Population evolution trend intelligent prediction method and system based on regional fertility intention modeling

Abstract

This application relates to an intelligent prediction method and system for population evolution trends based on regional fertility intention modeling. The method includes: acquiring multi-source population evolution data and cleaning the data to obtain a baseline year fusion dataset; aggregating regional features of each region in the baseline year fusion dataset using a graph neural network to obtain regional feature vectors; inputting the regional feature vectors into a fertility intention quantification model to obtain the regional fertility intention baseline probability; constructing an initial agent population set based on the regional fertility intention baseline probability and an agent behavior rule model; and performing agent event simulation based on the initial agent population set and future policy parameter assumptions to obtain a complete simulated event sequence list, and summarizing and statistically analyzing the complete simulated event sequence list to obtain a population prediction result table. This method can introduce multi-agent simulation to improve prediction accuracy.

Description

Technical Field

[0001] This invention belongs to the field of intelligent prediction, and in particular relates to an intelligent prediction method and system for population evolution trends based on regional fertility intention modeling. Background Technology

[0002] With the deep integration of big data and artificial intelligence technologies, demographic research is undergoing profound changes. Against this backdrop, a cutting-edge technological approach integrating multi-agent simulation and deep learning has emerged. Its core characteristic lies in its ability to emerge from the bottom up, starting from micro-level individual interactions, to reveal macro-level population patterns and to perform digital simulations of the impacts of complex social policies. Traditional population forecasting techniques primarily rely on macro-econometric models centered on the cohort factor method. This method treats the population as a homogeneous group, extrapolating based on historical trends by setting key parameters such as age-specific fertility rates, mortality rates, and migration rates to predict future population size and structure. However, current traditional macro-level models have several inherent limitations. First, the homogeneity assumption makes it difficult to characterize the significant heterogeneity of micro-level individuals in terms of family background, economic conditions, and social networks, and it also fails to simulate the interactions and influences between individual decisions. Second, the correlation between model parameters and macro-level socioeconomic variables is often mechanical and static, making it difficult to dynamically respond to the complex and non-linear impacts of exogenous shocks such as fertility incentive policies and economic fluctuations on micro-level decisions. This poses a significant challenge to the explanatory power and predictive accuracy of traditional methods when evaluating sophisticated, combined social policies. Summary of the Invention

[0003] Therefore, it is necessary to provide a method and system for intelligent prediction of population evolution trends based on regional fertility intention modeling, which can construct large-scale multi-agent systems for population prediction, to address the above-mentioned technical problems.

[0004] Firstly, this application provides an intelligent prediction method for population evolution trends based on regional fertility intention modeling, including:

[0005] Acquire multi-source population evolution data and perform data cleaning on the multi-source population evolution data to obtain a base year fused dataset; the multi-source population evolution data includes population census micro-samples and regional annual statistical reports;

[0006] Based on graph neural networks, regional feature aggregation is performed on each region in the baseline year fusion dataset to obtain regional feature vectors;

[0007] The regional feature vector is input into the fertility intention quantification model to predict fertility intention and obtain the baseline probability of regional fertility intention.

[0008] Based on the regional fertility intention baseline probability and agent behavior rule model, multiple agent populations are initialized to obtain the initial agent population set;

[0009] Based on the initial population set of the agents and the assumptions of future policy parameters, agent event simulation is performed to obtain a complete simulated event sequence table. The complete simulated event sequence table is then summarized and statistically analyzed to obtain a population prediction result table.

[0010] Furthermore, the agent population is constructed through the following methods:

[0011] Based on micro-samples from the census, key fields are extracted to obtain a baseline dataset of individual characteristics;

[0012] Based on the individual feature benchmark dataset, key fields are mapped to the internal state variables of the agent population to obtain the agent attribute architecture specification document.

[0013] Based on the quantitative model of fertility intention, the model is reduced in dimensionality to obtain the fertility intention function. Then, based on social impact rules and policy response rules, the fertility intention function is fine-tuned to obtain the agent behavior rule model.

[0014] By integrating the agent attribute architecture specification document and the agent behavior rule model, the basic agent population is obtained;

[0015] Based on the basic agent population, a lifecycle event rule set is added to obtain the agent population. The lifecycle event rule set is used to characterize the core events that drive changes in the agent's state.

[0016] Furthermore, based on the regional fertility intention baseline probability and agent behavior rule model, multiple agent populations are initialized to obtain an initial agent population set, including:

[0017] Based on the individual feature benchmark dataset, attribute values ​​are assigned to each intelligent agent population to obtain a list of basic intelligent agent objects;

[0018] Traverse the region code of each agent population in the list of basic agent objects, and associate the region features and the baseline probability of regional fertility intention of the corresponding region code in the region feature vector with the agent population to obtain the list of environmental connected agents.

[0019] Based on the list of connected agents in the environment, the probability of mutual connection between each agent population is calculated to obtain a gridded list of agents.

[0020] Based on the gridded list of agents, the overall sample is expanded to obtain the initial set of agent population.

[0021] Furthermore, based on the initial agent population set and future policy parameter assumptions, agent event simulations are performed to obtain a complete simulated event sequence list, including:

[0022] A simulated environment is constructed using the initial set of intelligent agent populations as the initial population data and the assumptions of future policy parameters as the external environment parameters.

[0023] Based on the simulation environment, the internal state variables of each agent population and the corresponding regional feature vector of the agent population are input into the agent behavior rule model for prediction to obtain the probability of individual fertility intention.

[0024] By comparing the probability of individual business willingness with the fertility threshold, a list of fertility decision results is obtained, and the initial intelligent agent population set is updated based on the list of fertility decision results to obtain the updated intelligent agent population set;

[0025] By integrating the list of fertility decision outcomes, a complete sequence of simulated events is obtained.

[0026] Furthermore, the updated agent population set is calculated using the following formula:

[0027]

[0028] in, Let be the new state vector of agent i at time period t+1. Let be the state vector of agent i during time period t. Let i be the result of agent i's reproductive decision in time period t. Let t be the external environmental parameters acting on the agent over a time period t, and R be the state transition function.

[0029] Furthermore, the graph neural network is obtained through the following method:

[0030] Based on the list of regions, a regional relationship graph is constructed, with regions as graph nodes and inter-regional relationship data as weighted edges.

[0031] Based on the region list, the region features corresponding to each graph node are extracted from the original region feature data to obtain the node feature matrix;

[0032] Align the historical label data with the graph nodes to obtain the node label vector;

[0033] Based on the region relation graph and node feature matrix, the core layer type and network depth of the basic graph neural network are selected to obtain the basic graph neural network architecture.

[0034] Based on node label vectors, region relationship graphs, and node feature matrices, the basic graph neural network architecture is iteratively trained to obtain a graph neural network.

[0035] Secondly, this application also provides an intelligent prediction system for population evolution trends based on regional fertility intention modeling, including:

[0036] The cleaning module is used to acquire multi-source population evolution data and clean the multi-source population evolution data to obtain a base year fused dataset; the multi-source population evolution data includes population census micro-samples and regional annual statistical reports;

[0037] The feature module is used to aggregate regional features of each region in the baseline year fusion dataset based on graph neural networks to obtain regional feature vectors.

[0038] The prediction module is used to input regional feature vectors into the fertility intention quantification model to predict fertility intentions and obtain the baseline probability of regional fertility intentions.

[0039] The intelligent module is used to initialize multiple agent populations based on the regional fertility intention baseline probability and agent behavior rule model to obtain an initial agent population set;

[0040] The simulation module is used to simulate agent events based on the initial agent population set and future policy parameter assumptions, obtain a complete simulated event sequence table, and summarize and statistically analyze the complete simulated event sequence table to obtain a population prediction result table.

[0041] Thirdly, this application also provides a computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any step of the method provided in the first aspect of this application.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the method provided in the first aspect of this application.

[0043] The aforementioned intelligent prediction method and system for population evolution trends based on regional fertility intention modeling acquires multi-source population evolution data and cleans it to obtain a baseline year fusion dataset. This multi-source data includes micro-samples from the census and regional annual statistical reports. Based on a graph neural network, regional features are aggregated for each region in the baseline year fusion dataset to obtain regional feature vectors. These regional feature vectors are then input into a fertility intention quantification model to predict fertility intentions and obtain the regional fertility intention baseline probability. Based on the regional fertility intention baseline probability and an agent behavior rule model, multiple agent populations are initialized to obtain an initial agent population set. Based on the initial agent population set and future policy parameter assumptions, agent event simulations are performed to obtain a complete simulated event sequence list. The complete simulated event sequence list is then summarized and statistically analyzed to obtain a population prediction result table. The ability to introduce multi-agent simulation fundamentally characterizes the heterogeneity of the population. Individual differences and social relationships are effectively incorporated into the simulation system, improving the reliability of population prediction. By utilizing machine learning technology, core behavioral decision-making models of agents are trained from micro-survey data and macro panel data, enabling decision parameters to adaptively adjust with individual life cycles and changes in the external environment, thereby capturing nonlinear responses and improving the accuracy of predictions. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the process of an intelligent prediction method for population evolution trends based on regional fertility intention modeling provided in an embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram of the structure of an intelligent prediction system for population evolution trends based on regional fertility intention modeling, provided in an embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] In one embodiment, such as Figure 1As shown, an intelligent prediction method for population evolution trends based on regional fertility intention modeling is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0049] Step 101: Obtain multi-source population evolution data and perform data cleaning on the multi-source population evolution data to obtain a base year fused dataset; the multi-source population evolution data includes population census micro-samples and regional annual statistical reports.

[0050] Multi-source population evolution data refers to datasets from multiple sources used to study population change trends. These primarily include census micro-samples and regional annual statistical reports. Census micro-samples are de-identified datasets containing individual-level information, while regional annual statistical reports summarize annual macro-indicators for various regions. Data cleaning refers to the preprocessing of the acquired raw multi-source data. Its purpose is to correct errors, handle missing values, and standardize formats and criteria to improve data quality. The baseline year fusion dataset refers to a unified dataset for a specific baseline year, formed after cleaning and integration. This dataset combines individual characteristics from micro-samples with macro-regional characteristics from statistical reports, serving as the foundation for all subsequent analyses.

[0051] The terminal obtains two main categories of raw data from different sources: micro-samples from the census and regional annual statistical reports. It performs data cleaning operations, checks and corrects outliers in the data, fills in missing data items, aligns and standardizes data fields from different sources, and ensures the consistency of key identifiers such as regional codes. The cleaned micro-individual data are then associated, matched and merged with the corresponding macro-regional statistical data through key fields such as regional codes, according to their respective regions, to generate a base year fusion dataset that can be directly used in subsequent steps.

[0052] Step 102: Based on graph neural networks, perform regional feature aggregation on each region in the baseline year fusion dataset to obtain regional feature vectors.

[0053] Specifically, a graph neural network (Graph Neural Network) is a deep learning model specifically designed for processing graph-structured data. In this embodiment, the Graph Neural Network treats each geographical region as a node in a graph, and the relationships between regions, including geographical adjacency, population flow, and economic ties, as edges connecting the nodes. Regional feature aggregation refers to the operation of comprehensively learning and condensing information about each region and its surrounding related regions using the Graph Neural Network model. A regional feature vector is a numerical vector used to represent the comprehensive characteristics of a region. The regional feature vector is a higher-level abstract feature representation that aggregates the region's attributes and their influence from surrounding regions in the regional relationship network through the Graph Neural Network.

[0054] The terminal uses a baseline year fusion dataset as a basis, extracts data from each region, and applies a graph neural network to the data in each region. Based on predefined inter-regional relationships, the graph neural network automatically learns how to integrate the various features of a region itself, and considers the features of related regions. Through regional feature aggregation operations, the graph neural network calculates and outputs a numerical representation with fixed dimensions and rich information for each region in the dataset, namely the regional feature vector.

[0055] Step 103: Input the regional feature vector into the fertility intention quantification model to predict fertility intention and obtain the baseline probability of regional fertility intention.

[0056] Specifically, the fertility intention quantification model is a trained mathematical model that takes a numerical vector describing regional characteristics as input and outputs a quantified value representing the average fertility intention level of that region. Fertility intention prediction is an operation that uses the fertility intention quantification model to estimate the level of fertility intention in a region. The regional fertility intention baseline probability is a probability value that represents the average likelihood or propensity for individuals in a region to have fertility intentions under a baseline year or current conditions; it is a macroscopic measure of the overall fertility climate of the region.

[0057] The terminal inputs the generated regional feature vectors for each region into a pre-built fertility intention quantification model. The fertility intention quantification model calculates fertility intentions based on the input feature vectors. For each input regional feature vector, the fertility intention quantification model outputs a corresponding probability value, which is the baseline probability of fertility intention for that region. The baseline probability of regional fertility intention reflects the predicted strength of fertility intention based on the comprehensive characteristics of the region's socio-economic situation, population structure, etc.

[0058] Step 104: Based on the regional fertility intention baseline probability and agent behavior rule model, initialize multiple agent populations to obtain the initial agent population set.

[0059] The agent behavior rule model is a set of computational rules and logic defining how micro-level individuals make behavioral decisions. It includes modules for individual state assessment, social impact calculation, and responses to policy stimuli. Initialization is the process of creating and setting the states of all micro-level individuals at the start of the simulation. The agent population refers to the computational entities used in the simulation to represent real individuals. Each agent population possesses a set of attributes and follows specific behavioral rules. The initial agent population set refers to the total number of agents created and assigned initial attributes at the start of the simulation; this initial set is the starting point for the dynamic simulation.

[0060] Based on the actual population structure distribution of the baseline year, the terminal generates a large number of intelligent agents with different attributes, allocates the intelligent agents to different regions, and assigns the baseline probability of the fertility intention of the region where the agent is located as part of its initial environmental pressure or social norm parameters. Each intelligent agent is configured with an intelligent agent behavior rule model, enabling it to make decisions in the simulation. All the created and configured intelligent agents together constitute the initial intelligent agent population set.

[0061] Step 105: Based on the initial agent population set and future policy parameter assumptions, conduct agent event simulation to obtain a complete simulated event sequence table, and summarize and statistically analyze the complete simulated event sequence table to obtain a population prediction result table.

[0062] The future policy parameter assumptions are a set of variable values ​​used to set the simulated future scenarios, representing the strength or type of various population-related policies that may be implemented. Agent event simulation is a computational process in which each agent in the initial agent population set autonomously interacts within the simulated time step to make decisions and drive time forward, based on its behavioral rules and the external environment. The complete simulation event sequence list records detailed information on various important events (such as births, deaths, migrations, etc.) triggered by all agents throughout the entire simulation period, arranged chronologically. Summary statistics are operations that process, classify, and calculate the raw event records in the complete simulation event sequence list. The population prediction results table is a structured data table obtained after summarizing and statistically analyzing the simulated events, containing predicted indicators such as total population, age structure, number of births, and number of deaths for each future year or time period.

[0063] The terminal starts the simulation with an initial set of agent populations and sets a set of future policy parameters as the external environmental conditions that change during the simulation. It then initiates agent event simulation. In multiple consecutive time steps, each agent evaluates and makes behavioral decisions based on its own attributes, social network, agent behavior rule model, and current future policy parameter assumptions. Each decision triggers an event, forming a complete time-ordered sequence of simulated events. After the simulation, the event list is summarized and statistically analyzed. This involves classifying, counting, and calculating events according to demographic standards, and then compiling the statistical results into a clear population prediction table that shows future trends.

[0064] This embodiment provides an intelligent prediction method for population evolution trends based on regional fertility intention modeling. It acquires multi-source population evolution data and cleans it to obtain a baseline year fusion dataset. This multi-source data includes micro-samples from the census and regional annual statistical reports. Based on a graph neural network, it aggregates regional features from each region in the baseline year fusion dataset to obtain regional feature vectors. These regional feature vectors are then input into a fertility intention quantification model to predict fertility intentions and obtain the regional fertility intention baseline probability. Based on the regional fertility intention baseline probability and an agent behavior rule model, it initializes multiple agent populations to obtain an initial agent population set. Based on the initial agent population set and future policy parameter assumptions, it simulates agent events to obtain a complete simulated event sequence list. Finally, it summarizes and statistically analyzes the complete simulated event sequence list to obtain a population prediction result table. Through the above methods, multi-agent simulation can be introduced to fundamentally characterize the heterogeneity of the population. Individual differences and social relationships are effectively incorporated into the simulation system, improving the reliability of population prediction. By using machine learning technology, the core behavioral decision-making model of the agent is trained from micro-survey data and macro panel data, so that the decision parameters can be adaptively adjusted with individual life course and changes in the external environment, thereby capturing nonlinear responses and improving the pertinence of prediction.

[0065] In one embodiment, the agent population is constructed using the following method:

[0066] Step 201: Based on the micro-samples of the population census, extract key fields to obtain the individual characteristic benchmark dataset.

[0067] The census micro-sample refers to a dataset containing numerous individual-level records that has undergone anonymization. Each record represents anonymized information about a real individual, including fields such as age, gender, years of education, occupation, marital status, place of residence, and relationship to the head of household. Key field extraction refers to the process of selecting the most important data items for population simulation research from the vast amount of raw micro-sample data. The individual characteristic benchmark dataset refers to the dataset obtained after key field extraction. This dataset contains a series of core characteristic variables used to describe and distinguish different individuals, serving as the original data foundation for constructing intelligent agent population profiles.

[0068] The terminal analysis of fertility intention research identifies which individual attributes are key influencing factors. It performs key field extraction, selecting pre-determined key fields from all fields in the original sample. The data records of all individuals corresponding to the selected key fields are organized, exported, and formed into a dedicated individual characteristic benchmark dataset for further use.

[0069] Step 202: Based on the individual feature benchmark dataset, map the key fields to the internal state variables of the agent population to obtain the agent attribute architecture specification document.

[0070] Specifically, mapping refers to the process of assigning a specific name, with a particular meaning and data type, to each key field in the dataset as a corresponding state variable name in the agent model. Internal state variables of an agent population refer to a set of variables in the computational model used to represent the internal attributes of an agent population, such as age, gender, education level, and marital status. The values ​​of these internal state variables can change over time. The agent attribute architecture specification document is a design document that formally defines the names, data types, value ranges, and correspondences between all internal state variables included in the agent and the original fields in the individual feature baseline dataset.

[0071] The terminal takes an individual characteristic benchmark dataset as input and maps each key field in the dataset that is useful for simulation to a corresponding internal state variable in the agent model. For example, the years of education in the data is mapped to a variable for the agent. After completing the mapping design for all key fields, the variable definitions, mapping relationships, and data constraints are formally recorded in a specification file, thus generating the agent attribute architecture specification document. The agent attribute architecture specification document serves as the basic blueprint for subsequent programming implementation of the agent class.

[0072] Step 203: Based on the fertility intention quantification model, perform model dimensionality reduction to obtain the fertility intention function, and fine-tune the fertility intention function based on social impact rules and policy response rules to obtain the agent behavior rule model.

[0073] Specifically, the fertility intention quantification model is a mathematical model capable of predicting the probability of macro-level fertility intentions based on regional characteristics. Model dimensionality reduction refers to simplifying or transforming the relatively complex fertility intention quantification model to obtain a more lightweight core function with individual characteristics as input. The fertility intention function is the output of the model dimensionality reduction operation; it is a mathematical function whose input is the agent's internal state variables, and whose output is an individual fertility intention probability value between 0 and 1, representing the macro-level model at the individual level. The social influence rule is a set of computational rules used to quantify the degree to which the attitudes or behaviors of other individuals in the agent's social network influence the agent's fertility intention. The policy response rule is a set of computational rules used to quantify the degree to which changes in the external policy environment affect the agent's fertility intention. Fine-tuning refers to the operation of correcting the final output value of the function based on the fertility intention function by introducing the computational results of the social influence rule and the policy response rule as adjustment factors. The agent behavior rule model is a complete set of rules integrating core computational logic and dynamic adjustment mechanisms, defining the computational process by which an agent makes the crucial decision of whether or not to have children based on its own state, social environment, and external policies.

[0074] The terminal performs dimensionality reduction on the macro-level quantitative model of fertility intentions. By analyzing the correlation between its key parameters and individual characteristics, a simplified function with individual characteristics as independent variables is derived, resulting in the basic fertility intention function. The fertility intention function is then fine-tuned by designing social impact rules and policy response rules. The proportion of friends who have already given birth within the agent's social circle is calculated and influenced by certain weights. Based on the current amount of childcare subsidies, the intention value is increased according to a specific formula. The calculation results of these two sets of rules are used as adjustment terms and integrated into the calculation process of the basic fertility intention function. This complete calculation system, including the core function and dynamic adjustment rules, constitutes the agent's behavioral rule model.

[0075] Step 204: Integrate the agent attribute architecture specification document and agent behavior rule model to obtain the basic agent population.

[0076] Integration refers to the process at the programming implementation level of combining the architecture document defining the static attributes of an agent with the rule model defining its dynamic behavior into a unified agent class or prototype. The basic agent population refers to the collection of agent instances created according to the agent attribute architecture specification document and equipped with the agent behavior rule model. These agents possess basic attributes and core behavioral logic, but do not yet include the driving mechanisms for other lifecycle events.

[0077] Based on the intelligent agent attribute architecture specification document, the terminal creates a data structure for the intelligent agent class, which contains all internal state variables. The complete computational logic of the intelligent agent behavior rule model is implemented as a core method of the intelligent agent class. Through integrated operations, each instantiated intelligent agent object has the specified attributes and executable behavioral decision logic. The collection of such intelligent agent objects is the basic intelligent agent population.

[0078] Step 205: Based on the basic agent population, add a lifecycle event rule set to obtain the agent population. The lifecycle event rule set is used to characterize the core events that drive changes in the agent's state.

[0079] The lifecycle event rule set defines the triggering conditions and state update rules for major events, excluding reproductive decisions, that fundamentally change the state of an agent. Events can include death, forming a partnership, divorce, migration, etc. Adding refers to programming the decision logic and state update logic of the lifecycle event rule set into the agent class of the basic agent population. The agent population refers to a fully functional set of agents that, based on the basic agent population, have further integrated and activated the lifecycle event rule set. At this point, the agents can respond to multiple events throughout a complete simulated lifecycle, and their states can be changed by these core events.

[0080] The terminal is based on a complete set of lifecycle event rules, which defines the probability model or triggering conditions for each key life event such as death and migration, and how the agent's state is updated after the event occurs. For example, if a death event is triggered, the agent is marked as removed; if a migration event is triggered, its residence code is updated and an addition operation is performed. The event judgment and processing logic is programmed into the agent model. The resulting set of agents with complete attributes, reproductive behavior rules and full lifecycle event processing capabilities is the agent population that can be used for long-term dynamic simulation.

[0081] In one embodiment, based on the regional fertility intention baseline probability and agent behavior rule model, multiple agent populations are initialized to obtain an initial agent population set, including:

[0082] Step 301: Based on the individual feature benchmark dataset, assign attribute values ​​to each agent population to obtain a list of basic agent objects.

[0083] The individual characteristic benchmark dataset is a structured collection of data containing core individual characteristics extracted from micro-samples of the census. Each record represents an attribute of a sampled individual, such as age, gender, and education level. The agent population is a computational entity used in the simulation to represent real individuals. Attribute assignment refers to retrieving specific values ​​from its corresponding data source to initialize the internal state variables of each newly created agent population instance. The basic agent object list is a list containing multiple agent population objects. Each agent object in the basic agent object list has already completed attribute assignment, meaning its attributes have been assigned actual values ​​derived from a specific record in the individual characteristic benchmark dataset.

[0084] The terminal creates a corresponding number of empty agent population objects based on the number of records in the dataset, performs attribute assignment operations, iterates through each individual record in the dataset, and assigns attribute values ​​to the corresponding agent object one by one. For example, the age field value in the data record is assigned to the agent's age attribute, and the education level field value is assigned to the agent's education level attribute. After processing all records in the dataset, a list of basic agent objects corresponding one-to-one with the sampled data, with all agent attributes initialized, is obtained.

[0085] Step 302: Traverse the region code of each agent population in the basic agent object list, and associate the region features and the baseline probability of regional fertility intention of the corresponding region code in the region feature vector with the agent population to obtain the environmental connected agent list.

[0086] Specifically, the basic agent object list is a collection of agent objects whose individual attributes have been initialized. The region code is a unique code identifying the administrative or geographical region to which the agent belongs. The region feature vector is a numerical vector representing the comprehensive characteristics of a region, obtained through graph neural network aggregation. The regional fertility intention baseline probability is a probability value representing the average fertility intention level of individuals within a region. Association refers to the operation of establishing references or assignments from macro-regional data to micro-level agent objects, enabling agents to obtain macro-environmental information about their region. The environment-connected agent list is a new list obtained by updating the basic agent object list. Each agent object in the list, in addition to possessing individual attributes, is also associated with the region feature vector and the regional fertility intention baseline probability as its perceived environmental parameters.

[0087] The terminal iterates through each agent population in the basic agent object list, reads its regional code attribute, performs association operations, and finds the corresponding regional feature vector and regional fertility intention baseline probability in all stored regional information based on the read regional code. These two macro data values ​​are then assigned or linked to the currently iterated agent object as environmental variables. After completing the traversal and association of all agents, the updated agent list is the environment-connected agent list.

[0088] Step 303: Based on the environmental connected agent list, calculate the probability of mutual connection between each agent population to obtain a gridded agent list.

[0089] Specifically, the environmental connected agent list is a list of agent objects already associated with a macro-regional environment. The probability of mutual connection is a value used to calculate the likelihood of a social association between any two agents, calculated based on the similarity of certain agent attributes or specific rules. Calculation refers to the process of determining whether a connection relationship is established between agents according to preset rules or probability models. The gridded agent list is the result of further processing based on the environmental connected agent list. The agents in the list have established network connections simulating social relationships based on the calculated mutual connection probabilities, thus forming an agent group with a specific topology.

[0090] The terminal defines a set of rules based on research hypotheses or prior knowledge to determine the probability of social connections between two agents. That is, it defines a method for calculating the probability of mutual connection, performs calculation operations, traverses the list of agents, and calculates the probability value of forming a connection relationship between each pair of agents according to the defined rules. Based on the calculated probability value, it decides whether to actually establish a connection edge between the pair of agents through random sampling or other determination methods. When the connection relationships between all pairs of agents have been determined and established, the entire set of agents is transformed from a simple list into a network structure. The agent network with embedded connection relationships is the gridded agent list.

[0091] Step 304: Based on the gridded list of agents, perform overall expansion to obtain the initial agent population set.

[0092] The gridded agent list is a list of agents with established internal social network connections. Population expansion is a statistical processing method that, based on sampling weights, extends the results obtained from a single sample to the complete population it represents. The initial agent population set is the set of agents obtained after population expansion, representing the entire population of the target area. Its size is much larger than the input sample list, serving as the initial population basis for large-scale population simulation.

[0093] The terminal determines the sampling weight represented by each agent in the sample and replicates and expands the gridded agent list according to the sampling weight. For example, an agent with a weight of 100 represents 100 individuals with similar characteristics in the population. During the expansion process, 99 copies of the agent need to be created, and these copies, along with the original agent, are incorporated into a new set. Simultaneously, the network connections between the newly generated agent copies are adjusted according to rules. Through weighted replication and expansion, a larger initial agent population set is obtained, designed in both quantity and structure to represent the real population distribution of the target area.

[0094] In one embodiment, based on an initial agent population set and assumptions about future policy parameters, agent event simulation is performed to obtain a complete simulated event sequence list, including:

[0095] Step 401: Using the initial intelligent agent population set as the initial population data and the future policy parameter assumptions as the external environment parameters, construct a simulation environment.

[0096] The initial agent population set refers to the set of agents that, after overall expansion, can represent the entire population of the target area at the start of the simulation. Each agent already possesses individual attributes, environmental associations, and potential social network connections. Future policy parameter assumptions are a set of variable values ​​used to set the simulated future scenarios. They represent the hypothetical content and intensity of various population-related policies that may be implemented or change within the simulation time span, such as childcare subsidies, maternity leave days, and educational resource allocation. External environment parameters, in the simulation context, specifically refer to those external variables uniformly set by the simulation system that can affect the decisions and states of all or some agents. In this embodiment, future policy parameter assumptions are specifically designated as external environment parameters in the simulation. The simulation environment is a virtual computational framework whose core includes the initial agent population set and a set of external environment parameters that evolve over time. The simulation environment provides the spatiotemporal and rule-based foundation for subsequent agent interactions and event simulations.

[0097] The terminal uses the initial set of intelligent agent populations as the total population data at the start of the simulation, and uses a set of preset future policy parameters as the external influencing factors to be applied during the simulation. It creates a management entity in the calculation program, loads all intelligent agents in the initial set of intelligent agent populations, and initializes variables such as future policy parameters as external environment parameters that can be perceived and responded to by intelligent agents in this environment. A simulation environment containing the initial state subjects and external driving conditions is thus established.

[0098] Step 402: Based on the simulation environment, input the internal state variables of each agent population and the corresponding regional feature vector of the agent population into the agent behavior rule model for prediction to obtain the individual fertility intention probability.

[0099] Specifically, internal state variables refer to the set of variables used to describe the current intrinsic attributes of an agent, such as age, income, education level, marital status, and number of children. Regional feature vectors refer to the comprehensive numerical representation of the features of the agent's region, obtained through graph neural network aggregation. The agent behavior rule model refers to a complete computational logic integrating the core fertility intention function, social influence rules, and policy response rules, used to drive the agent's fertility decisions. Prediction specifically refers to the operation of using the agent behavior rule model to calculate a quantified probability value. The individual fertility intention probability is a probability value between 0 and 1, representing the likelihood that a specific agent, based on its own state, regional characteristics, and the social and policy influences it experiences, will have a fertility intention at the current simulation point in time and environment.

[0100] The terminal acquires the agent's current internal state variables and its associated regional feature vectors, and inputs them into the agent's behavior rule model for prediction. The model integrates individual characteristics, regional macro-characteristics, built-in social impact calculations, and policy parameters in the current simulation environment, runs its internally defined calculation process, and outputs a specific probability value for the agent, namely, the probability of an individual's fertility intention.

[0101] Step 403: Compare the individual business willingness probability with the fertility threshold to obtain a list of fertility decision results, and update the initial agent population set based on the list of fertility decision results to obtain the updated agent population set.

[0102] Specifically, the fertility threshold is a preset critical value or decision rule used to transform the continuous probability of an individual's fertility intention into a discrete yes / no decision. The fertility threshold can be a fixed value, a function that varies with individual characteristics, or a randomly obtained value. Comparison refers to comparing an individual's fertility intention probability with the fertility threshold and making a binary decision based on the comparison result. The fertility decision result list is a list, ordered by agent number or sequence, recording the fertility decision result generated by each agent after comparison in the current simulation cycle, typically indicating whether to fertilize or not. Update refers to modifying the state of relevant agents in the initial agent population set or increasing / decreasing the number of agents based on the agents' decision results and other lifecycle rules. Updating the agent population set is the process of modifying the state and adjusting the members of the initial agent population set at the end of the current simulation cycle, based on the fertility decision result list and other parallel-processed lifecycle event results. This updated agent population set reflects the population state after the simulation has progressed one time step.

[0103] The terminal performs a comparison operation on the probability of each individual's fertility intention, comparing it with a preset fertility threshold. If the probability value is greater than or meets the threshold condition, the agent is determined to decide to conceive; otherwise, they are determined not to conceive. The decisions of all agents are collected and recorded to form a fertility decision result list, and the state is updated. Based on the fertility decision result list, all agents are traversed. For agents that have decided to conceive, their fertility behavior is simulated, including adding agents representing newborns to the population set and updating the state such as the number of children of their parents. At the same time, other non-fertility events are processed in parallel. After completing all state changes and member additions and subtractions, the updated agent population set is obtained.

[0104] Step 404: Integrate the list of fertility decision results to obtain a complete simulated event sequence list.

[0105] The reproductive decision results list records the reproductive decisions of all agents within the current simulation period. Integration refers to the operation of merging newly generated event records into the total event record. The complete simulation event sequence list is a detailed list of all important events recorded throughout the entire simulation process, arranged chronologically. Each record typically includes information such as the time the event occurred, the agents involved, and the event type.

[0106] The terminal converts each fertility decision in the fertility decision result list into a formatted fertility event record. Newly generated fertility event records are added to or appended to the overall complete simulation event sequence table in chronological order of occurrence. If other types of events are processed within the same time period, their records are also formatted and integrated into this sequence table. The complete simulation event sequence table grows continuously, forming a comprehensive log containing all key events throughout the simulation process.

[0107] In one embodiment, the updated agent population set is calculated using the following formula:

[0108]

[0109] in, Let be the new state vector of agent i at time period t+1. Let be the state vector of agent i during time period t. Let i be the result of agent i's reproductive decision in time period t. Let t be the external environmental parameters acting on the agent over a time period t, and R be the state transition function.

[0110] Specifically, It is a collection of multiple elements used to characterize all the intrinsic attributes of an agent at a specific moment. Attributes include age, gender, education level, marital status, number of children, health status, occupation, income level, etc. It defines a complete snapshot of the agent at simulation time t.

[0111] It is a discrete decision variable generated by the output of the agent's behavior rule model and compared with the fertility threshold. Its value is usually a binary result and is one of the key event inputs that drive the change of the agent's state within the period t.

[0112] It is a set of variables uniformly set by the simulation system, which the individual intelligent agent cannot control but can perceive and be affected by. The most typical component is the assumption of future policy parameters, such as the current amount of maternity subsidies, the proportion of personal income tax reduction, and the accessibility index of childcare services. It represents the external macro-environment in which the intelligent agent exists.

[0113] R is a complex set of functions or rules that encapsulates all state transition logic. It defines how to calculate the agent's new state at the next moment based on the agent's current state, the latest events, and the external environment.

[0114] This represents the set of all updated attribute values ​​for the agent after one simulation period. This represents the specific new state of the agent within the updated agent population set.

[0115] In one embodiment, the graph neural network is obtained through the following method:

[0116] Step 601: Based on the list of regions, construct a region relationship graph with regions as graph nodes and inter-regional relationship data as weighted edges.

[0117] The region list is a list containing unique identifiers for all regions under study. A graph node is the basic unit representing an entity or object in a graph data structure. In this embodiment, each graph node corresponds to a specific region in the region list. Inter-regional relationship data is used to quantify the strength or closeness of the relationship between any two regions. This can be population migration flow, economic exchange intensity, geographical adjacency, or commuting time, etc. A weighted edge is a line segment connecting two graph nodes in a graph data structure. This line segment is assigned a value to represent a measure of the relationship between the nodes, and the weight value comes from the inter-regional relationship data. The regional relationship graph is a graph-structured data object where each region is represented as a graph node, and the inter-regional relationship data between any two regions is quantified as a weighted edge connecting these two nodes, thus depicting the interconnected network of all regions.

[0118] The terminal creates a corresponding graph node for each region in the region list. For each pair of regions, it queries the inter-region association data between the graph nodes and uses the association data value as the weight value of the edge connecting the two nodes. If there is no association between two regions or the data is zero, no edge is established between the nodes. All nodes and weighted edges together constitute a complete region relationship graph.

[0119] Step 602: Based on the region list, extract the region features corresponding to each graph node from the original region feature data to obtain the node feature matrix.

[0120] Specifically, raw regional feature data is a collection of raw data describing various attributes of each region, obtained from various statistical reports or surveys, including indicators such as average years of schooling, density of medical resources, and child dependency ratio. Regional features are a set of representative indicator values ​​extracted from the raw regional feature data for a specific region. The node feature matrix is ​​a two-dimensional matrix. Each row of the matrix corresponds to a graph node in the regional relationship graph, and the order of the rows is consistent with the order of the region list. Each column of the matrix represents a specific regional feature indicator; the value of the element in the i-th row and j-th column of the matrix represents the value of the i-th region on the j-th feature indicator. The matrix serves as the input feature for nodes in a graph neural network.

[0121] The terminal determines a set of key regional feature indicators needed for model training and processes each region sequentially according to the region list. For the current region, it searches and extracts the specific values ​​of the region for all selected indicators from the original region feature data. The set of values ​​for the selected indicators constitutes the region features. All region features are stacked into a two-dimensional array in the same order as the region list; this two-dimensional array is the node feature matrix. The rows of the matrix correspond to regions, and the columns correspond to feature dimensions.

[0122] Step 603: Align the historical label data with the graph nodes to obtain the node label vector.

[0123] Specifically, historical label data refers to known, real-valued data directly related to the prediction target, used to supervise model training. In this embodiment, it typically refers to the actual observed values ​​of indicators to be predicted, such as fertility rates and population growth rates, for each region in historical years. Graph nodes are nodes in a regional relationship graph, with each node representing a region. The alignment operation refers to the process of accurately associating each data record in the historical label data with its corresponding graph node in the regional relationship graph, using a region identifier. The node label vector is a one-dimensional vector with the same length as the region list. The value at the i-th position in the vector is the historical label data value of the region corresponding to the i-th graph node, found through the alignment operation. The vector represents the target value during graph neural network training.

[0124] The terminal ensures that the nodes in the historical label data and the regional relationship graph use the same regional identifier system. It iterates through each node in the regional relationship graph, and for each node, based on its regional identifier, searches for the corresponding label value in the historical label data. Once found, the label value is recorded in a list according to the node's order. This list of label values ​​arranged in node order constitutes the node label vector. The node label vector provides each node with a true target value for supervised learning.

[0125] Step 604: Based on the region relationship graph and node feature matrix, select the core layer type and network depth of the basic graph neural network to obtain the basic graph neural network architecture.

[0126] The core layer type of a basic graph neural network refers to the types of basic computational units used to construct the graph neural network, such as graph convolutional layers, graph attention layers, and graph sampling and aggregation layers. Different layer types determine the specific methods of information aggregation and transmission between nodes. Network depth refers to the number of core layers or stacked layers in the graph neural network. Depth affects how far the model can capture neighborhood information. The basic graph neural network architecture is a graph neural network model framework whose structure is determined before training. It specifies how many layers the network will use, what type of layer each layer will use, and how the layers will be connected. It defines the computational blueprint of the model.

[0127] Depending on the task complexity and data scale, the terminal may decide to use two graph convolutional layers as the core layers, i.e., a network depth of 2. Based on this selection, a specific, uninitialized graph neural network model with the corresponding layer type and number of layers is instantiated. The input to the graph neural network model architecture is designed to receive the node feature matrix and the topology of the region relation graph.

[0128] Step 605: Based on the node label vector, the region relationship graph, and the node feature matrix, iteratively train the basic graph neural network architecture to obtain the graph neural network.

[0129] Iterative training refers to using optimization algorithms to repeatedly input data into the model, calculate the error between the predicted and actual values, and adjust the model's internal parameters based on the error to make the model's predictions increasingly accurate. Graph neural networks are usable machine learning models obtained through iterative training of a basic graph neural network architecture, with optimized and determined parameters. The model can learn spatial dependencies between regions and output predictions of node labels based on the input node feature matrix and region relationship graph.

[0130] The terminal uses node label vectors (as training targets), region relation graphs (providing structural information), and node feature matrices (providing node attribute information) as training data. The basic graph neural network architecture, training data, a loss function, and an optimizer are configured. The node feature matrix and region relation graph are repeatedly input into the current model to obtain predicted node labels; the loss between the predicted values ​​and the true node label vectors is calculated; the gradient of the loss with respect to the model parameters is calculated using the backpropagation algorithm; and the optimizer updates the model parameters based on the gradient. After multiple iterations, training ends when the model performance stabilizes or reaches a preset stopping condition. A graph neural network suitable for prediction is obtained.

[0131] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0132] Based on the same inventive concept, this application also provides an intelligent prediction system for population evolution trends based on regional fertility intention modeling, used to implement the aforementioned intelligent prediction method for population evolution trends based on regional fertility intention modeling. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent prediction system for population evolution trends based on regional fertility intention modeling provided below can be found in the limitations of the intelligent prediction method for population evolution trends based on regional fertility intention modeling described above, and will not be repeated here.

[0133] In one exemplary embodiment, such as Figure 2 As shown, a population evolution trend intelligent prediction system 700 based on regional fertility intention modeling is provided, including:

[0134] The cleaning module 701 is used to acquire multi-source population evolution data and clean the multi-source population evolution data to obtain a base year fused dataset; the multi-source population evolution data includes population census micro-samples and regional annual statistical reports;

[0135] Feature module 702 is used to aggregate regional features of each region in the baseline year fusion dataset based on graph neural network to obtain regional feature vectors.

[0136] The prediction module 703 is used to input the regional feature vector into the fertility intention quantification model to predict fertility intention and obtain the baseline probability of regional fertility intention.

[0137] The intelligent module 704 is used to initialize multiple agent populations based on the regional fertility intention baseline probability and agent behavior rule model to obtain an initial agent population set.

[0138] The simulation module 705 is used to simulate agent events based on the initial agent population set and future policy parameter assumptions, obtain a complete simulated event sequence table, and summarize and statistically analyze the complete simulated event sequence table to obtain a population prediction result table.

[0139] Furthermore, the system also includes a model module for:

[0140] Based on micro-samples from the census, key fields are extracted to obtain a baseline dataset of individual characteristics;

[0141] Based on the individual feature benchmark dataset, key fields are mapped to the internal state variables of the agent population to obtain the agent attribute architecture specification document.

[0142] Based on the quantitative model of fertility intention, the model is reduced in dimensionality to obtain the fertility intention function. Then, based on social impact rules and policy response rules, the fertility intention function is fine-tuned to obtain the agent behavior rule model.

[0143] By integrating the agent attribute architecture specification document and the agent behavior rule model, the basic agent population is obtained;

[0144] Based on the basic agent population, a lifecycle event rule set is added to obtain the agent population. The lifecycle event rule set is used to characterize the core events that drive changes in the agent's state.

[0145] Furthermore, the intelligent module 704 is also used for:

[0146] Based on the individual feature benchmark dataset, attribute values ​​are assigned to each intelligent agent population to obtain a list of basic intelligent agent objects;

[0147] Traverse the region code of each agent population in the list of basic agent objects, and associate the region features and the baseline probability of regional fertility intention of the corresponding region code in the region feature vector with the agent population to obtain the list of environmental connected agents.

[0148] Based on the list of connected agents in the environment, the probability of mutual connection between each agent population is calculated to obtain a gridded list of agents.

[0149] Based on the gridded list of agents, the overall sample is expanded to obtain the initial set of agent population.

[0150] Furthermore, the analog module 705 is also used for:

[0151] A simulated environment is constructed using the initial set of intelligent agent populations as the initial population data and the assumptions of future policy parameters as the external environment parameters.

[0152] Based on the simulation environment, the internal state variables of each agent population and the corresponding regional feature vector of the agent population are input into the agent behavior rule model for prediction to obtain the probability of individual fertility intention.

[0153] By comparing the probability of individual business willingness with the fertility threshold, a list of fertility decision results is obtained, and the initial intelligent agent population set is updated based on the list of fertility decision results to obtain the updated intelligent agent population set;

[0154] By integrating the list of fertility decision outcomes, a complete sequence of simulated events is obtained.

[0155] Furthermore, the updated agent population set is calculated using the following formula:

[0156]

[0157] in, Let be the new state vector of agent i at time period t+1. Let be the state vector of agent i during time period t. Let i be the result of agent i's reproductive decision in time period t. Let t be the external environmental parameters acting on the agent over a time period t, and R be the state transition function.

[0158] Furthermore, the model building module is also used for:

[0159] Based on the list of regions, a regional relationship graph is constructed, with regions as graph nodes and inter-regional relationship data as weighted edges.

[0160] Based on the region list, the region features corresponding to each graph node are extracted from the original region feature data to obtain the node feature matrix;

[0161] Align the historical label data with the graph nodes to obtain the node label vector;

[0162] Based on the region relation graph and node feature matrix, the core layer type and network depth of the basic graph neural network are selected to obtain the basic graph neural network architecture.

[0163] Based on node label vectors, region relationship graphs, and node feature matrices, the basic graph neural network architecture is iteratively trained to obtain a graph neural network.

[0164] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the above-described intelligent prediction method for population evolution trends based on regional fertility intention modeling.

[0165] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0166] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0167] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligently predicting population evolution trends based on regional fertility intention modeling, characterized in that, The method includes: Acquire multi-source population evolution data and perform data cleaning on the multi-source population evolution data to obtain a baseline year fused dataset; the multi-source population evolution data includes population census micro-samples and regional annual statistical reports; Based on graph neural networks, regional feature aggregation is performed on each region in the baseline year fusion dataset to obtain regional feature vectors. The regional feature vector is input into the fertility intention quantification model to predict fertility intention and obtain the baseline probability of regional fertility intention. Based on the baseline probability of fertility intention in the region and the agent behavior rule model, multiple agent populations are initialized to obtain an initial agent population set. Based on the initial agent population set and future policy parameter assumptions, agent event simulation is performed to obtain a complete simulated event sequence table. The complete simulated event sequence table is then summarized and statistically analyzed to obtain a population prediction result table.

2. The method according to claim 1, characterized in that, The agent population is constructed using the following method: Based on the aforementioned census micro-samples, key fields are extracted to obtain an individual characteristic benchmark dataset; Based on the individual feature benchmark dataset, the key fields are mapped to the internal state variables of the agent population to obtain the agent attribute architecture specification document; Based on the fertility intention quantification model, the model dimensionality is reduced to obtain the fertility intention function. Then, based on social impact rules and policy response rules, the fertility intention function is fine-tuned to obtain the agent behavior rule model. By integrating the aforementioned agent attribute architecture specification document and the aforementioned agent behavior rule model, a basic agent population is obtained; Based on the basic agent population, a lifecycle event rule set is added to obtain the agent population. The lifecycle event rule set is used to characterize the core events that drive changes in the agent's state.

3. The method according to claim 2, characterized in that, The method, based on the baseline probability of fertility intention in the region and the agent behavior rule model, initializes multiple agent populations to obtain an initial agent population set, including: Based on the individual feature benchmark dataset, attribute values ​​are assigned to each of the intelligent agents to obtain a list of basic intelligent agent objects. Traverse the region code of each agent population in the list of basic agent objects, and associate the region feature corresponding to the region code in the region feature vector with the region fertility intention baseline probability to the agent population to obtain the list of environment-connected agents; Based on the environmental connected agent list, the probability of mutual connection between each agent population is calculated to obtain a gridded agent list. Based on the gridded list of agents, a general expansion is performed to obtain the initial agent population set.

4. The method according to claim 3, characterized in that, Based on the initial agent population set and future policy parameter assumptions, agent event simulation is performed to obtain a complete simulated event sequence list, including: Using the initial set of intelligent agent populations as initial population data and the assumptions about future policy parameters as external environment parameters, a simulation environment is constructed. Based on the simulated environment, the internal state variables of each agent population and the corresponding regional feature vector of the agent population are input into the agent behavior rule model for prediction to obtain the probability of individual fertility intention. By comparing the individual business willingness probability with the fertility threshold, a fertility decision result list is obtained, and based on the fertility decision result list, the initial agent population set is updated to obtain the updated agent population set; By integrating the list of fertility decision outcomes, the complete simulated event sequence list is obtained.

5. The method according to claim 4, characterized in that, The updated agent population set is calculated using the following formula: in, Let be the new state vector of agent i at time period t+1. Let be the state vector of agent i during time period t. Let i be the result of agent i's reproductive decision in time period t. Let t be the external environmental parameters acting on the agent over a time period t, and R be the state transition function.

6. The method according to claim 1, characterized in that, The graph neural network is obtained through the following method: Based on the list of regions, a regional relationship graph is constructed, with the regions as graph nodes and the inter-regional association data between the regions as weighted edges. Based on the region list, the region features corresponding to each graph node are extracted from the original region feature data to obtain a node feature matrix; Align the historical label data with the graph nodes to obtain the node label vector; Based on the region relationship graph and the node feature matrix, the core layer type and network depth of the basic graph neural network are selected to obtain the basic graph neural network architecture. Based on the node label vector, the region relationship graph, and the node feature matrix, the basic graph neural network architecture is iteratively trained to obtain the graph neural network.

7. A population evolution trend intelligent prediction system based on regional fertility intention modeling, characterized in that, The system includes: The cleaning module is used to acquire multi-source population evolution data and clean the multi-source population evolution data to obtain a baseline year fused dataset; the multi-source population evolution data includes population census micro-samples and regional annual statistical reports; The feature module is used to aggregate regional features of each region in the baseline year fusion dataset based on a graph neural network to obtain regional feature vectors. The prediction module is used to input the regional feature vector into the fertility intention quantification model to predict fertility intention and obtain the baseline probability of regional fertility intention. The intelligent module is used to initialize multiple agent populations based on the baseline probability of fertility intentions in the region and the agent behavior rule model to obtain an initial agent population set; The simulation module is used to simulate agent events based on the initial agent population set and future policy parameter assumptions, obtain a complete simulated event sequence table, and summarize and statistically analyze the complete simulated event sequence table to obtain a population prediction result table.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images