Prediction method and system for fertility resource demand, and storage medium

Abstract

The invention relates to the technical field of demand prediction, in particular to a method and a system for predicting a growth resource demand, and a storage medium. The method comprises the following steps: obtaining real-time number of people waiting for delivery and real-time fertility resource data, carrying out space-time alignment, constructing a number-of-delivery-resource curve, collecting historical number of people waiting for delivery and fertility resource consumption data to identify a historical association relationship, analyzing the historical change influence degree based on the association relationship, and determining the number of people waiting for delivery. The method comprises the following steps: simulating a growth resource circulation condition in combination with a current resource curve, deducing a resource change trend according to real-time resources and a simulation result, reversely predicting the number change of people to be born through the resource change trend and a historical change influence degree, and calculating required matched resources in combination with historical associated data, thereby completing accurate prediction of growth resource demands. According to the method, structured and standardized resource prediction output is realized, and the adaptability of the system to high-frequency input, periodic fluctuation and structural variation is enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of demand forecasting, and in particular to a method, system and storage medium for forecasting reproductive resource demand. Background Art

[0002] Traditional fertility resource allocation relies heavily on fixed plans or periodic data scheduling, lacking a deep linkage mechanism between the real-time number of expectant mothers and dynamic resource changes. Traditional methods often rely on simple demographic data for rough resource allocation, making it difficult to meet differentiated regional needs. This is especially true during periods of instantaneous concentrated expectant mothers or sudden population fluctuations, where precise resource allocation cannot be achieved. This leads to resource shortages, waste, or uneven distribution in some regions. Medical resources are under severe pressure during peak periods, while existing facilities and manpower cannot be fully utilized during off-peak periods, resulting in low overall scheduling efficiency. In typical application scenarios, such as areas with accelerated urbanization and frequent demographic changes, existing forecasting models generally lack the ability to link and analyze historical and real-time data. Most systems lack a complete spatiotemporal data alignment mechanism, making it impossible to construct a dynamic mapping relationship between the actual number of expectant mothers and resource consumption. Furthermore, they lack a reverse inference model for resource change trends, resulting in delayed and distorted forecasting methods, making it difficult to promptly identify potential resource mismatches, and making it difficult to form an effective closed-loop mechanism for dynamic scheduling and pre-configuration. Summary of the Invention

[0003] Based on this, it is necessary to provide a method, system and storage medium for predicting fertility resource demand to solve at least one of the above technical problems.

[0004] To achieve the above purpose, the method for predicting fertility resource demand includes the following steps: Step S1: obtaining the real-time number of expectant mothers and real-time fertility resources; performing spatiotemporal joint alignment on the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; converting the spatiotemporal aligned data into a number of expectant mothers-resource curve; Step S2: Collect historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; identify historical associated data through the historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; and analyze the degree of impact of historical changes in the number of expectant mothers on historical changes in historical reproductive resource consumption data based on the historical associated data; Step S3: Simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; infer the change trend of fertility resources based on the current flow of fertility resources and real-time fertility resources; Step S4: Reversely predict the change data of the number of expectant mothers through the change trend of fertility resources and the impact of historical changes; calculate the corresponding supporting fertility resources required for the number of expectant mothers based on historical correlation data and the predicted change data of the number of expectant mothers, so as to obtain the predicted fertility resource demand.

[0005] The present invention constructs a structured data model under a unified reference by performing spatiotemporal joint alignment processing on the real-time number of expectant mothers and real-time fertility resources, which can enhance the spatial consistency and dynamic response capabilities between data. By generating a curve of the number of expectant mothers-resources, a continuous mapping relationship between resources and the number of people is realized, which is convenient for subsequent dynamic modeling of resource allocation. After fusing the historical data on changes in the number of expectant mothers and the historical data on fertility resource consumption, the stability of the cycle structure can be improved by identifying the historical associated data. By analyzing the impact of historical changes and constructing a cross-cycle resource response model, it is conducive to extracting the deep dependency between the number of people fluctuations and resource consumption. By combining the impact of historical changes with the current number of expectant mothers-resource curve to simulate the resource flow structure, a highly consistent flow state path can be formed, thereby more accurately By accurately inferring the changing trend of fertility resources, and further reversely reasoning the relationship between the changing trend of resources and historical impacts, the changing data of the number of expectant mothers can be predicted and the potential incremental logic of the population structure can be restored. By combining the predicted changing data of the number of expectant mothers with historical related data, the demand for supporting resources can be calculated, and structured and standardized resource forecast output can be achieved. The overall process enhances the system's adaptability to high-frequency inputs, periodic fluctuations and structural changes through continuous nesting of data structures and closed-loop design of result logic, improves the accuracy and timeliness of the system's forecast results and the flexibility of resource scheduling, has good portability, controllability and structural fusion performance, is suitable for the deployment of multi-scenario and multi-period resource forecasting systems, and builds a forecasting and control model support system with high adaptability, high versatility and high structural stability.

[0006] The present invention further provides a prediction system for fertility resource demand, which is used to execute the above-mentioned prediction method for fertility resource demand. The prediction system for fertility resource demand includes: The spatiotemporal alignment module is used to obtain the real-time number of expectant mothers and real-time fertility resources; perform spatiotemporal joint alignment of the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; and convert the spatiotemporal aligned data into a number of expectant mothers-resource curve. The historical correlation module is used to collect historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; identify historical correlation data through historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; and analyze the impact of historical data on changes in the number of expectant mothers on historical data on reproductive resource consumption based on historical correlation data; The resource deduction module is used to simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; and to infer the trend of fertility resource changes based on the current flow of fertility resources and real-time fertility resources; The reverse prediction module is used to reversely predict the change data of the number of expectant mothers based on the change trend of fertility resources and the impact of historical changes; based on historical correlation data and the predicted change data of the number of expectant mothers, the corresponding supporting fertility resources required for the number of expectant mothers are calculated, thereby obtaining the predicted demand for fertility resources.

[0007] The present invention realizes the joint processing of the real-time number of expectant mothers and fertility resources by constructing a spatiotemporal alignment module, which can form a dynamic matching structure in a multi-dimensional coordinate system, enhance the integration and synchronization of data on the time axis and the space axis, and construct a continuous mapping relationship for subsequent modeling analysis by converting the alignment result into a number of expectant mothers-resource curve. The historical number of expectant mothers change data and resource consumption data are introduced into the historical association module for joint identification, which can extract historical response trajectories with cyclical stability and form a resource evolution chain with upstream and downstream dependency logic. The long-term impact weight matrix of the number of expectant mothers on resource allocation can be obtained by constructing the impact degree of historical changes, so that data in different historical stages have quantifiable and interoperable trend projection capabilities. The historical impact and current structure are integrated in the resource deduction module to simulate the evolution direction of resources in the current state, so that the current resource distribution trend can be structured. The system can express the expected number of births and make inferences based on the expected number of births, and pair it with real-time resource input to obtain the path of future resource changes. In the reverse prediction module, the changing process of the number of expectant mothers is reversed based on the resource trend, forming a reverse mapping network from structural causality to structural demand. The system can also combine and deduce the historical correlation data with the predicted number of births to ensure the predicted amount of supporting fertility resources has periodic continuity and historical consistency. The overall system realizes modular division of labor from data collection, relationship identification, structural simulation to demand forecasting, with the characteristics of structural closed-loop, prediction robustness and output standardization. It is suitable for forecasting and allocation scenarios with complex periodic structures, frequent data fluctuations and strong resource dependence. It has the advantages of strong scalability, high computing efficiency and strong system integration. It still has high precision, high dynamic response capability and high time series adaptability in the context of multi-source heterogeneous data, and can effectively support the intelligent evolution of resource allocation systems and the deep integration of data-driven predictive modeling.

[0008] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed, implements the method for predicting reproductive resource demand as described in any one of the above.

[0009] The present invention embeds a complete method for predicting fertility resource demand into a computer-readable storage medium in the form of a computer program, thereby realizing modular encapsulation of various sub-processes such as data collection, spatiotemporal combination, historical identification, trend deduction and reverse prediction, and constructing a program architecture with deployability and reusability. It still has complete functional adaptability in scenarios with limited computing resources or complex deployment environments. By embedding the collaborative calling logic of real-time data and historical data into the program instruction structure, it can ensure that the program dynamically loads data modules and computing modules during execution, realizes automatic identification of data structures, automatic alignment of time series structures and automatic mapping of spatial structures, and improves the overall data processing efficiency and the degree of information fusion. By predefining the multi-dimensional coordinate projection logic of fertility resources and the number of expectant mothers in the program, the system can automatically generate a spatiotemporal data association map during operation, thereby enhancing the parseability and Updatability: The program supports joint modeling operations of historical evolution paths and current data features, and can dynamically switch between resource response models and population change models in the prediction engine to improve prediction stability and result adaptability. Based on the curve fitting strategy and multi-dimensional mapping mechanism solidified in the program, nonlinear distribution relationships can be converted into continuous fitting expressions, improving the computability and subsequent reasoning capabilities of the output structure. The program module has built-in resource trend backtracking mechanism and population reverse evolution logic. By calling a unified data access interface and time period sliding window structure, it realizes the stable flow of timing logic, ensuring that the prediction output has high timeliness and stable continuity. The instruction set in the computer-readable storage medium has cross-platform deployment capabilities and can be standardized in different hardware architectures, enhancing the system compatibility and resource management capabilities of the technical solution, and is suitable for integrated deployment and stable execution in multi-terminal, multi-node, and multi-stage fertility resource collaborative systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic flow chart of the steps of a method for predicting reproductive resource requirements; Figure 2 Detailed implementation flow chart of step S1; The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0011] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0012] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0013] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0014] To achieve this, please refer to Figures 1 to 2 , a method for predicting fertility resource demand, comprising the following steps: Step S1: obtaining the real-time number of expectant mothers and real-time fertility resources; performing spatiotemporal joint alignment on the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; converting the spatiotemporal aligned data into a number of expectant mothers-resource curve; Step S2: Collect historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; identify historical associated data through the historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; and analyze the degree of impact of historical changes in the number of expectant mothers on historical changes in historical reproductive resource consumption data based on the historical associated data; Step S3: Simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; infer the change trend of fertility resources based on the current flow of fertility resources and real-time fertility resources; Step S4: Reversely predict the change data of the number of expectant mothers through the change trend of fertility resources and the impact of historical changes; calculate the corresponding supporting fertility resources required for the number of expectant mothers based on historical correlation data and the predicted change data of the number of expectant mothers, so as to obtain the predicted fertility resource demand.

[0015] The present invention constructs a structured data model under a unified reference by performing spatiotemporal joint alignment processing on the real-time number of expectant mothers and real-time fertility resources, which can enhance the spatial consistency and dynamic response capabilities between data. By generating a curve of the number of expectant mothers-resources, a continuous mapping relationship between resources and the number of people is realized, which is convenient for subsequent dynamic modeling of resource allocation. After fusing the historical data on changes in the number of expectant mothers and the historical data on fertility resource consumption, the stability of the cycle structure can be improved by identifying the historical associated data. By analyzing the impact of historical changes and constructing a cross-cycle resource response model, it is conducive to extracting the deep dependency between the number of people fluctuations and resource consumption. By combining the impact of historical changes with the current number of expectant mothers-resource curve to simulate the resource flow structure, a highly consistent flow state path can be formed, thereby more accurately By accurately inferring the changing trend of fertility resources, and further reversely reasoning the relationship between the changing trend of resources and historical impacts, the changing data of the number of expectant mothers can be predicted and the potential incremental logic of the population structure can be restored. By combining the predicted changing data of the number of expectant mothers with historical related data, the demand for supporting resources can be calculated, and structured and standardized resource forecast output can be achieved. The overall process enhances the system's adaptability to high-frequency inputs, periodic fluctuations and structural changes through continuous nesting of data structures and closed-loop design of result logic, improves the accuracy and timeliness of the system's forecast results and the flexibility of resource scheduling, has good portability, controllability and structural fusion performance, is suitable for the deployment of multi-scenario and multi-period resource forecasting systems, and builds a forecasting and control model support system with high adaptability, high versatility and high structural stability.

[0016] In an embodiment of the present invention, the method for predicting reproductive resource demand includes the following steps: Step S1: obtaining the real-time number of expectant mothers and real-time fertility resources; performing spatiotemporal joint alignment on the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; converting the spatiotemporal aligned data into a number of expectant mothers-resource curve; In this embodiment, the obstetric hospitalization registration module interface in the regional medical information system is called to extract data on the number of pregnant women who have registered as awaiting delivery within the past hour. The data fields include the pregnant woman's ID, registration time, department, and geographic coordinates. Real-time reproductive resource data, including obstetric bed utilization, midwifery and nursing staff schedules, maternal and child supply inventory, and equipment operating status, is accessed through the medical resource management platform. All data is structured based on timestamps and institution identifiers. A joint alignment algorithm based on time synchronization is used to align the real-time number of awaiting delivery data with the reproductive resource data according to time windows and spatial regions. This generates a spatiotemporally aligned data matrix containing the number of awaiting delivery and their associated resource quantities at each time point. A polynomial fitting method is then applied to the spatiotemporally aligned data to generate a number-of-arriving-delivery-resources curve. The curve has the number of awaiting delivery as the horizontal axis and the total number of reproductive resources and various resource classification indicators as the vertical axis. Spline interpolation is used to improve the smoothness and continuity of the curve. The entire process uses the distributed data processing platform Spark for batch and streaming data fusion processing to achieve dynamic conversion of the spatiotemporally aligned data into the curve.

[0017] Step S2: Collect historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; identify historical associated data through the historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; and analyze the degree of impact of historical changes in the number of expectant mothers on historical changes in historical reproductive resource consumption data based on the historical associated data; In this embodiment, obstetric outpatient and inpatient delivery records from the hospital's historical database over the past five years were extracted, and the monthly change in the number of expectant mothers and children was structuredly extracted. Simultaneously, data on maternal and infant supplies consumed during the same period, human resource scheduling and usage time data, and delivery room equipment operation and maintenance records were exported from the material management system. The data was cleaned and normalized using an ETL (Extract-Transform-Load) process. The correlation between the historical data on changes in the number of expectant mothers and childbirth resource consumption was identified using statistical analysis methods based on the Pearson correlation coefficient and Granger causality test. A historical correlation dataset was constructed. This dataset performed synchronous sliding window analysis on each time series of changes in the number of expectant mothers and childbirth resource consumption, calculating correlation and lag effects. Subsequently, a multivariate linear regression model was used to estimate the impact of historical changes in the number of expectant mothers on changes in childbirth resource consumption. Ridge regression was used in model training to prevent multicollinearity, and model parameters were optimized using cross-validation. All processing steps were implemented using the Pandas and Statsmodels libraries in Python to obtain the impact of historical changes.

[0018] Step S3: Simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; infer the change trend of fertility resources based on the current flow of fertility resources and real-time fertility resources; In this example, a resource flow simulation model is constructed by inputting a real-time, generated curve of the number of people waiting to be born versus resources and a weight matrix of the impact of historical changes. This model uses a discrete event simulation framework with a simulation cycle set to one hour. The model simulates the allocation of resource demand corresponding to each time node in the curve of the number of people waiting to be born. The model includes parameters for resource scheduling delay and resource replenishment cycle. Real-time fertility resource data is introduced as state input to adjust resource occupancy and release rates. During the simulation, queue management and priority scheduling are performed for each resource type. The Monte Carlo method is used to simulate resource usage fluctuations, and indicators such as current resource occupancy, queue length, and resource turnover time are output. The simulation results are time-series smoothed using a sliding window. Based on the simulation results and real-time resource data, the Kalman filter algorithm is applied to infer fertility resource change trends. All models are run using the AnyLogic simulation platform combined with a Python API for parameter control and result extraction.

[0019] Step S4: Reversely predict the change data of the number of expectant mothers through the change trend of fertility resources and the impact of historical changes; calculate the corresponding supporting fertility resources required for the number of expectant mothers based on historical correlation data and the predicted change data of the number of expectant mothers, so as to obtain the predicted fertility resource demand.

[0020] In this embodiment, the fertility resource change trend vector is input into a multidimensional time series stitching module and cross-mapped with the weight matrix of the historical change impact. The module uses tensor decomposition technology to stitch the data to obtain a stitched evolution benchmark tensor. This tensor is then used to perform a time series reverse backtracking of the resource change trend. The backtracking uses an integral accumulation algorithm to decompose the resource change trend signal segment by segment along the time axis, extracting the starting point of the resource change. The starting point is matched with the resource change amount in the historical change impact using a dynamic time warping (DTW) algorithm. The matched resource change amount is assigned a corresponding sensitivity coefficient, and a weighted backpropagation algorithm is used to map the resource change amount to the corresponding change interval of the number of expectant mothers. Finally, all change intervals are accumulated and superimposed to form a complete data sequence of predicted changes in the number of expectant mothers. The backtracking and mapping process is implemented using a trained residual network model in the TensorFlow deep learning framework. The model weights remain unchanged based on the historical training data. The prediction results are normalized and output to provide basic data for subsequent resource demand calculations.

[0021] Preferably, step S1 includes the following steps: Step S11: obtaining the real-time number of expectant mothers and real-time fertility resources; performing distribution mapping processing on the real-time number of expectant mothers to obtain spatiotemporal distribution data of the number of expectant mothers; Step S12: normalize the labels of real-time fertility resources, where the normalized label dimensions do not exceed 12 resource types, thereby generating spatiotemporal distribution data of fertility resources; perform multidimensional coordinate projection on the spatiotemporal distribution data of the number of expectant mothers and the spatiotemporal distribution data of fertility resources to obtain spatiotemporal matching data; Step S13: performing spatial alignment on the spatiotemporal matching data, wherein the alignment error threshold is limited to a spatial offset error not exceeding 0.25 grid units, thereby obtaining spatiotemporal aligned data; Step S14: performing surface fitting processing on the spatiotemporal alignment data to obtain spatiotemporal surface structure data; based on the spatiotemporal surface structure data, depicting the spatiotemporal alignment data as a number of expectant mothers-resource curve.

[0022] In this embodiment, the real-time number of expectant mothers is extracted by calling the database interface of the regional hospital management system. The data obtained includes the unique identity of the pregnant woman, the registration timestamp, the coordinates of the medical institution where she is located, and the delivery status. The number of expectant mothers is gridded in the spatial dimension using geographic information system (GIS) technology and divided into grid units with a grid side length of 100 meters. The density of the number of expectant mothers in each grid is calculated and combined with the timestamp to generate a spatiotemporal distribution matrix. The matrix elements represent the number of expectant mothers at a specific time point and spatial unit. The spatiotemporal data is normalized and output as the spatiotemporal distribution data of the number of expectant mothers. The real-time fertility resource data comes from the hospital material management platform, and the resource types are limited to 12 categories or more. The data includes the number of obstetric beds, the number of midwives, the number of delivery room equipment and the inventory of medical consumables. A unified normalized label is defined for each resource. The normalization process uses the maximum-minimum normalization method to map the actual value of the resource to the range of 0 to 1. The normalized resource data is associated with the geographical location to generate the corresponding spatiotemporal distribution data of fertility resources. The normalized label dimension is limited to no more than 12 resource types. The data storage adopts a multidimensional tensor structure. The spatiotemporal distribution data of the number of people waiting to give birth and the spatiotemporal distribution data of fertility resources are input into the multidimensional coordinate projection module. The module is based on the high-dimensional space dimensionality reduction algorithm to map the two spatiotemporal tensors to a unified multidimensional coordinate system and use principal component analysis (PCA, Principal Component Analysis (CSA) was used to extract features and reduce the dimensionality of the original multidimensional data. After dimensionality reduction, the data dimensions were kept within six dimensions to ensure computational efficiency. The projection coordinate axes covered spatial location, time node, density of the number of expectant mothers, and normalized resource values. The projection results were output as a matrix representing spatiotemporal matching data. Each row of the matrix corresponds to a specific spatiotemporal point and its matching between the number of expectant mothers and reproductive resources. All operations were implemented in the Python environment using the NumPy and Scikit-learn machine learning libraries. Spatial alignment was performed on the spatiotemporal matching data. The iterative closest point (ICP) algorithm was used to align the spatiotemporal distribution of the number of expectant mothers with that of reproductive resources. The spatial offset error threshold was set to 0.25 grid units, the grid unit corresponds to an actual space length of 100 meters, and the upper limit of the error is 25 meters. During the registration process, the Euclidean distance between the corresponding points of each grid unit is calculated, and the rigid transformation matrix is ​​optimized using the least squares method to minimize the distance between the two sets of point clouds. The iterative process includes the nearest point matching, transformation matrix calculation and error evaluation. The iteration is terminated if the error does not exceed the threshold. The data after completion of the alignment forms a spatiotemporal alignment data structure, which includes each spatial grid and its corresponding number of expectant mothers and normalized values ​​of fertility resources. All registration operations are called in the C++ environment PCL (Point Cloud The spatiotemporal data were implemented using the Point Cloud Library (Point Cloud Library). Surface fitting was applied to the spatiotemporal aligned data using a cubic spline surface fitting algorithm. Parameterization was performed on the X and Y dimensions of the spatial coordinate system and the time dimension T. The spline surface fitted the functional relationship between the number of expectant mothers and normalized fertility resources at each time slice. Locally weighted scatterplot smoothing (Loess) was used to smooth the spatiotemporal surface. The surface grid resolution was set to 50 m × 50 m spatial units, and the time dimension slice interval was 15 minutes. The fitting results formed continuous and smooth three-dimensional spatiotemporal surface structure data. This structure data was visualized as a dynamic surface, mapping the strength of the match between the number of expectant mothers and resources. Based on the spatiotemporal surface structure data, contour extraction was used to convert key data points on the surface into a two-dimensional expectant mother-resource curve. The curve is plotted using the Matplotlib library. The curve reflects the dynamic relationship between resources and the number of expectant mothers. The entire fitting and curve plotting process was completed in the Python environment.

[0023] Preferably, in step S2, identifying historical associated data through historical data on changes in the number of expectant mothers and historical data on consumption of reproductive resources includes: The historical data on changes in the number of expectant mothers are processed into time series layers, where the length of the layered time window is limited to no more than 72 hours, thereby obtaining a layered series of the number of expectant mothers; Identify resource consumption patterns in historical reproductive resource consumption data, where the minimum continuous usage period for a single resource type must not be less than 4 hours; Perform sequence collaborative retrieval on the stratified sequence of the number of people waiting to be born and the resource consumption pattern to obtain historical correlation data.

[0024] In this embodiment, a complete time series is established based on the original timestamp data of the change in the number of expectant mothers. The time granularity of the time series is set to 15 minutes per gear. Based on the continuous time window length of 72 hours, the entire historical time series is divided into multiple subsequences of fixed length. Time overlap is allowed between subsequences, and the overlap length is 12 hours. The time series is cut using a sliding window algorithm. After the cutting is completed, each subsequence is smoothed. The smoothing adopts the moving average method. The moving window size is 3 time gears. The smoothing result reduces the sudden data fluctuation. Then, trend analysis is performed on each layered subsequence. The trend analysis adopts the linear regression fitting method. The slope parameter of the fitting is used to judge the trend of the number of people. The change characteristics in the layered sequence, such as peaks and troughs, are extracted by the local extreme point detection algorithm. The extreme point detection is implemented based on the first-order difference method. All time series processing algorithms are implemented based on Python's Pandas and NumPy libraries. Timestamp conversion uses ISO 8601 format. After processing, a hierarchical time series matrix is ​​output. Each row corresponds to a change sequence of the number of expectant mothers within a 72-hour time window. Historical resource consumption records are extracted from the hospital material management system database. The record content includes resource type, consumption time starting point, consumption time ending point and consumption amount. A continuous consumption time series is constructed for each resource type. The time unit is unified as 1 hour. The threshold method is used to determine the continuous usage cycle in the resource consumption time series. The specific method is that when the resource consumption amount is continuously greater than zero, the continuous consumption duration is calculated. The consumption segments with a continuous duration of less than 4 hours are eliminated, and the remaining continuous consumption segments are marked as effective consumption cycles. The consumption rate is calculated for each effective consumption cycle. The consumption rate is defined as the total resource consumption during a period divided by its duration. A sliding window with a length of 24 hours is used to calculate the frequency distribution of consumption periods. The statistical data is stored as a three-dimensional array consisting of time period, resource type, and consumption rate. The Python SciPy library is used to detect continuous segments and perform frequency statistics. All times are in UTC. The extracted consumption patterns are used as resource consumption features. The stratified series of the number of expectant mothers and the resource consumption feature data are timestamped uniformly using the NTP network time protocol. The timestamps of the corrected timestamps ensure that the start and end points of the two series correspond. The dynamic time warping (DTW) algorithm is used to nonlinearly align the time axes of the two series. During the DTW distance matrix calculation, the distance function is defined as the absolute value of the difference between the corresponding values ​​at two time points. The DTW distance is calculated to assess the similarity between the series of expectant mothers and the series of resource consumption. Based on the DTW calculation results, the matching sequence segments with the smallest distance are selected as candidate association data. The cross-correlation function (CCF) is then applied to the candidate sequence segments to calculate the correlation coefficient, with a threshold of no less than 0.7. Sequence segments exceeding the threshold are further included in the historical association data set. All retrieval and matching processes are implemented using Python's dtaidistance and statsmodels libraries. The resulting historical association data structure includes the start and end time, the change pattern of the number of expectant mothers, the corresponding resource consumption pattern, and the correlation coefficient index. The association data is stored in a time series data table based on a relational database.

[0025] Preferably, in step S2, analyzing the impact of historical changes in the number of expectant mothers on historical fertility resource consumption data based on historical correlation data includes: Based on the historical correlation data, query the corresponding correlation features in the historical data on changes in the number of people expecting to give birth and the historical data on consumption of reproductive resources; Locate the relevant change segments in the historical data on changes in the number of expectant mothers and the historical data on consumption of reproductive resources based on the corresponding correlation features; Analyze the historical change data of the number of expectant mothers and the relative change data of historical reproductive resource consumption data based on the corresponding correlation features and correlation change segments; The relative change data is used to determine the impact of historical changes in the number of expectant mothers on historical fertility resource consumption data, and quantify the degree of historical change impact.

[0026] In this embodiment, historical associated data is stored in a relational time series database. The database table structure design includes a timestamp field, a field for the change in the number of people waiting to be delivered, a field for resource consumption, and a field for associated feature labels. The associated feature labels are generated based on the previous dynamic time warping (DTW) and cross-correlation function (CCF) calculation results. The query is implemented by constructing a structured query language (SQL) statement, limiting the time range to the past year. The feature labels include categories such as peaks, troughs, growth trends, and downward trends. The WHERE condition is used in the SQL statement to filter out records that meet the feature labels, and the feature time points and corresponding values ​​of the two types of sequence data are output. The query efficiency is optimized by the index field, which includes the timestamp and feature label. The entire query operation is performed in the database management. The data processing system uses multi-threaded parallel processing technology to ensure millisecond response times for massive data queries. Query results are organized into a two-dimensional table structure, corresponding to feature sequences of changes in the number of expectant mothers and feature sequences of reproductive resource consumption. The data format is unified as a list of timestamp-feature value pairs. Correlated change segments are located for the queried historical features of changes in the number of expectant mothers and reproductive resource consumption. This location is achieved using a sliding window method with a window size of 24 hours and a window sliding step of 1 hour. The two feature sequences within each window are synchronously compared, and the similarity between the two sequences within that time window is calculated using the cosine similarity algorithm. The formula for similarity is equal to the dot product of the two sequence vectors within the window divided by the modular product of the two sequence vectors. The threshold is set to 0.8. Windows exceeding the threshold are identified as associated change segments. The positioning process is implemented using Python's NumPy library for efficient matrix operations. After positioning, the start and end time tags of the associated change segments are saved in the database table. At the same time, the change series of the number of expectant mothers and the resource consumption series data in the corresponding time period are extracted. The series data are arranged in timestamp order. The baseline is adjusted for the two types of time series data. The baseline is defined as the data value at the starting time point of the corresponding segment. The baseline value is subtracted from the data at all subsequent time points to obtain the relative change amplitude. The baseline-adjusted data form relative change sequences, and the difference sequence of the relative change sequence is calculated. The change rate of each time point relative to the previous time point is obtained by the first-order difference formula. The change rate data is used to capture dynamic change trends. The change rate calculation uses the sliding window averaging method, and the window length is set to 3 hours. The smoothing result reduces short-term noise interference. The Python Pandas library is used to perform difference and smoothing operations on the time series. After the calculation is completed, the relative change series of the number of expectant mothers and the smoothed relative change series of resource consumption are generated. The relative change series are statistically analyzed to determine the impact of historical changes in the number of expectant mothers on changes in reproductive resource consumption. The Pearson correlation coefficient (Pearson correlation coefficient) is used to calculate the relative change series. The Correlation Coefficient (CC) measures the degree of linear correlation between two series. The calculation formula is the covariance of the two series divided by the product of their respective standard deviations. The correlation coefficient ranges from -1 to 1, with a threshold of ±0.5 or greater considered significant. The calculation process is performed separately for all correlation change segments, generating a corresponding correlation coefficient value for each segment. The correlation coefficient serves as a quantitative indicator of the impact of historical changes. The correlation coefficient calculation is implemented using the Python SciPy library. All calculation results are recorded in the corresponding correlation change segment table in the database, including the time interval, correlation coefficient, and statistical significance p-value. The p-value is obtained through a two-tailed t-test with a significance level set at 0.05. The correlation coefficient and p-value jointly determine the final determination of the impact of historical changes.

[0027] Preferably, in step S3, simulating the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve includes: Confirm current resource occupancy data based on the number of expectant mothers-resource curve; Calculate resource reflow cycle based on current resource occupancy data; Infer available fertility resources through real-time fertility resources and resource return cycles; Conduct time series analysis on current resource occupancy data and available reproductive resources to generate resource return dynamic data; Correcting the resource return dynamic data based on the impact of historical changes to obtain corrected resource return dynamic data; The current flow of reproductive resources is simulated by correcting the dynamic data of resource return.

[0028] In this embodiment, the real-time number of people waiting to give birth and the corresponding fertility resource data are accurately matched through timestamps. The data source is the real-time interface between the hospital obstetrics management system and the fertility resource allocation system. The time series database query engine is used to read the curve data according to the time point. The time window of the current moment is set to the sampling data within the last hour. The interpolation algorithm is used to fill the sampling points to ensure data continuity. The current resource occupancy is calculated by accumulating the resource demand corresponding to the real-time number of people waiting to give birth. The resource demand value is determined by the standardized index. In the standardization process, the minimum-maximum normalization method is used to unify the resource consumption standards at different time points to obtain the current resource occupancy. The data format is a time series array, which includes the time point and the corresponding resource occupancy value. The operation is performed through Python The Pandas library implements data reading, interpolation, and calculation to ensure that the data structure meets the needs of subsequent resource flow calculations. The resource return cycle is calculated based on the current resource occupancy data. The resource return cycle is defined as the time required for resources to be occupied, released, and returned to an available state. It is obtained by statistically calculating the difference between the resource release time and the occupancy time in the historical reproductive resource usage log. The moving average method is used to smooth the historical cycle data. The time window is the past 30 days, and the release cycle mean is calculated every day. The noise is processed by excluding outliers. The outlier identification uses the box plot method to identify extreme values ​​outside the upper and lower quartiles. The calculation formula for the return cycle is the mean of the time difference of each release cycle, and the unit is set to hours. Finally, the resource return cycle parameters are input into the resource scheduling model. The scheduling model is based on the discrete event simulation (DIS) This is achieved using the DES (Dysfunctional Experiential Learning) method to ensure that the calculation results conform to actual resource usage patterns. The available reproductive resources are inferred by combining real-time reproductive resource data with the resource return cycle. Real-time reproductive resource data comes from the reproductive resource inventory management system, including the number of beds, medical staff, and equipment. This data is updated every five minutes. The inference process is based on a dynamic inventory model. The dynamic inventory changes are described by the following expression: available resources equal the current total resources minus resource usage plus estimated return resources. The estimated return resources are calculated based on the resource return cycle parameters. Exponential smoothing is used to predict future return resources. The forecast time range matches the resource return cycle, and the smoothing coefficient is set to 0.3. The exponential smoothing prediction method uses weighted historical return flow data to calculate future return flow trends. After the calculation is completed, the time series data of available fertility resources is output. The data includes time points and corresponding available resource amounts. The data format is compatible with the front-end resource scheduling system to ensure the accuracy of the scheduling model input. Time series analysis is performed on the current resource occupancy data and available fertility resources to generate resource return dynamic data. The time series analysis uses the autoregressive moving average model (ARMA, Autoregressive Moving Average) to jointly model the two sets of time series data. The model parameters p and q are set to 2 and 1 respectively. The parameters are estimated by the maximum likelihood estimation method. The input sequence includes the resource occupancy and available resource amounts in the last 72 hours. The data preprocessing steps include normalization and detrending operations. Normalization uses the mean-variance standardization method to convert the data into a standard normal distribution with a mean of zero and a variance of one. Detrending uses the difference method to eliminate the non-stationarity of the time series. Model training uses Python The Statsmodels library outputs resource return dynamic data as a series of predicted values ​​for resource occupancy and release over the next 24 hours. This data reflects the dynamic return process of resources on the timeline. The prediction accuracy is evaluated using the root mean square error (RMSE), which is kept within 5% to meet the needs of subsequent analysis. The resource return dynamic data is corrected for the return impact based on the impact of historical changes. The historical impact is introduced as a weight factor into the correction model. The correction model uses the weighted moving average method, with the weight determined by the correlation coefficient of the historical impact. The correlation coefficient is linearly mapped between 0 and 1 to the weight. The correction calculation process multiplies the resource return dynamic value at each time point by the weight factor for the corresponding time period to obtain the corrected resource return dynamic data. The data format remains consistent, and the corrected data is aligned with the original dynamic data in the time dimension. The correction algorithm is implemented through matrix multiplication, using the NumPy library for efficient calculation, ensuring that the correction operation is completed within seconds. The correction process limits abnormal weight values ​​within a range of 0.1 to 0.9. To prevent data from deviating from reality due to overcorrection, the current fertility resource flow situation is simulated using discrete event simulation methods by correcting resource return dynamic data. The simulation model design includes three core events: resource application, resource occupation, and resource release. The event occurrence times strictly correspond to the time points of the corrected resource return dynamic data. The model performs time-series event scheduling within the simulation environment. The scheduling algorithm is implemented based on priority queues, and event priorities are dynamically adjusted based on resource scarcity. Resource flow status is expressed as a state machine, including idle, occupied, and released states. The simulation process is built using the Python SimPy library. The input data is the corrected resource return dynamic time series, and the output is a real-time state data stream of resource flow. The state data includes the occupation time, release time, and standby time of each resource unit. The simulation results are used in the subsequent resource demand forecasting module. The simulation operation parameter configuration includes the simulation duration set to the next 48 hours and the time step of 5 minutes.

[0029] It is particularly important to calculate the resource reflow cycle based on current resource usage data, including: Perform resource type layering on the current resource occupancy data to obtain a resource type occupancy structure; Extract the time window of resource type occupancy structure to generate resource release time period; Perform frequency aggregation on the resource release time period to obtain a resource release frequency curve; Combine the resource release frequency curve and the impact of historical changes to perform cycle fitting and generate a preliminary return cycle; The preliminary reflow cycle data is processed for multi-source consistency correction to obtain the resource reflow cycle.

[0030] In this embodiment, resource occupancy data is classified according to resource types, including but not limited to beds, medical staff, medical equipment and other categories. The data structure is stored in the form of a multidimensional array, and each dimension corresponds to a resource type. The classification is based on the resource attribute field in the resource management system. Data preprocessing tools such as Python's Pandas library are used to read the original resource occupancy time series data. Data stratification is achieved by screening the resource type field. An independent time series data set is established for each resource type. The time series is sorted by timestamp, and missing data is filled by linear interpolation. The hierarchical data structure facilitates subsequent fine-grained analysis. The sampling frequency is set to every 5 minutes for resources, and every 15 minutes for medical staff resources. The classified resource type occupancy structure is stored in a relational database. The index is used to optimize the retrieval speed to ensure that the occupancy of specific resources can be quickly located in a large-scale data environment. The overall processing flow is automatically executed by writing Python scripts, defining the time window length as 1 hour and the window sliding step as 10 minutes. By traversing the time series data of the resource type occupancy structure, the resource occupancy status changes are counted for each time window to determine the time point when the resource release event occurs. The judgment condition for the release event is that the resource occupancy status changes from "occupied" to "idle". , the state transition matrix is ​​used to encode the time series state of each resource type. The state code includes 0 for idle and 1 for occupied. The state difference between adjacent time points is calculated to identify the release boundary, the timestamps of all release events are extracted, and the continuous release time periods are summarized. The release time period format is defined as a start time and end time pair. All release time periods are stored as a time interval list. The time window extraction and release time period generation process adopts the sliding window algorithm and is combined with the state machine model to achieve high-speed traversal of the time series using NumPy arrays. After the release time period is generated, the time interval is merged. The merging rule is that if the interval between adjacent release time periods is less than 5 minutes, it is considered to be the same release time period. The merged dataset is written into the time series database to support fast query. The resource release time period is frequency aggregated to obtain the resource release frequency curve. The aggregation operation counts the release event frequency in hours. The specific method is to segment and count all resource release time periods according to the time axis, and count the number of release events every hour. The release frequency curve is defined as the ratio of the release event count to the total number of resources in the time period, and the value range is between 0 and 1. The release frequency curve is smoothed by the sliding average method, and the smoothing window length is set to 3 hours. The weighted moving average algorithm is used in the smoothing process, and the weight decreases linearly according to the distance from the current time point. The signal processing module in the Python SciPy library is used to implement the smoothing calculation. The output result is the resource release frequency value and the corresponding timestamp per hour. The release frequency curve data format is a two-dimensional array, the first column is the timestamp, and the second column is the frequency value.Frequency curve visualization uses the Matplotlib drawing library to reflect the dynamic trend of resource release by drawing a line graph. The frequency data is structured and stored in the database. The resource release frequency curve and the impact of historical changes are combined to perform period fitting to generate a preliminary reflux cycle. The period fitting process uses Fourier transform technology to analyze the periodic characteristics of the release frequency curve. First, a discrete Fourier transform is performed on the release frequency data to extract the main frequency component. The period corresponding to the main frequency is the main period of resource release. The impact of historical changes is introduced as a regulating factor in the fitting process. The impact degree is assigned different weights through the previously calculated correlation coefficient. The frequency components in the frequency spectrum are weighted and adjusted. The weight is linearly mapped from the correlation coefficient to the frequency amplitude. The weight range is limited to between 0.1 and 1 to ensure that no noise is introduced. The fitting process uses Python's NumPy The FFT module is implemented to extract the period corresponding to the maximum weight frequency as the preliminary reflux period. The period unit is in hours. The fitting result is restored to the time domain through inverse transformation to verify the fitting accuracy. The fitting residual is controlled within the threshold through mean square error (MSE) calculation. The preliminary reflux period data is subjected to multi-source consistency correction processing to obtain the resource reflux period. The correction operation integrates information from multiple data sources, including current resource occupancy data, historical resource usage records, and related environmental parameters. First, the period parameters of each data source are standardized. Standardization is achieved by subtracting the mean and then dividing by the standard deviation. The standardized period parameters are synthesized through a weighted fusion algorithm. The weights are based on the reliability of the data source and the data Integrity is dynamically adjusted, and the weight distribution rules are based on the recency of data collection time and data quality indicators, with the sum of weights being 1. During the correction process, the Bayesian fusion method is used to calculate the joint probability of the period distributions of different data sources. The fusion result is the posterior period probability distribution. The final period value is determined by maximum a posteriori estimation (MAP). The period value is accurate to one decimal place. The corrected resource reflow period is stored as time series data. The period data supports dynamic updates and is updated once a day. The correction process is implemented by Python scripts involving the statistical library Statsmodels and the probability calculation library PyMC3 to ensure that the period data conforms to actual resource usage.

[0031] Preferably, in step S3, inferring the change trend of fertility resources based on the current fertility resource flow situation and real-time fertility resources includes: Nest resource nodes for the current fertility resource flow situation to obtain nested flow node data; Perform fertility resource mapping based on nested flow node data to obtain mapped fertility resources; The mapped fertility resources and real-time fertility resources are asynchronously spliced ​​to generate asynchronous trajectory splicing data; Capturing trend anchors in asynchronous trajectory splicing data; Identify reproductive resource transition trends in asynchronous trajectory splicing data based on trend anchor points; The changing trend of reproductive resources is inferred through the transition trend of reproductive resources.

[0032] In this embodiment, the time series information of a single resource node is extracted from the resource flow status data obtained by discrete event simulation. The time series includes the start time, end time and resource status identifier of the resource occupancy. By setting the time window length to 10 minutes, the continuous or overlapping resource status nodes are grouped according to the time sequence, and the resource nodes are nested using the hierarchical clustering algorithm. The nesting process is implemented based on the hierarchy module in the Python Scipy library, and the single linkage method is used. linkage) aggregates resource nodes. The clustering threshold is set to no more than 5 minutes based on the resource occupancy time difference. The resource nodes obtained after nesting contain multi-layer nested relationships. Each nested node is defined by the union of its child node time intervals. The node structure is stored in a tree data structure. Node attributes include start time, end time, number of child nodes, and resource type. By traversing the tree structure, a multi-level description of resource flow nodes is achieved. The nested node data format is a JSON structure. The nested flow node data is used as mapping input. The core basis of the mapping is the resource type label and time interval. The mapping operation uses a hash mapping table to achieve fast matching. The hash table key value is the resource type and time period identifier, and the value is the resource attribute vector. The resource attribute vector includes parameters such as resource capacity, resource priority, and resource consumption rate. The vector space mapping technology is used to map the nested node time interval to the fertility resource. In state space, similarity is measured using Euclidean distance. Efficient matrix operations are performed using the Python NumPy library to identify the most similar mapped resource nodes. The mapping process is limited to a time interval error of no more than 2 minutes. The mapping result contains the mapped resource ID and the corresponding state vector, which contains information such as occupancy rate and the proportion of remaining available resources. The mapped resource data structure is stored in a dictionary format. Time alignment is performed on the time series of mapped and real-time fertility resources using linear time interpolation with a fixed interpolation step of 1 minute to ensure high-resolution time series. Asynchronous flow trajectory splicing uses a dual-sequence merging algorithm to match and fuse the resource states at each time point in the two time series. The fusion rule uses a weighted average method, with weights set based on the credibility of the data source. The weight of mapped fertility resources is 0.6, and the weight of real-time fertility resources is 0.4. Use the merge_asof function of the Pandas library to implement time alignment and merging. During the merging process, the resource occupancy rate and availability rate in the overlapping time period are weighted averaged. The trajectory data generated after splicing contains timestamps, resource occupancy rate, resource release rate and total resource change rate. The trajectory data is stored as a time series DataFrame structure. The data time coverage range is the past 24 hours and the resolution is 1 minute. After the splicing operation is completed, the data integrity is checked and the asynchronous trajectory splicing data is smoothed. The weighted moving average algorithm is used. The smoothing window length is set to 15 minutes and the weight decreases linearly. The processed smoothed data is used to detect trend anchor points. The trend anchor point is defined as the local extreme point of the resource state change. The detection method uses a peak detection algorithm based on the zero point of the first-order derivative. Specifically, it calculates the difference in resource change rate at adjacent time points and uses the find_peaks function in the SciPy signal processing module to implement it. The minimum height threshold of the peak is set to 0.05 and the peak interval is 0. The minimum interval is set to 10 minutes. The anchor point information includes timestamp, resource change amplitude and change direction. The change direction is distinguished by positive and negative signs. Positive values ​​indicate an upward trend in resource occupancy, and negative values ​​indicate a downward trend. Trend anchor point data is stored in a list format. Each element in the list contains a timestamp and the corresponding trend type. The capture operation excludes outliers at the same time. Outlier identification is eliminated by the standard deviation 3 times method. The fertility resource transition trend in the asynchronous trajectory splicing data is identified based on the trend anchor point. The identification process uses the trend anchor point as the key node. The resource state change between consecutive anchor points is defined as the transition interval. The transition trend is determined by calculating the average value and change direction of the resource change rate in the transition interval. The transition trend types include rapid rise, slow rise, rapid decline and slow decline. The classification standard is divided by the change rate threshold. The threshold parameter is set to 0.1 per minute for the rapid change threshold and 0.03 per minute for the slow change threshold. The sliding window algorithm is used for data processing. The window size is consistent with the transition interval. Python is used The Pandas library calculates the average rate of change. Transition trend results are output as a dictionary of time intervals and trend types. The dictionary includes the start time, end time, trend type, and average rate of change. The identification process is verified using a jump point detection algorithm. A jump point is defined as the point in time within the transition interval where the change in resource usage exceeds a threshold, which is set to 0.2. The jump point locations are compared with trend anchor points to ensure the accuracy of trend identification. The identification results are output to a time series database. K-means clustering is performed on multiple transition trend sample data. The number of clusters is preset to 4, and the clustering features are trend type and average change rate. The clustering results form a typical change pattern. A linear regression model is then used to fit the resource change trend for each pattern. The regression model inputs are the length of the time series transition interval and the amount of resource change, and the output is the resource change rate over a period of time. The regression model parameters are optimized using the least squares method. Model training is implemented using the Scikit-learn library, and the time window is set to the next 12 hours. The inference results are displayed as resource change trend curves. The curves contain predicted resource utilization values ​​and their confidence intervals. The confidence intervals use a 95% confidence level. The inference results are stored in a structured data table. The data fields include timestamps, predicted resource utilization, and change rates. The inference results serve as an important input for subsequent resource demand forecasting.

[0033] Preferably, in step S4, reversely predicting the change data of the number of expectant mothers based on the change trend of fertility resources and the impact of historical changes includes: The trend of reproductive resource changes and the impact of historical changes are interactively stitched together in time series to obtain a stitched evolution benchmark; Based on the suture evolution benchmark, the change trajectory of reproductive resources is traced back to generate a retrospective resource change trajectory; Confirm the starting data of resource changes based on the retrospective resource change trajectory; Match the resource change amount in the resource change starting data based on the impact of historical changes; The impact of historical changes is used to reversely infer the impact of resource changes, thereby generating data on changes in the number of people waiting to give birth.

[0034] In this embodiment, the fertility resource transition trend data obtained in the previous step is used as the main time series input. The main time series data structure is in DataFrame format. The fields include timestamp, resource type, resource change direction, resource change speed and resource change intensity. The timestamp resolution is 1 minute, and the time range covers the last 72 hours. At the same time, the historical change impact data in the historical fertility resource change record database is called. The historical impact data structure includes event time, resource change intensity, change value of the number of expectant mothers at the same time and the impact propagation time difference. The merge_asof function in the Pandas library is used to merge the two time series. The sequences are time aligned, and the alignment tolerance is set to 3 minutes. Then, the resource type is used as the keyword, and the vector inner product method is used to calculate the cosine similarity between the resource change direction and the historical event impact direction. The similarity threshold is set to no less than 0.85. The resource change segments that meet the similarity conditions are spliced ​​with their corresponding historical impact data, and the spliced ​​data are time-stitched. The stitching method is to correct the current trend data segment based on the time difference of the impact propagation of the historical data, and finally generate the stitched evolution benchmark data. The data format is a nested structure, including the stitching segment start time, stitching segment end time, and resource change synthetic trend value. , the impact vector and the parameters related to the impact of the number of expectant mothers, call the suture trend data saved in the "fusion_baseline" table, and reversely expand the time series of each suture segment. The expansion operation uses the time reversal algorithm to reconstruct the time field in descending time order, set the backtracking step to 1 minute, and set the backtracking interval length to no more than 1.5 times the impact propagation time difference. In the backtracking operation, the resource change value at each time node is reversed and the resource occupancy rate sequence is reconstructed. The resource occupancy rate backtracking value is equal to the current node value minus the change trend multiplied by the time difference. The rolling window function of Pandas is used for calculation The moving average of the backtracking sequence is used to eliminate data jumps. The moving average window size is 5 minutes, and the window function is set to triangular weighting to ensure that recent time points have higher weights. The backtracking trajectory data format generates an independent trajectory record for each stitched segment. Each record includes a timestamp, backtracked resource utilization, backtracked change speed, and reconstructed resource status classification. Based on the resource status sequence of each backtracking trajectory, the resource change starting point is defined as the location where the resource utilization change direction first reverses polarity. This location is achieved by calculating the first-order difference of the sequence with the NumPy library and detecting its positive and negative transition points. The change threshold is set to the resource utilization difference greater than 0.05 and the direction of the difference is reversed. The timestamp of the confirmed starting point is marked as the starting point of the change. 5-minute data segments are extracted before and after the starting point for starting data description. The extracted starting data segments include the original resource status, change speed, resource margin and resource compression rate. The resource compression rate is defined as the ratio of resource usage to available resources. This data is used to measure resource pressure. The starting data structure format is a dictionary structure. Each dictionary item contains the starting time, resource type, starting resource usage rate, corresponding compression rate and change speed. All starting point data items in the "start_point_profile" collection are read. The resource type and resource compression rate are used as matching conditions. The impact event data of the same resource type in the historical change impact data set is called. The data structure includes the event impact starting point, resource change amount, response change value of the number of expectant mothers and impact maintenance time. The K-Nearest Neighbor algorithm (K-Nearest Neighbor) is used. Neighbors (KNN) is used to perform similarity matching of historical events. The feature vector is a two-dimensional vector composed of the initial resource compression rate and the resource change speed. The distance measurement method uses the standardized Euclidean distance, and the K value is set to 3. The three closest historical events are selected and their average resource change is calculated as the matching result. The matched resource change is accompanied by a confidence score, and the score is based on the weighted inverse normalization result of the matching distance. The final matching data includes the predicted resource change, the matching historical event index and the confidence score. The recorded predicted resource change and historical event index are called to obtain the proportional factor of the resource change and the change value of the number of people waiting to be born in the historical event as a regression template. The regression model selects the linear weighted average regression model. The model input is the resource change, and the output is the change value of the number of people waiting to be born. The weight setting is based on the historical events. The confidence scores are linearly normalized, a weighted mapping relationship is established, and the change in the number of expectant mothers is calculated as the weighted sum of all sample predictions. The weighted formula is: the predicted change in the number of expectant mothers is equal to the weighted sum of the resource change in each matching historical event, multiplied by the change coefficient of the number of expectant mothers, and then multiplied by the confidence weight. The final output structure includes the predicted time, the predicted change in the number of expectant mothers, and the index of the historical event with the largest contribution. The output data is formatted as a JSON structure array and written to the Kafka message queue for invocation and update by the real-time prediction service module. The output frequency is set to once every 30 minutes and is automatically archived to the daily prediction log file. The log format is Parquet compressed to save storage space. The fields include the prediction time, predicted value, number of usage events, and the leading event identifier. The file name is named by date to ensure uniqueness.

[0035] It is particularly important to reversely infer the impact of resource changes through the impact of historical changes, including: Divide the resource change into time segments to generate segmented resource change; Perform change sensitivity mapping on the segmented resource change amount and the impact of historical changes to obtain change sensitive response data; Performing backward regression processing based on the change-sensitive response data to generate the shifted expectant labor associated interval data; Perform weight fitting on the data of the shifted waiting period to obtain the corresponding predicted waiting period change interval; The predicted change intervals of the number of people waiting to be delivered are integrated in time series to generate data on changes in the number of people waiting to be delivered.

[0036] In this embodiment, the original resource allocation data is imported into a structured database based on time series management. The resource allocation data includes the daily recorded medical resource input, bed change number, service manpower value and equipment usage change value. The data index field is in a continuous timestamp format with a granularity of once a day. A time period segmentation mechanism is constructed for this type of resource data. 10 days is selected as a basic analysis window length. The starting point of the division is the 1st, 11th and 21st of each month as the time anchor point. A high overlapping rate window is constructed by setting the sliding step to 1 day. A bilateral overlapping strategy is used to construct a resource change absorption segment. All change records in each window are summed up to obtain the total resource change data corresponding to a single time segment, and a segmented resource change is constructed. The sequence is constructed, and each segment contains the time number, window start and end time, total number of changes, maximum change type and dominant change factor. All segment data are written into a sub-table named "resource_segment_block" through structured query language. At the same time, the change trend data graph within the time segment is generated through the drawing engine. The historical event impact score corresponding to the resource change amount of each segment is extracted from the archived data set. The score is calculated based on the change trend evaluation model between resource input and service load in the same period. A three-year sliding evaluation window is used to construct a change trajectory curve for the service demand response data of the same period in history, extract the change value of the coordination degree between resource change amount and service change, and establish a change impact factor sequence. Column, each resource change segment corresponds to a historical impact intensity scalar, and the impact intensity value is unified to the [0,1] interval through the Min-Max normalization method. In order to prevent extreme values ​​from causing fitting offset, the upper and lower bounds are set to the 5th percentile and the 95th percentile using the upper and lower boundary truncation method. Then, the normalized resource change amount and the historical impact factor are input into the multidimensional transformation function, and the change sensitivity mapping value is extracted using the weighted superposition method. Each mapping value is obtained by combining the change amplitude parameter and the historical intensity parameter with weights of 0.65 and 0.35 respectively. All change sensitivity mapping results are saved in the form of floating-point sequences in "response_mapping_data", and each entry is accompanied by the window number and original index The position is used for back-mapping operations. In the process of backward regression processing based on change-sensitive response data, all change-sensitive mapping values ​​are extracted as regression target results. At the same time, a feature data group for each window is constructed. The feature data group consists of the three-segment sliding mean of the resource change amplitude in the current time period, the historical impact residual amplitude, and the change rate, the change category frequency coding, and the sensitive mapping change gradient. All feature data are processed by Z-score standardization and input into the sliding regression prediction network based on the time recursive structure. The network adopts a two-layer recursive structure with 64 units in each layer. The time window length is 5, the number of training iterations is set to 100 rounds, and the regression output is the predicted value of the lag effect of past resource changes on future windows.The backcasting results are expressed in the form of time series association intervals, recorded as "associate_trace_set". The predicted output interval of each time period contains the lagged resource response amplitude and the transition path number. All interval data in "associate_trace_set" are extracted and analyzed one-to-one with the current time series feature data. A change weight estimation model is constructed for each pair of transition association intervals and original resource response data. The partial least squares method is selected to construct the fitting function. Different weight coefficients are assigned to the three types of variables: resource change amplitude, service response lag, and change direction trend. The weight coefficients are dynamically selected through cross-validation. The number of verification folds is set to 10, and the average fitting error value in each fold is back-propagated and adjusted. A penalty term is introduced in each fitting process to suppress overfitting of the change weight. All fitting outputs are predicted change interval sequences. Each interval contains the resource change intensity fitting value, fitting error estimate, time passage index and prediction residual extension segment. Three consecutive prediction intervals are used as a fusion window. The rolling window mechanism with a step size of 1 constructs a time series superposition structure, performs consistency analysis on the prediction output of each interval in the window, extracts the high-correlation change value group, and fits and compresses the fusion interval based on the multi-layer perception network (MLP) structure. The input layer dimension is set to 5 feature dimensions per interval, with a total of 15-dimensional input. The network hidden layer uses the ReLU activation function, with a total of 3 layers, and the dimensions of each layer are 32, 16, and 8 respectively. The output layer is the fused prediction value sequence. The training iteration is 500 rounds through the minimum square error loss function. The change trend value of each time point is extracted from the fusion result, and the sliding mean is used to remove the negative coefficients. Except for instantaneous high-frequency disturbances, a data sequence of changes in the number of expectant mothers is finally generated. Each point contains a time index, a predicted number of changes, a change rate, a change stability, and a prediction confidence interval. The generated results are written to a structured table named "fused_birth_predict_result" for terminal call and visual presentation. All data are output in a structured format for docking and calling by the decision-making system. The graphical results are output in SVG dynamic graphic format for loading by the front-end display module. All processes do not involve human intervention or rule correction mechanisms. The system independently completes data calculation and sequence generation based on the previous modeling parameters, ensuring the logical closure of the prediction process and the traceability of the data generation path. The data structure conforms to the full-process consistency agreement on field length, precision, and time index order in the custom data interface. The output results do not contain non-structural items and implicit data fields, and are suitable for direct deployment in medium and large-scale prediction system modules for regular updates and prediction display calls. All operations do not rely on external platform call modules for execution, but are independently completed based on locally built data processing components and algorithm clusters. The operating environment is a local deployment environment under multi-core parallel computing nodes. All operation logs are regularly cached and version backed up.It is convenient for future data retraining and parameter optimization and updating, avoiding the problem of unnoticed historical model drift, ensuring the stability and accuracy of the model prediction logic. The fusion process is completely automatically tuned by the internal parameters of the model, without the need to manually set fixed weight ratios and time adjustment parameters, ensuring that the time series integration results can synchronously respond to trend disturbances caused by changes in external input data, meeting the timeliness requirements of high-frequency prediction needs and cross-window changes. The output frequency is set to once a day, and the fusion processing flow does not introduce a ring dependency structure and lag state accumulation, avoiding the error accumulation and interval drift problems caused by the cross-time stacking of prediction results, and ultimately forming a high timeliness and high A consistent and complete time-indexed sequence of expected births prediction data results. All fields are standardized, normalized, and structurally encrypted for use in downstream systems for resource matching and service planning response deployment. The output data format is compatible with the structural docking standards defined in the standard data exchange interface. All output fields contain no ambiguous items or fuzzy structural descriptions, ensuring complete consistency between field identification and data alignment during external system docking, preventing system identification conflicts or field call anomalies caused by structural differences. All structural descriptions do not contain polysemous structural tags or ambiguous field indexes, and the output logic is consistent with the forecast generation process.

[0037] Preferably, in step S4, calculating the supporting fertility resources required for the number of expectant mothers based on the historical correlation data and the predicted change data of the number of expectant mothers includes: Perform bidirectional matching and aggregation on historical correlation data and predicted change data on the number of expectant mothers to obtain matching and aggregated expectant mother nodes; Construct a virtual real-time environment based on the real-time number of expectant mothers and real-time fertility resources; Through the virtual real-time environment, the auxiliary resource demand of the matched and collected production nodes is retrieved to generate the auxiliary resource demand data; Based on the demand auxiliary resource data, the corresponding supporting reproductive resources required for the number of expectant mothers are calculated to obtain the predicted demand for reproductive resources.

[0038] In this embodiment, historical data on the number of people waiting to be delivered and corresponding fertility resource allocation data for more than three years are extracted from an archived structured database. The historical data structure includes sixteen dimensional fields, including a timestamp field, a region identification field, a daily number of people waiting to be delivered field, a daily number of obstetric beds, obstetric medical resources, the frequency of delivery surgery, and the amount of auxiliary medication used. All data are integrated through a unique time index and a geographic tag to construct a multidimensional resource allocation matrix. Subsequently, the predicted data sequence of the number of people waiting to be delivered generated in the previous stage is extracted. The predicted sequence field includes six fields: time number, predicted number value, change rate, change acceleration, and prediction confidence lower and upper bounds. A matching algorithm based on multidimensional cosine similarity is used to calculate the similarity between each predicted change record of the waiting to be delivered and the historical record set. In the similarity calculation process, a weight of 0.6 is set for the number of people waiting to be delivered, a weight of 0.3 is set for the change rate, and a weight of 0.1 is set for the prediction confidence. In the matching process, a Top-k strategy is used to select the 10 most similar historical data samples as the collection items, and each prediction period is combined with its most matching historical resource allocation sample set to form a matching collection. A collection of labor nodes is constructed. Each node structure contains a time index, a predicted number of people field, a collection of historical configuration index sets, and a corresponding resource configuration value sequence. The total number of people expected to deliver in the current region's obstetric reception system is received from the real-time monitoring interface at a refresh rate of every 5 minutes. This data comes from the "Real-time Delivery Registration Details" database synchronized by the access-type regional health information exchange node. This database field includes registration time, reception institution number, expected delivery time, obstetric classification label, risk category grouping field, and consultation queue sequence information. At the same time, real-time fertility resource present value data is obtained from the regional health resource allocation node. This data includes resource dimensions such as the current number of available beds, the number of obstetricians on duty, the number of midwives on duty, the number of delivery rooms in use, equipment availability, and anesthesia surgery queue time. The current number of people expected to deliver and real-time resource data are input into the constructed spatial state matrix. Based on the five-dimensional vector field modeling method, each resource dimension is mapped to the feature dimension spatial axis, forming a high-dimensional tensor matrix containing population demand and resource response. During the tensor construction process, the TensorFlow tensor construction interface is used, and the input dimension is set to [number of regions, The matrix state is initialized to zero. The update rule is based on the dynamic changes of resources and the input frequency, and exponential smoothing is performed within the time window. The smoothing factor is set to 0.7. After the construction is completed, the virtual real-time environment field is saved in the tensor array "VIRTUAL_ENV_FRAME" in the local memory, which is used for the resource demand retrieval operation of the matching and collection of the waiting birth nodes. In the process of auxiliary resource demand retrieval of the matching and collection of the waiting birth nodes through the virtual real-time environment field, the generated "matched_birth_nodes" structure table is traversed and compared with the virtual real-time environment tensor "VIRTUAL_ENV_FRAME" item by item. For each waiting birth node, the predicted number of people and the resource configuration records of the historical collection samples are used to construct a demand estimation function. This function takes the resource intensity demand factor as the output target, and the predicted number of people, historical resource allocation rate, real-time available resource status and regional service response speed as input variables. Auxiliary demand value estimation is performed through the resource offset calculation model constructed based on the logistic regression function structure. In this model, the predicted number of people is the main variable, and the proportion of available values ​​in the real-time resource status tensor is used as the correction factor. The logistic regression function outputs Each dimension represents a specific resource demand fit. For example, if the predicted population is 42, the historical average bed allocation is 4.2 beds per 100 people, and the current regional available bed ratio is 0.65, the fitted output bed demand value is (42 / 100*4.2) / 0.65 = 2.71 beds. All auxiliary resource demand values ​​are output and recorded as "ASSIST_RESOURCE_ESTIMATES". This data structure contains the predicted gap values ​​for all resource dimensions under each matching node. Each record is accompanied by the predicted time index, geographic tag, and resource dimension code. All resource demand gap values ​​in "ASSIST_RESOURCE_ESTIMATES" are extracted as target references. First, a weighted fusion operation is performed on the predicted gap data from all matching collection nodes under each time index. The fusion weight is determined by the predicted population confidence value and the historical sample similarity score. The higher the prediction confidence and the greater the similarity, the higher the weight. The maximum weight does not exceed 0.6, and the minimum weight is not less than 0.1. Each resource dimension is synthesized using a linear weighted calculation. The weighted result is then fitted and adjusted with the available resource values ​​in the real-time resource tensor. If the predicted gap exceeds the real-time maximum, the gap value is recorded as a high-voltage resource demand node, which is individually marked for resource scheduling warnings. The final output structure is "PREDICTED_RESOURCE_DEMAND", where fields include the predicted delivery time, the number of required obstetric beds, the number of required birth doctors, the number of required midwives, the required surgical time, the intensity of auxiliary drug use, the equipment scheduling index, and the service load score. Each resource field undergoes structural standardization and precision processing, and the value is recorded as a floating-point structure. All results are written to a structured data table named "final_birth_support_demand", which supports periodic updates and dynamic overwrite write operations. The output structure is completely consistent with the data reception format of the downstream resource scheduling engine, with strict consistency in field length, precision, and field order.

[0039] The present invention further provides a prediction system for fertility resource demand, which is used to execute the above-mentioned prediction method for fertility resource demand. The prediction system for fertility resource demand includes: The spatiotemporal alignment module is used to obtain the real-time number of expectant mothers and real-time fertility resources; perform spatiotemporal joint alignment of the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; and convert the spatiotemporal aligned data into a number of expectant mothers-resource curve. The historical correlation module is used to collect historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; identify historical correlation data through historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; and analyze the impact of historical data on changes in the number of expectant mothers on historical data on reproductive resource consumption based on historical correlation data; The resource deduction module is used to simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; and to infer the trend of fertility resource changes based on the current flow of fertility resources and real-time fertility resources; The reverse prediction module is used to reversely predict the change data of the number of expectant mothers based on the change trend of fertility resources and the impact of historical changes; based on historical correlation data and the predicted change data of the number of expectant mothers, the corresponding supporting fertility resources required for the number of expectant mothers are calculated, thereby obtaining the predicted demand for fertility resources.

[0040] The present invention realizes the joint processing of the real-time number of expectant mothers and fertility resources by constructing a spatiotemporal alignment module, which can form a dynamic matching structure in a multi-dimensional coordinate system, enhance the integration and synchronization of data on the time axis and the space axis, and construct a continuous mapping relationship for subsequent modeling analysis by converting the alignment result into a number of expectant mothers-resource curve. The historical number of expectant mothers change data and resource consumption data are introduced into the historical association module for joint identification, which can extract historical response trajectories with cyclical stability and form a resource evolution chain with upstream and downstream dependency logic. The long-term impact weight matrix of the number of expectant mothers on resource allocation can be obtained by constructing the impact degree of historical changes, so that data in different historical stages have quantifiable and interoperable trend projection capabilities. The historical impact and current structure are integrated in the resource deduction module to simulate the evolution direction of resources in the current state, so that the current resource distribution trend can be structured. The system can express the expected number of births and make inferences based on the expected number of births, and pair it with real-time resource input to obtain the path of future resource changes. In the reverse prediction module, the changing process of the number of expectant mothers is reversed based on the resource trend, forming a reverse mapping network from structural causality to structural demand. The system can also combine and deduce the historical correlation data with the predicted number of births to ensure the predicted amount of supporting fertility resources has periodic continuity and historical consistency. The overall system realizes modular division of labor from data collection, relationship identification, structural simulation to demand forecasting, with the characteristics of structural closed-loop, prediction robustness and output standardization. It is suitable for forecasting and allocation scenarios with complex periodic structures, frequent data fluctuations and strong resource dependence. It has the advantages of strong scalability, high computing efficiency and strong system integration. It still has high precision, high dynamic response capability and high time series adaptability in the context of multi-source heterogeneous data, and can effectively support the intelligent evolution of resource allocation systems and the deep integration of data-driven predictive modeling.

[0041] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed, implements the method for predicting reproductive resource demand as described in any one of the above.

[0042] The present invention embeds a complete method for predicting fertility resource demand into a computer-readable storage medium in the form of a computer program, thereby realizing modular encapsulation of various sub-processes such as data collection, spatiotemporal combination, historical identification, trend deduction and reverse prediction, and constructing a program architecture with deployability and reusability. It still has complete functional adaptability in scenarios with limited computing resources or complex deployment environments. By embedding the collaborative calling logic of real-time data and historical data into the program instruction structure, it can ensure that the program dynamically loads data modules and computing modules during execution, realizes automatic identification of data structures, automatic alignment of time series structures and automatic mapping of spatial structures, and improves the overall data processing efficiency and the degree of information fusion. By predefining the multi-dimensional coordinate projection logic of fertility resources and the number of expectant mothers in the program, the system can automatically generate a spatiotemporal data association map during operation, thereby enhancing the parseability and Updatability: The program supports joint modeling operations of historical evolution paths and current data features, and can dynamically switch between resource response models and population change models in the prediction engine to improve prediction stability and result adaptability. Based on the curve fitting strategy and multi-dimensional mapping mechanism solidified in the program, nonlinear distribution relationships can be converted into continuous fitting expressions, improving the computability and subsequent reasoning capabilities of the output structure. The program module has built-in resource trend backtracking mechanism and population reverse evolution logic. By calling a unified data access interface and time period sliding window structure, it realizes the stable flow of timing logic, ensuring that the prediction output has high timeliness and stable continuity. The instruction set in the computer-readable storage medium has cross-platform deployment capabilities and can be standardized in different hardware architectures, enhancing the system compatibility and resource management capabilities of the technical solution, and is suitable for integrated deployment and stable execution in multi-terminal, multi-node, and multi-stage fertility resource collaborative systems.

[0043] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0044] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting reproductive resource demand, characterized in that: The following steps are involved: Step S1: Obtain the real-time number of expectant mothers and real-time fertility resources; Perform spatiotemporal joint alignment of the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; Converting spatiotemporally aligned data into a waiting list-resource curve; Step S2: Collect historical data on changes in the number of expectant mothers and historical data on consumption of reproductive resources; identify historical related data through the historical data on changes in the number of expectant mothers and historical data on consumption of reproductive resources; Analyze the impact of historical changes in the number of expectant mothers on historical changes in reproductive resource consumption data based on historical correlation data; Step S3: Simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; infer the change trend of fertility resources based on the current flow of fertility resources and real-time fertility resources; Step S4: reversely predict the change data of the number of expectant mothers based on the change trend of fertility resources and the impact of historical changes; Based on historical correlation data and predicted changes in the number of expectant mothers, the corresponding supporting fertility resources required for the number of expectant mothers are calculated to obtain the predicted demand for fertility resources.

2. The method for predicting reproductive resource demand according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: obtaining the real-time number of expectant mothers and real-time fertility resources; performing distribution mapping processing on the real-time number of expectant mothers to obtain spatiotemporal distribution data of the number of expectant mothers; Step S12: normalize the labels of real-time fertility resources, where the normalized label dimensions do not exceed 12 resource types, thereby generating spatiotemporal distribution data of fertility resources; perform multidimensional coordinate projection on the spatiotemporal distribution data of the number of expectant mothers and the spatiotemporal distribution data of fertility resources to obtain spatiotemporal matching data; Step S13: performing spatial alignment on the spatiotemporal matching data, wherein the alignment error threshold is limited to a spatial offset error not exceeding 0.25 grid units, thereby obtaining spatiotemporal aligned data; Step S14: performing surface fitting processing on the spatiotemporal alignment data to obtain spatiotemporal surface structure data; based on the spatiotemporal surface structure data, depicting the spatiotemporal alignment data as a number of expectant mothers-resource curve.

3. The method for predicting reproductive resource demand according to claim 1, characterized in that: In step S2, identifying historical related data through historical data on changes in the number of expectant mothers and historical data on consumption of reproductive resources includes: The historical data on changes in the number of expectant mothers are processed into time series layers, where the length of the layered time window is limited to no more than 72 hours, thereby obtaining a layered series of the number of expectant mothers; Identify resource consumption patterns in historical reproductive resource consumption data, where the minimum continuous usage period for a single resource type must not be less than 4 hours; Perform sequence collaborative retrieval on the stratified sequence of the number of people waiting to be born and the resource consumption pattern to obtain historical correlation data.

4. The method for predicting reproductive resource demand according to claim 1, wherein: In step S2, the impact of the historical change in the number of expectant mothers on the historical change in the historical fertility resource consumption data is analyzed based on the historical correlation data, including: Based on the historical correlation data, query the corresponding correlation features in the historical data on changes in the number of people expecting to give birth and the historical data on consumption of reproductive resources; Locate the relevant change segments in the historical data on changes in the number of expectant mothers and the historical data on consumption of reproductive resources based on the corresponding correlation features; Analyze the historical change data of the number of expectant mothers and the relative change data of historical reproductive resource consumption data based on the corresponding correlation features and correlation change segments; The relative change data is used to determine the impact of historical changes in the number of expectant mothers on historical fertility resource consumption data, and quantify the degree of historical change impact.

5. The method for predicting reproductive resource demand according to claim 1, wherein: In step S3, the simulation of the current fertility resource flow based on the impact of historical changes and the number of expectant mothers-resource curve includes: Confirm current resource occupancy data based on the number of expectant mothers-resource curve; Calculate resource reflow cycle based on current resource occupancy data; Infer available fertility resources through real-time fertility resources and resource return cycles; Conduct time series analysis on current resource occupancy data and available reproductive resources to generate resource return dynamic data; Correcting the resource return dynamic data based on the impact of historical changes to obtain corrected resource return dynamic data; The current flow of reproductive resources is simulated by correcting the dynamic data of resource return.

6. The method for predicting reproductive resource demand according to claim 1, characterized in that: In step S3, the change trend of fertility resources is inferred based on the current fertility resource flow situation and real-time fertility resources, including: Nest resource nodes for the current fertility resource flow situation to obtain nested flow node data; Perform fertility resource mapping based on nested flow node data to obtain mapped fertility resources; The mapped fertility resources and real-time fertility resources are asynchronously spliced ​​to generate asynchronous trajectory splicing data; Capturing trend anchors in asynchronous trajectory splicing data; Identify reproductive resource transition trends in asynchronous trajectory splicing data based on trend anchor points; The changing trend of reproductive resources is inferred through the transition trend of reproductive resources.

7. The method for predicting reproductive resource demand according to claim 1, characterized in that: In step S4, the reverse prediction of the change data of the number of expectant mothers based on the change trend of fertility resources and the impact of historical changes includes: The trend of reproductive resource changes and the impact of historical changes are interactively stitched together in time series to obtain a stitched evolution benchmark; Based on the suture evolution benchmark, the change trajectory of reproductive resources is traced back to generate a retrospective resource change trajectory; Confirm the starting data of resource changes based on the retrospective resource change trajectory; Match the resource change amount in the resource change starting data based on the impact of historical changes; The impact of historical changes is used to reversely infer the impact of resource changes, thereby generating data on changes in the number of people waiting to give birth.

8. The method for predicting reproductive resource demand according to claim 1, characterized in that: In step S4, the calculation of the supporting fertility resources required for the number of expectant mothers based on the historical correlation data and the predicted change data of the number of expectant mothers includes: Perform bidirectional matching and aggregation on historical correlation data and predicted change data on the number of expectant mothers to obtain matching and aggregated expectant mother nodes; Construct a virtual real-time environment based on the real-time number of expectant mothers and real-time fertility resources; Through the virtual real-time environment, the auxiliary resource demand of the matched and collected production nodes is retrieved to generate the auxiliary resource demand data; Based on the demand auxiliary resource data, the corresponding supporting reproductive resources required for the number of expectant mothers are calculated to obtain the predicted demand for reproductive resources.

9. A prediction system for reproductive resource demand, characterized in that: The method for predicting fertility resource demand according to claim 1 is used to implement the fertility resource demand prediction system, comprising: The spatiotemporal alignment module is used to obtain the real-time number of expectant mothers and real-time fertility resources; perform spatiotemporal joint alignment of the real-time number of expectant mothers and real-time fertility resources to obtain spatiotemporal aligned data; and convert the spatiotemporal aligned data into a number of expectant mothers-resource curve. The historical correlation module is used to collect historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; identify historical correlation data through historical data on changes in the number of expectant mothers and historical data on reproductive resource consumption; and analyze the impact of historical data on changes in the number of expectant mothers on historical data on reproductive resource consumption based on historical correlation data; The resource deduction module is used to simulate the current flow of fertility resources based on the impact of historical changes and the number of expectant mothers-resource curve; and to infer the trend of fertility resource changes based on the current flow of fertility resources and real-time fertility resources; The reverse prediction module is used to reversely predict the change data of the number of expectant mothers based on the change trend of fertility resources and the impact of historical changes; based on historical correlation data and the predicted change data of the number of expectant mothers, the corresponding supporting fertility resources required for the number of expectant mothers are calculated, thereby obtaining the predicted demand for fertility resources.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed, the method for predicting reproductive resource requirements according to any one of claims 1 to 8 is implemented.

Images