, hash collisions and hotspot partition writes further increase; when the number of partitions is higher than 4096, the number of partition index entries and memory addressing counts will increase accordingly. The increased time slot width constant is used to maintain a balance between data residence time in the high-speed buffer, the number of indexes, and the disk write frequency. In the above test, as network latency increased, the disk write throughput of the control group continued to decrease, and transaction deadlock occurred under high latency conditions. In this embodiment, the sample group determines the write address through integer spatiotemporal addressing tokens, uses a circular buffer array to complete the in-situ smooth rearrangement of out-of-order data, and performs parallel dual-track persistent storage for structured text data and time-series parameter data. Within the test range that does not exceed the high-speed buffer capacity boundary, data frames can be continuously written to disk in time sequence, and the foreground write path no longer depends on frame-by-frame cross-database joint transaction queries.

[0046] Example 4: This embodiment combines Figures 1 to 2 This document explains the methods and systems for automating the creation of perinatal follow-up health records, such as... Figure 1 As shown, the filing method includes steps S1, S2 and S3 that flow in sequence. Step S1 extracts features and timestamps to generate anonymous feature codes. Then, the processing logic is passed to step S2 to perform feature splicing and time slot generation to generate address tokens. Finally, the output is sent to step S3 to perform address correction based on the tokens and rearrange the disk.

[0047] like Figure 2As shown, the acquisition node inputs the signal to the receiving multi-source asynchronous data stream processing unit. This receiving multi-source asynchronous data stream processing unit sequentially transfers the signal to the unstructured state data frame conversion processing unit, the fixed-length anonymous feature code generation processing unit, the integer spatiotemporal addressing token generation processing unit, and the in-situ smooth rearrangement processing unit. Finally, the processing result is transmitted to the dual-track persistent storage processing unit. In this processing path, the storage middleware establishes interactive connections with the receiving multi-source asynchronous data stream processing unit and the dual-track persistent storage processing unit, respectively. At the same time, the storage audit unit establishes a connection with the cold and hot hierarchical storage migration processing unit to achieve corresponding data management collaboration.

[0048] Example 5: In this embodiment, the network receiving port of the distributed computing system continuously receives multi-source asynchronous data streams during perinatal follow-up. The data bus parses each data frame, separates application layer business fields, and converts the received content into unstructured state data frames. The computing node extracts the original identity feature vector and time-series metadata timestamp from the unstructured state data frames. A one-way hash algorithm is used to perform feature mapping on the original identity feature vector to generate a fixed-length anonymous feature code. In memory, the computing node establishes a data mapping channel and calls the algebraic addressing operator. The time-series metadata timestamp is divided by the time-slice span constant and rounded down to obtain the discrete time slot sequence number. The fixed-length anonymous feature code and the discrete time slot sequence number are then concatenated. A hash operation is performed on the concatenation result to generate a first hash value. The first hash value is then XORed with the inherent hash of the acquisition node's hardware to generate an integer spatiotemporal addressing token. The calculation relationship is as follows: ,in, It is an integer spacetime addressing token. For fixed-length anonymous signatures, This is a timestamp for time-series metadata, with a value range of 0 to 20000000000000, and the unit is ms; This is the time segmentation span constant, with a fixed value of 86,400,000 ms; This refers to the discrete-time slot sequence number. For string concatenation operations, For bitwise XOR operation, This is the hardware-inherent hash of the data acquisition node.

[0049] The storage middleware separates the structured text data from the time-series parameter data based on the attributes of the fields in the unstructured state data frame. The structured text data is written to the distributed relational storage engine using an integer spatiotemporal address token as the unique primary key; the time-series parameter data is appended to the distributed time-series storage engine using the same token as the aggregated physical partition key. Both types of data are written in parallel through their respective storage paths, without needing to query the entity identity in the other storage engine beforehand. The distributed time-series storage engine determines the physical time-series block based on the integer spatiotemporal address token, and its logical physical partition index value is calculated using the following formula: ,in, For logical physical partition index values, For Cyclic Redundancy Check (CRC) hash operators, This is a constant representing the total number of physical partitions in the cluster, with a fixed value of 1024. The timing slot width is a constant, fixed at 10000ms, and bmod is the modulo operation. During data writing, the network status monitoring unit calculates the absolute delay difference between the system physical master clock and the timing metadata timestamp in real time. When the absolute delay difference does not exceed 3000ms, the storage middleware determines the writing position according to the integer spatiotemporal address token and writes the unstructured state data frame to the corresponding storage path. When the absolute delay difference exceeds 3000ms but does not exceed 300000ms, the system determines that the multi-source asynchronous data stream has out-of-order backflow, suspends the foreground writing to the main path, and transfers the subsequently arriving unstructured state data frames to the high-speed buffer.

[0050] A circular buffer array of length 3600 is allocated within the high-speed buffer. The storage middleware modulo the value of the integer spacetime addressing token with the capacity of the circular buffer array, and uses the remainder as the physical index of the slot. Based on the array's starting address, the slot offset is determined by multiplying the slot physical index by the number of bytes occupied by a single slot, and this slot offset is used to correct the physical memory address before writing the unstructured state data frame to the corresponding slot. When multiple unstructured state data frames exist in the same slot, the compute node performs linear scan and insertion sort based on the time-series metadata timestamps, and adjacent slots are read according to their respective recorded time-series metadata timestamps. First, data frames with smaller timestamps are output. Data frames with the same timestamp are arranged in the order they entered the high-speed buffer. After in-situ smooth rearrangement, each data frame is stored continuously in the non-decreasing order of the time-series metadata timestamps, forming a monotonically continuous linear sequence. When the absolute latency difference falls back to within 3000ms, the background thread reads the unstructured state data frames according to the arrangement order in the circular buffer array and continuously writes the rearranged data sequence into the underlying time-series storage block. Structured text data is synchronously written to the distributed relational storage engine, and time-series parameter data is written to the distributed time-series storage engine, thus completing parallel dual-track persistent storage.

[0051] When the logical physical partition index value reaches or exceeds 1024, the distributed time-series storage engine performs a modulo operation on the index value again at the driver layer, using 1024 as the modulo, and uses the remainder as the actual physical partition number, keeping the actual physical partition number within the range of 0 to 1023. This wrapback process is only used to determine the physical partition in the distributed time-series storage engine and does not change the slot physical index determined by the integer space-time addressing token in the high-speed buffer. Through the above processing, the data frames of each follow-up file are first determined by the integer space-time addressing token to determine the memory write location and persistent storage location, and then the out-of-order data is smoothly rearranged in place by the circular buffer array. Under normal latency, the data frames are directly written to the corresponding storage path. When out-of-order backflush occurs, the data frames are sorted in the high-speed buffer and then continuously written to disk. The distributed relational storage engine uses primary key writing, and the distributed time-series storage engine uses partition append writing. The foreground concurrent writing process no longer relies on frame-by-frame cross-database queries and transaction lock waits.

[0052] Example 6: In this embodiment, the acquisition unit continuously generates high-frequency time-series physiological parameters and discrete clinical text. The storage middleware converts the received multi-source asynchronous data streams into unstructured state data frames and determines the memory write location of the data frames based on integer spatiotemporal addressing tokens. When concurrent data frames arrive in a concentrated manner, the high-speed buffer first accepts data that has not yet completed timing calibration, avoiding mutex lock waiting caused by multiple write threads entering the underlying storage engine simultaneously. The computing node allocates 64MB of contiguous physical memory as a high-speed buffer at the data access layer and divides it into a circular buffer array containing 3600 fixed slots. Each slot is set with a head pointer pointing to the linked list of unstructured state data frames written to that slot. When multiple data frames are received in the same slot, they are processed according to... Data frames are linked to a linked list in the order they enter the high-speed buffer. The physical input of the data access module is connected to the network receiving port of the distributed network to receive multi-source asynchronous data streams. Its output is connected to the data image channel of the algebraic addressing module through a serial peripheral interface bus. The algebraic addressing module is equipped with an arithmetic logic unit to perform time slicing, string concatenation, hashing, and bitwise XOR operations. The output of the arithmetic logic unit is connected to the memory physical space control terminal of the in-situ rearrangement and disk writing module through a high-speed external bus. After receiving the integer time-space addressing token, the in-situ rearrangement and disk writing module drives the memory controller of the high-speed buffer to correct the physical memory address and writes the data frames that have completed in-situ smooth rearrangement to the underlying storage block through the high-speed storage bus.

[0053] After receiving data at the network receiving port, the data access module parses the application layer business fields, converts the received content into unstructured state data frames, and extracts the original identity feature vector and time-series metadata timestamp. The algebraic addressing module generates an integer spatiotemporal addressing token based on the fixed-length anonymous feature code, discrete-time slot sequence number, and the hardware-inherent hash of the acquisition node. This token is then passed to the in-situ rearrangement disk module. The system monitoring module synchronously calculates the absolute delay difference between the system physical master clock and the time-series metadata timestamp, using the following formula: ,in, For the system's physical master clock, For time-series metadata timestamps, when the absolute latency difference does not exceed 3000ms, unstructured state data frames directly enter the subsequent write process according to the address determined by the integer spatiotemporal addressing token; when the absolute latency difference exceeds 3000ms but does not exceed 300000ms, the routing mechanism suspends foreground writing to the main path and transfers the data frame to the high-speed buffer; the in-situ rearrangement disk writing module performs a modulo operation on the value of the integer spatiotemporal addressing token with the capacity of the circular buffer array 3600 to obtain the slot physical index: ,in, For slot physical index, The integer spatiotemporal addressing token is used. Then, the slot physical index is multiplied by the array start address to obtain the slot offset. The physical memory address is then corrected based on the slot offset, and the unstructured state data frame is written to the corresponding slot.

[0054] The circular buffer array uses a timing slot width constant of 10000ms. Within the corresponding slot, the processing unit performs a linear scan according to the 10000ms time segment to which the time-series metadata timestamp belongs, and inserts data frames that arrive later but have smaller timestamps forward. Data frames with the same timestamp are arranged in the order they entered the high-speed buffer. After insertion, the data frames in each valid slot form a monotonically continuous linear sequence according to the non-decreasing order of the time-series metadata timestamps. When the absolute latency difference falls back to within 3000ms, or when the number of valid slots that have completed time alignment reaches 60, the background persistence thread reads the data frames according to the arrangement order in the circular buffer array and writes them continuously to the underlying time-series storage block through the high-speed storage bus. After each batch of data frames is written to disk, the storage middleware synchronously updates the status bit in the global status bit map corresponding to the integer time-space addressing token.

[0055] In this embodiment, 300,000 ms is used as the upper limit for latency processing of the high-speed buffer. During the test, when the absolute latency difference approaches this upper limit, the memory occupancy of the circular buffer array increases non-linearly. When the absolute latency difference continues to increase, unrefreshed slots will overlap with subsequent write positions, causing data overwriting. Therefore, when the absolute latency difference is between 3,000 ms and 300,000 ms, the high-speed buffer takes over and rearranges the data. After the refresh condition is met, the data is written to the underlying storage block in a monotonically continuous linear sequence. Through the above processing, The data access module, algebraic addressing module, and in-situ rearrangement disk writing module form a continuous data processing path. Integer spatiotemporal addressing tokens are used to determine the physical index of the slot in the circular buffer array, and the slot offset is used to correct the physical memory address. The circular buffer array is used to perform in-situ smooth rearrangement of out-of-order unstructured state data frames. After the rearrangement is completed, the structured text data and time-series parameter data are written to the distributed relational storage engine and the distributed time-series storage engine, respectively, thereby maintaining the continuous writing of perinatal follow-up health records under high concurrency and network latency fluctuation conditions.

[0056] Example 7: In this embodiment, a distributed industrial hybrid storage cluster continuously receives high-frequency time-series physiological parameters and discrete clinical texts generated by the acquisition unit. The computing node acquires multi-source asynchronous data streams sent by the front-end acquisition node through a preset data receiving port, parses the application-layer business fields, and converts the received data into unstructured state data frames. Subsequently, it extracts the original identity feature vector and time-series metadata timestamp from the unstructured state data frames, performs feature mapping on the original identity feature vector using a one-way hash algorithm, and generates a fixed-length anonymous feature code. The computing node establishes a data mapping channel in memory and calls an algebraic addressing operator to divide the time-series metadata timestamp by the time slicing span constant 86,400,000 ms and rounds down to obtain the discrete time slot sequence number. The fixed-length anonymous feature code and the discrete time slot sequence number are then concatenated to generate a first hash value. The first hash value is then XORed with the hardware-specific hash of the acquisition node to generate an integer spatiotemporal addressing token. The calculation relationship is as follows: ,in, It is an integer spacetime addressing token. For fixed-length anonymous signatures, For time-series metadata timestamps, The time partitioning span constant, The inherent hash of the data acquisition node's hardware. For string concatenation operations, This is a bitwise XOR operation.

[0057] The storage middleware determines the write position of unstructured state data frames in the high-speed buffer based on integer spatiotemporal address tokens. The compute node allocates physical memory space and initializes a circular buffer array of length 3600. The value of the integer spatiotemporal address token is used to perform an algebraic modulo operation on the capacity of the circular buffer array, and the remainder is used as the physical index of the slot. The slot offset is calculated by multiplying the physical index of the slot by the first address of the array, and then the physical memory address is corrected according to the slot offset. The unstructured state data frame is written to the corresponding image address. In the high-speed buffer, the compute node uses the time-series metadata timestamp as the sorting basis and performs linear scan and insertion sort on the unstructured state data frames written to the circular buffer array. Data frames that arrive later and have smaller time-series metadata timestamps are adjusted forward. Data frames with the same timestamp are arranged according to the order in which they entered the high-speed buffer. After in-situ smooth rearrangement, the data frames are stored continuously in non-decreasing order of time-series metadata timestamps, forming a monotonically continuous linear sequence.

[0058] The system monitoring module synchronously calculates the absolute latency difference between the system's physical master clock and the timestamp of the time-series metadata. When the absolute latency difference falls below 3000ms, the background thread reads the rearranged data frames according to the order in the circular buffer array and atomically writes the data sequence to the underlying time-series storage block, thus completing the ordered solidification of out-of-order data. In this process, integer spatiotemporal addressing tokens are used to determine the physical index of the slot, slot offsets are used to correct the physical memory address, and the circular buffer array is used to receive and rearrange the out-of-order unstructured state data frames. High-concurrency write operations are executed by the storage middleware according to the determined memory address and data arrangement order, reducing transaction lock waiting between foreground write threads and enabling continuous writing of perinatal follow-up health records under high concurrency and network latency fluctuations.

Claims

1. A method for automated filing of perinatal follow-up health records, characterized in that, Includes the following steps: Step S1: Receive multi-source asynchronous data streams, parse the multi-source asynchronous data streams to separate application layer business fields and convert them into unstructured state data frames, extract the original identity feature vectors and time-series metadata timestamps carried in the unstructured state data frames, perform feature mapping on the original identity feature vectors through a one-way hash algorithm, and generate fixed-length anonymous feature codes. Step S2: Establish a data mapping channel in memory and call the built-in algebraic addressing operator to divide the time-series metadata timestamp by the preset time-slice span constant and round down to separate the discrete time slot sequence number. Concatenate the fixed-length anonymous feature code with the discrete time slot sequence number to generate the first hash value. Perform a bitwise XOR operation between the first hash value and the hardware inherent hash of the acquisition node to generate an integer spatiotemporal addressing token. Step S3: Allocate physical memory space and construct a high-speed buffer through the storage middleware. Perform algebraic modulo operation on the capacity of the circular buffer array based on the value of the integer space-time addressing token to establish the slot physical index. Calculate the slot offset using the product of the slot physical index and the array's starting address. Correct the physical memory address based on the slot offset. Perform in-situ smooth rearrangement of the unstructured state data frame within the high-speed buffer using the circular buffer array. calibrate the rearranged unstructured state data frame into a monotonically continuous linear sequence before storing it on disk. After the physical memory address correction and in-situ smooth rearrangement are completed, the storage middleware separates the structured text data and time-series parameter data from the unstructured state data frame. The structured text data is written to the distributed relational storage engine using the integer space-time addressing token as the unique primary key. The time-series parameter data is appended to the physical time-series block of the distributed time-series storage engine using the integer space-time addressing token as the aggregation physical partition key.

2. The method for automated filing of perinatal follow-up health records according to claim 1, characterized in that, Step S3 is further refined into the following sub-steps: Step S31, calculate the image address of the unstructured state data frame in the high-speed buffer using the slot offset, and write the unstructured state data frame into the high-speed buffer; Step S32, in the high-speed buffer, use the circular buffer array to perform in-situ smooth rearrangement of the asynchronously out-of-order unstructured state data frames, and calibrate them into a monotonically continuous linear sequence.

3. The method for automated filing of perinatal follow-up health records according to claim 1, characterized in that, The method also includes step S4, which involves a cold / hot tiered storage migration process for data through a storage audit unit in a distributed industrial hybrid storage cluster. This process includes the following sub-steps: Step S41, the storage audit unit scans the update frequency vector of the global state bitmap and the block read / write count according to a preset period to obtain the read / write count of the data corresponding to the integer spatiotemporal addressing token within a continuous preset time period; Step S42, the storage audit unit compares the read / write count with a preset low-frequency threshold. When the read / write count is determined to be lower than the low-frequency threshold, the storage data state is switched. The historical fragmented data in the time-series storage area is aggregated through a merging algorithm, and the historical fragmented data is merged with the baseline snapshot data in the relational storage area to rewrite and generate columnar storage blocks. In step S43, the storage audit unit migrates the columnar storage block to the magnetic storage cold zone and establishes a reset pointer at the original index node to establish a single sequential read path for subsequent data queries.

4. The method for automated filing of perinatal follow-up health records according to claim 3, characterized in that, Step S42 is further refined into the following sub-steps: Step S421, the storage audit unit extracts the baseline snapshot data in the relational storage area as the reference axis, and arranges the historical fragment data in the time-series storage area according to the integer spatiotemporal address token for time alignment; Step S422, the storage audit unit merges the aligned historical fragment data into the columnar storage structure of the baseline snapshot data in an incremental append form, and rewrites to generate columnar storage blocks.

5. The method for automated filing of perinatal follow-up health records according to claim 1, characterized in that, The method also includes step S5, which involves writing historical archived data with delay compensation through a data storage middleware. This process includes the following sub-steps: Step S51, the data storage middleware receives out-of-order unstructured state data frames and calculates the historical address tokens corresponding to the out-of-order unstructured state data frames. Step S52: The data storage middleware opens a delay compensation storage slot in the high-speed buffer, introduces the main data block index head pointer and the incremental patch block mounted on the main data block index head pointer in the magnetic storage cold area, and writes the parameter information in the out-of-order unstructured state data frame into the incremental patch block in an incremental append form; Step S53: The data storage middleware updates the checksum in the global state bitmap through lock-free atomic bitmap operation and establishes the first-order data check path.

6. The method for automated filing of perinatal follow-up health records according to claim 5, characterized in that, Step S52 is further refined to include the following sub-steps: Step S521, the data storage middleware locates the main data block index head pointer in the magnetic storage cold zone and reads the mount path of the incremental patch block pointed to by the main data block index head pointer. In step S522, the data storage middleware writes the parameter information from the out-of-order unstructured state data frame into the incremental patch block and updates the file size attribute of the incremental patch block within the delay compensation storage slot.

7. The method for automated filing of perinatal follow-up health records according to claim 3, characterized in that, The continuous preset time period is 30 to 90 days, the low frequency threshold is no more than 5 times within the preset period, and the time segmentation span constant in step S2 is 86,400,000 ms.

8. The method for automated filing of perinatal follow-up health records according to claim 3, characterized in that, After establishing a reset pointer at the original index node, the method also includes the following query steps: when a global history query request for an integer spacetime addressing token is received, the query path is reset to the columnar storage block in the magnetic storage cold zone through the reset pointer, and a single continuous read addressing is performed on the columnar storage block in the magnetic storage cold zone.

9. An automated perinatal follow-up health record filing system, used to implement the automated perinatal follow-up health record filing method as described in claim 1, characterized in that, include: The data access module is used to receive multi-source asynchronous data streams, parse the multi-source asynchronous data streams to separate application layer business fields and convert them into unstructured state data frames, extract the original identity feature vector and time-series metadata timestamp carried in the unstructured state data frames, perform feature mapping on the original identity feature vector through a one-way hash algorithm and generate a fixed-length anonymous feature code. The algebraic addressing module, connected to the data access module, is used to establish a data image channel in memory and call the built-in algebraic addressing operator to divide the time-series metadata timestamp by the preset time segmentation span constant and round down to separate the discrete time slot sequence number. The fixed-length anonymous feature code is concatenated with the discrete time slot sequence number to generate the first hash value. The first hash value is XORed with the hardware inherent hash of the acquisition node to generate an integer spatiotemporal addressing token. The in-situ rearrangement and disk write-to-disk module, connected to the algebraic addressing module, is used to allocate physical memory space and build a high-speed buffer through the storage middleware. It performs an algebraic modulo operation on the capacity of the circular buffer array based on the value of the integer spatiotemporal addressing token to establish the slot physical index. It calculates the slot offset using the product of the slot physical index and the array's starting address, corrects the physical memory address based on the slot offset, and performs in-situ smooth rearrangement of the unstructured state data frame within the high-speed buffer using the circular buffer array. The rearranged unstructured state data frame is then calibrated into a monotonically continuous linear sequence before being written to disk. After the physical memory address correction and in-situ smooth rearrangement are completed, the storage middleware separates the structured text data and time-series parameter data from the unstructured state data frame. The structured text data is written to the distributed relational storage engine using the integer spatiotemporal addressing token as the unique primary key; the time-series parameter data is appended to the physical time-series block of the distributed time-series storage engine using the integer spatiotemporal addressing token as the aggregation physical partition key.

Images