Method and system for predicting regional population migration volume

Abstract

The invention discloses a regional population migration volume prediction method and system, and the method comprises the steps: obtaining the user condition of an intelligent terminal in a research region, judging migrated population and non-migrated population, and classifying the migrated population and the non-migrated population according to population portraits; constructing a bidirectional migration network of all population and various types of population in each time period; obtaining independent features and correlation features of the region, and constructing respective driving factor systems of population immigration and population emigration; setting constraint conditions of a population migration network, constructing a series of alternative simulation models for preliminarily simulating bidirectional migration volumes of all populations and various types of populations based on the driving factors, training and testing the alternative simulation models, and selecting an optimal simulation model; and adopting the optimal simulation model to predict the migration population in the preset planning period, and performing correction based on the constraint condition. According to the method, the population migration global features and the sub-population migration local features are integrated, migration network constraint conditions are added, and the population migration amount can be predicted more finely and accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of population statistics, and in particular relates to a method and system for predicting regional population migration. Background Art

[0002] Compared to natural population growth, population migration is having a more dramatic impact on regional population size and structure, becoming one of the core drivers of regional economic and social development. Existing population migration forecasts primarily rely on traditional historical statistical data, such as censuses and population mobility surveys, with less use of mobile phone signaling data. Technically, these methods primarily employ time-series recursion and push-pull models to predict population mobility. Most of these methods focus solely on net migration between two locations, insufficiently considering bidirectional migration and its driving mechanisms. Furthermore, focusing solely on the overall population movement between two locations, they overlook the differentiated migration mechanisms of different types of migrants. This makes it difficult to accurately predict regional population distribution flows and their evolving patterns. In reality, population migration involves two dynamic processes, in-migration and out-migration, which are integrated to form a net migration representation. This process offsets the amount of bidirectional migration. Furthermore, the driving factors of in-migration and out-migration vary significantly, which net migration ignores. Furthermore, the factors influencing in-migration and out-migration vary significantly across different population groups. For example, young people are more likely to consider the impact of work and income, while older households are more likely to consider factors such as living environment and service facilities. Therefore, it is difficult to accurately predict the amount of population inflow and outflow through net population migration forecasts or total migration population forecasts, and there will be large deviations. In addition, existing methods are mostly based on the simulation of population migration between two regional units, lacking the multi-unit-to-multi-unit population migration network constraints and calibration from the perspective of the migration network. For example, the simulated number of people migrating into a unit from all units must be as close as possible to the total number of actual inflows to the unit, and the simulated number of people migrating out of the unit to other units must be as close as possible to the total number of actual outflows from the unit. Finally, existing technologies rarely involve the impact of the population attributes of the inflow and outflow areas (size, age structure, education level, employment and unemployment scale, etc.) on migration, which needs further in-depth exploration. Summary of the Invention

[0003] One purpose of the present invention is to address the shortcomings of the existing technology and provide a method for predicting regional population migration. The method is based on a two-way migration network of in-migration and out-migration, and integrates the global characteristics of population migration and the local characteristics of sub-group migration. It can predict the population migration amount more finely and accurately, and can also predict the changes in population structure after population migration.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A method for predicting regional population migration volume includes the following steps:

[0006] Step 1: Obtain the situation of smart terminal users in the study area, determine the migrant population and non-migrant population, and further classify the migrant population or non-migrant population according to user portrait attributes;

[0007] Step 2: Based on the obtained user information, a migration vector is constructed for each user. Then, a bidirectional migration network is constructed for the entire population and each type of population in each time period, and the in-migration and out-migration volumes are counted.

[0008] Step 3: Obtain the independent characteristics and inter-related characteristics of each unit in the study area, and construct the driving factor system of population in-migration and out-migration based on them;

[0009] Step 4: Set the constraints of the population migration network and, based on the above driving factors, construct a series of alternative simulation models for preliminary simulation of bidirectional population migration, train and test them, and select the optimal simulation model;

[0010] Step 5: Use the optimal simulation model to predict the migrant population within the preset planning period, set constraints, revise the predicted value of the migrant population, and summarize the revised predicted population in-migration and out-migration situation.

[0011] Furthermore, obtaining user conditions includes: based on the smart terminal positioning data, obtaining the user's spatial position at the beginning and end of a period within multiple time periods with the same time interval, dividing the study area into different spatial units, and setting spatial unit attributes, wherein the spatial unit includes multiple types of spatial scales.

[0012] Furthermore, the method for determining whether the user is a migrant population in step 1 is as follows: within multiple time periods with the same time interval, if the Euclidean distance between the user's initial and final spatial positions meets the distance threshold and the spatial unit to which the user belongs changes, the user is determined to be a migrant population across spatial units; otherwise, the user is determined to be a non-migrant population.

[0013] Furthermore, the specific implementation method of step 2 includes:

[0014] Constructing user migration vectors based on the spatial positions of the migrating users at the beginning and end of each time period;

[0015] For all pairs of spatial units A and B, the migration volume V of all population from A→B and B→A in each period is summarized separately. ij and the migration of various groups Where k is the number of each type of group, i and j are the numbers of spatial units;

[0016] Using the bidirectional migration flows of two spatial units as edges and spatial units as nodes, a bidirectional migration network of all populations and all types of groups in each period is constructed;

[0017] Count the total inflow of all migrant population in each spatial unit and each time period Migration and the influx of various groups Migration k is the number of each type of group, and i is the number of the spatial unit.

[0018] Furthermore, the independent and interrelated features of the population obtained in step 3 include:

[0019] Obtain the independent demographic characteristics P of the population in-migration and out-migration spatial units during each period, including the total population, the population base of each age group, the total number of employed persons, the number of newly employed persons in urban areas, the number of re-employed persons in urban areas, and the unemployment rate;

[0020] Obtain the independent ecological and environmental characteristics E of the spatial units of population inflow and outflow during each period, including all independent ecological and environmental characteristics of the air quality index, water resource quality index, green space and ecological space index, biodiversity index, carbon emission and energy structure index;

[0021] Obtain the independent characteristics L of the living environment of the spatial unit of population in-migration and out-migration during each period, including all independent characteristics of the living environment such as infrastructure index, living conditions index, public service index, public safety index, and social security index;

[0022] Obtain the independent characteristics M of the production environment of the spatial unit of population inflow and outflow during each period, including all independent characteristics of the production environment, such as the number of employed people, the number of unemployed people, the economic base index, the employment and entrepreneurship index, the industrial supporting index, the business environment index, and the scientific and technological innovation index;

[0023] Obtain the spatial connection correlation characteristics R between the population in-migration and out-migration spatial units during each time period, including all spatial connection correlation characteristics of geographical proximity, location, and transportation connection, as well as the population migration connection characteristics within the preset time period in step 2.

[0024] Furthermore, the method of constructing the simulation model in step 4 includes:

[0025] The total population and the migration of various groups of people in all time periods or between two spatial units in a certain time period are used as explained variables, and the driving factors of population migration in and out of spatial units and related units are used as explanatory variables. Mathematical modeling methods are used to construct a series of alternative simulation models, among which mathematical modeling methods include regression analysis, machine learning, deep learning, neural networks, and graph neural networks.

[0026] When constructing the above-mentioned alternative simulation models, according to the simulation requirements and accuracy, the number of samples, and the computing power of the computer, set the population migration network constraints and conduct constrained modeling, or first conduct modeling and then use the constraints to modify the established simulation model;

[0027] After the simulation model is established, the performance of each alternative simulation model is compared through evaluation indicators, and the simulation model with the best performance is selected as the final bidirectional migration simulation model.

[0028] Furthermore, the population migration network constraints mainly include:

[0029] ① The sum of the simulated inflow of population from other units to any spatial unit is equal to the actual inflow of population to the unit.

[0030] ② The sum of all simulated population outflows from any spatial unit to other units is equal to the actual population outflow from that unit.

[0031] ③ The sum of the simulated influx of any type of population from other units into any spatial unit is equal to the actual influx of that specific type of small population into that unit.

[0032] ④ The sum of the simulated outflow of any type of population from any spatial unit to other units is equal to the actual outflow of that specific type of small population in that unit

[0033] ⑤ The sum of the simulated migration of each type of population between any two spatial units is equal to the actual in-migration of the entire population between those two units;

[0034] ⑥ For the entire population, the difference between the simulated in-migration and out-migration of each unit between any two spatial units is equal to the actual net migration;

[0035] ⑦ For each type of population, the difference between the simulated in-migration and out-migration of each unit between two spatial units is equal to the actual net migration;

[0036] ⑧For the entire population and all types of groups, the simulated migration amount is greater than or equal to 0;

[0037] Select one or more of the above constraints to perform constrained modeling or post-modeling correction.

[0038] Furthermore, the implementation of step 5 includes:

[0039] During the preset planning period, the predicted values of driving factors are set according to the future development conditions of each spatial unit;

[0040] The predicted values of driving factors are input into the above simulation models of the two-way migration of the entire population and all types of groups, and the predicted values of the two-way migration of the entire population between two spatial units i and j are generated. ij , and the predicted value S of the population bidirectional migration of each type of group k between i and j ij,k ;

[0041] With the constraints that the total number of migrants between two units is equal to the sum of the number of migrants of each type of group and that the amount of each migration is greater than or equal to 0, a mathematical method is used to jointly correct the predicted values of the total number of migrants and the number of migrants of each type of group to obtain the final predicted value T′. ij , S′ ij,k ;

[0042] The predicted inflow of all population in each spatial unit after statistical correction is summarized and calculated M′ in , outflow amount M' out , net inflow M′ net , synchronously output the immigration volume of each type of group in each spatial unit Migration and the difference between the two as net inflow

[0043] Furthermore, the mathematical method used for correction includes one of a mathematical programming method, a proportional distribution method, a weighted least squares adjustment, and a Bayesian adjustment.

[0044] Another object of the present invention is to provide a system for implementing the above-mentioned method for predicting regional population migration, comprising:

[0045] The data acquisition and processing module is used to obtain the situation of smart terminal users in the study area, determine the migrant population and the non-migrant population, and further classify the migrant population or the non-migrant population according to the user portrait attributes;

[0046] The bidirectional migration network construction module is used to construct the migration vector of each user based on the acquired user information, and then construct the bidirectional migration network of the entire population and various types of groups in each time period, and count the in-migration and out-migration volumes;

[0047] The driving factor measurement module is used to obtain the independent characteristics and inter-related characteristics of each unit in the study area, and accordingly construct the driving factor system of population in-migration and out-migration;

[0048] The simulation model construction module is used to set the constraints of the population migration network and, based on the above-mentioned driving factors, to construct a series of alternative simulation models for preliminary simulation of the two-way population migration volume, train and test them, and select the optimal simulation model;

[0049] The prediction module is used to use the optimal simulation model to predict the migrant population within the preset planning period, set constraints, revise the predicted value of the migrant population, and summarize the revised predicted population inflow and outflow.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention meticulously deconstructs population migration into two dynamic processes, population in-migration and population out-migration, explores the core driving factors of population in-migration and population out-migration respectively, and supplements the population attribute factors of the in-migration and out-migration places; and the present invention constructs simulation systems for the in-migration network and the out-migration network of the entire population and a specific type of population respectively, and performs cross-constraints and corrections among the four, thereby increasing the degree of conformity between the simulation and reality in terms of network traffic consistency and population structure consistency; in addition, based on the perspective of the migration network, the present invention adds multi-unit to multi-unit population migration network constraints in the above modeling process, thereby improving the global accuracy, including making the number of simulated populations migrating into a certain unit as equal as possible to the actual total number of populations migrating into the unit, and the number of simulated populations migrating out of the unit to other units must be as equal as possible to the actual total number of populations migrating out of the unit, etc.

[0052] The present invention integrates the global characteristics of population migration and the local characteristics of sub-group migration to construct a simulation system for the in-migration network and the out-migration network of the entire population and specific types of population. By setting many-to-many constraints in the simulation system of the two-way migration network of the population, it can achieve more detailed local and more accurate global prediction of regional population migration volume, and also provides technical support for predicting regional population size and structure, evaluating population introduction policies, increasing or decreasing various resources that match population development trends, and guiding the layout of industries and real estate markets. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method for predicting regional population migration volume disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0056] The present invention will be further described below with reference to specific examples, but they are not intended to limit the present invention.

[0057] like Figure 1 As shown, the embodiment of the present invention discloses a method for predicting regional population migration, comprising the following steps:

[0058] Step 1: Obtain the situation of smart terminal users in the study area, determine the migrant population and non-migrant population, and further classify the migrant population or non-migrant population according to user portrait attributes;

[0059] In this embodiment, mobile phones are selected as smart terminals. Based on mobile phone positioning big data, the spatial locations and spatial unit attributes of users within a preset study area, including their residence and employment locations, at the beginning and end of multiple time periods with equal time intervals are obtained to identify migrant populations. In this embodiment, the residential locations of users in a certain province at the beginning and end of each period are obtained for 2022, 2023, and 2024.

[0060] The study area is divided into different spatial units, and spatial unit attributes are set. The spatial units include multiple spatial scales, such as national, provincial, urban agglomeration, metropolitan area, city, county, district, township, street, and regular network. In this example, the in- and out-migration behavior of a province is used as research data. The data is divided into five equal, non-overlapping spatial units: a, b, c, d, and e, based on the administrative scope of prefecture-level cities.

[0061] Within multiple time periods with the same time interval, if the Euclidean distance between the user's initial and final spatial locations meets the distance threshold and the spatial unit to which they belong changes, they are determined to be cross-spatial unit migrants; otherwise, they are non-migrants. In this embodiment, within the time periods of 2022-2023 and 2023-2024, if the Euclidean distance between the user's initial and final spatial locations meets the distance threshold (in this embodiment, the threshold is greater than 1000 meters) and the zoning attributes change, they are determined to be cross-spatial unit migrants. Based on the change in the user's spatial location within the spatial unit, they are further classified as in-migrants and out-migrants.

[0062] Based on operator-provided demographic profiles such as user age, gender, employment type, and educational background, as well as identification of urban and rural populations based on permanent residence, the entire migrant population and the non-migrant population are divided into several non-overlapping subgroups whose sum totals equal the total migrant population or the non-migrant population. In this example, age is used as the basis for division: within the total migrant population, migrant users under 35 are classified as the young migrant population, those aged 36-59 are classified as the middle-aged migrant population, and those aged 60 and above are classified as the elderly migrant population. These three subgroups are non-overlapping and their sum totals equal the total migrant population.

[0063] Step 2: Based on the obtained user information, a migration vector is constructed for each user. Then, a bidirectional migration network is constructed for the entire population and specific small groups in each time period, and the in-migration and out-migration volumes are counted.

[0064] In this step, based on the spatial positions of the migrating users at the beginning and end of each period, the user migration vectors are constructed. For all pairs of spatial units A and B, the migration volume V of all population from A→B and B→A in each period is summarized. ij and the migration of specific types of small groups Where k is the number of each type of small group, i and j are the numbers of the spatial unit. In this example, the young migrant population is marked as y, the middle-aged migrant population is marked as m, and the elderly migrant population is marked as o. For spatial unit a, the total influx of population V in 2023 ba 、V ca 、V da 、V ea They are 120,000, 170,000, 130,000 and 150,000 respectively, and all the people who moved out are V ab 、V ac 、V ad 、V ae 150,000, 25,000, 110,000 and 130,000 respectively; young migrant population The number of young people migrating out of the city is 31,000, 32,000, 31,000 and 34,000 respectively. 27,000, 26,000, 25,500 and 27,000 respectively; middle-aged migrant population The numbers of middle-aged migrants are 62,000, 109,000, 71,000 and 88,000 respectively. 93,000, 60,000, 58,000 and 73,000 respectively; elderly migrant population The number of elderly people migrating out is 27,000, 29,000, 28,000 and 28,000 respectively. They are 30,000, 29,000, 26,500 and 30,000 respectively;

[0065] In this embodiment, we use this as an edge and spatial units as nodes to construct a bidirectional migration network of all populations and groups in each period. Specifically, we count the inflow of all migrant populations in each spatial unit in each period. Migration and the influx of various groups Migration k is the number of each type of group, i is the number of the spatial unit. In this embodiment, taking spatial unit a as an example, see Table 1, the total inflow of migrant population in 2023 570,000 people, outflow 505,000 people; see Table 2, the number of young people migrating 100,000 people, outflow 87,000 people; see Table 3, the number of middle-aged people migrating 330,000 people, outflow 284,000 people; see Table 4, the number of elderly people migrating in 140,000 people, outflow It is 134,000 people.

[0066] Table 1 Summary of all population migration numbers (2023)

[0067]

[0068] Table 2 Summary of Youth Migration (2023)

[0069]

[0070] Table 3 Summary of Migrants in Middle Ages (2023)

[0071]

[0072] Table 4 Summary of Migration of Elderly Population (2023)

[0073]

[0074] Step 3: Obtain the independent characteristics and inter-related characteristics of each unit in the study area, and construct the driving factor system of population in-migration and out-migration based on them;

[0075] In this step, obtaining independent features and interrelated features of regional units includes:

[0076] Obtain the independent population attribute characteristics P of the population in-migration and out-migration spatial units at the end of each period; in this embodiment, the independent population attribute characteristics P include all independent population attribute characteristics such as the population size index and the population employment index;

[0077] Obtain the independent ecological and environmental characteristics E of the spatial units of population inflow and outflow at the end of each period; in this embodiment, the independent ecological and environmental characteristics E include all independent ecological and environmental characteristics such as air quality index, water resource quality index, green space and ecological space index, biodiversity index, carbon emission and energy structure index;

[0078] Obtain the independent living environment characteristics L of the spatial units of population inflow and outflow at the end of each period; in this embodiment, the independent living environment characteristics L include all independent living environment characteristics such as infrastructure index, living conditions index, public service index, public safety index, and social security index;

[0079] Obtain the independent production environment characteristics M of the population in-migration and out-migration spatial units at the end of each period; in this embodiment, the independent production environment characteristics M include all independent production environment characteristics, such as the number of employed people, the number of unemployed people, the economic base index, the employment and entrepreneurship index, the industrial supporting index, the business environment index, and the scientific and technological innovation index;

[0080] Obtain the spatial connection correlation characteristics R between the spatial units of population in-migration and out-migration at the end of each period; in this embodiment, the spatial connection correlation characteristics include all spatial connection correlation characteristics such as geographical proximity, location, and transportation connection, as well as the population migration connection characteristics from 2022 to 2024;

[0081] The feature system and sources for this embodiment are shown in Table 5 below. There are 47 independent features, and five inter-city correlation features, consisting of spatial connections and historical migration data (historical in- and out-migration vector data, as time series features, reflects the historical migration intensity as part of the input features, capturing the inertia or periodicity of migration behavior). Each sample is the data for "city pair A→B" at the end of a particular year, and the input dimension is a (47 × 2 + 5) = 95-dimensional vector.

[0082] Table 5 Characteristic system of influencing factors

[0083]

[0084]

[0085] Data preprocessing such as data cleaning, missing value processing, standardization and normalization are performed on various indicators. In this embodiment, the missing values of the data are filled by filling the mean / median of features with a low data missing rate (<10%) and a distribution close to normal (such as GDP and per capita income), and interpolating the time series data using linear interpolation or time window mean (such as using the mean of the previous and next three years to fill the missing unemployment rate in a certain year). Outliers are detected and corrected using the 3σ principle, quantile pruning, business logic correction, etc.; and the feature data are standardized and normalized to complete the data preprocessing.

[0086] Step 4: Set the constraints of the population migration network. Based on the above driving factors, build a series of alternative simulation models for preliminary simulation of bidirectional population migration, train and test them, and select the optimal simulation model.

[0087] Using the total population and the migration of various groups of people between any two spatial units over all time periods or within a certain time period as the explained variables, and the driving factors of population in-migration and out-migration in spatial units and related units as the explanatory variables, a series of mathematical modeling techniques such as regression analysis, machine learning, deep learning, neural networks, and graph neural networks were used to train and construct a series of alternative models that can preliminarily simulate the bidirectional migration of the total population and various groups of people between all spatial units.

[0088] In this embodiment, the population inflow between the five spatial units a, b, c, d, and e in 2023 and 2024 is used as the explained variable, and the driving factors of population inflow and outflow into and out of the spatial units and related units are used as explanatory variables. Neural network and multivariate linear regression models are used for modeling. In other embodiments, machine learning, deep learning, and other methods can also be used for modeling:

[0089] For the neural network model, taking the training process of a fully connected network as an example, the structure is as follows: input layer (95 dimensions) → Dropout layer (to prevent overfitting) → fully connected layer (128 nodes, ReLU) → fully connected layer (64 nodes, ReLU) → output layer (1 node, linear activation); the loss function uses mean square error (MSE) for regression tasks and cross entropy for classification tasks; the optimizer prefers Adam (adaptive learning rate), and the initial learning rate is set to 1e-3. Parameters are adjusted from three key points: network structure optimization, training process parameter adjustment, and feature and loss optimization: the number of layers and nodes is gradually increased from the shallowest layer (2 layers), and the number of nodes increases from 64 to 128 to 256. The optimal complexity is selected through the validation set loss; the regularization Dropout ratio (0.2-0.5) is combined with the L2 regularization coefficient (1e-4 to 1e-2) to suppress overfitting; the learning rate is initially set to 1e-3 and is reduced to 1e-4 if the loss fluctuates; the batch size is commonly 32 or 64, and can be increased to 128 when the data volume is large; an early stopping mechanism is set to monitor the validation set loss, and training is terminated if there is no improvement for 5-10 consecutive rounds; feature engineering introduces lagged variables (migration amount in the previous year) and interaction terms (such as unemployment rate × target city salary) to enhance explanatory power; for scenarios sensitive to extreme values, the Huber loss is used instead of the MSE to improve the loss function.

[0090] For the multiple linear regression model, the construction example is as follows: data cleaning, processing missing values and standardizing continuous variables, then screening significant variables through correlation analysis, VIF test and stepwise regression to eliminate multicollinearity interference. When testing the model, it is necessary to verify the normality and homoscedasticity of the residuals, and use the adjusted R 2 The explanatory power and overall significance of the model were evaluated using the p-value (e.g., 0.68) and F-test (P<0.001). The constructed multiple linear regression equation is as follows:

[0091]

[0092] Among them, V ij is the total population inflow from city i to city j during the period or the inflow of a specific type of small group of people; is the independent characteristic driving factor of the migration location i at the end of the period; is the independent characteristic driving factor of the immigration location j at the end of the period; D ij is the driving factor of the correlation characteristics between the two places;∈ ij is the error term.

[0093] When constructing the above simulation model, it is necessary to set the population migration network constraints of the simulation model. According to the simulation requirements and accuracy, sample size, and computer computing power, select 0-7 of the following population migration network constraints to perform constrained modeling or post-modeling correction: ① The sum of all simulated population inflows from other units in any spatial unit is equal to the actual population inflow of the unit.

[0094] ② The sum of all simulated population outflows from any spatial unit to other units is equal to the actual population outflow from that unit.

[0095] ③ The sum of the simulated influx of any type of population from other units into any spatial unit is equal to the actual influx of that specific type of small population into that unit.

[0096] ④ The sum of the simulated outflow of any type of population from any spatial unit to other units is equal to the actual outflow of that specific type of small population in that unit

[0097] ⑤ The sum of the simulated migration of each type of population between any two spatial units is equal to the actual in-migration of the entire population between those two units;

[0098] ⑥ For the entire population, the difference between the simulated in-migration and out-migration of each unit between any two spatial units is equal to the actual net migration;

[0099] ⑦ For each type of population, the difference between the simulated in-migration and out-migration of each unit between two spatial units is equal to the actual net migration;

[0100] ⑧For the entire population and all types of groups, the simulated migration amount is greater than or equal to 0.

[0101] This embodiment adopts constrained modeling, selecting three constraints ①, ②, and ⑧. For the neural network model, the constraints are added to the loss function; for the multivariate linear regression model, the model is first built, and then the data planning method is used to adopt the three constraints ①, ②, and ⑧ to correct the model with the goal of minimizing the correction amplitude.

[0102] After the model is established, a multi-stage, multi-dimensional systematic approach is used to comprehensively evaluate the pros and cons of each model. Specifically, R 2 , mean square error, AIC, BIC and other evaluation indicators are used to compare the performance of each alternative simulation model, and the simulation model with the best performance is selected as the final simulation model for predicting the two-way migration of the entire population and each specific type of population.

[0103] In this embodiment, the first stage is based on the prediction accuracy indicators (MSE, R 2 ) Quickly eliminate models with poor performance; in the second stage, further analyze the remaining models, calculate AIC / BIC for traditional statistical models (linear regression, gravity model) to quantify the simplicity of the model, and cross-validate all models, and calculate the standard deviation of the indicators to evaluate stability; in the third stage, the selected candidate models will be screened out for final generalization performance verification on an independent test set. Differentiated treatment is performed for special models: due to the large parameter scale and the difficulty in defining the likelihood function of the deep learning model, the AIC / BIC indicator is abandoned, and overfitting is controlled by early stopping and regularization technology, and the number of parameters and training time are additionally recorded as a reference for complexity; for classic statistical models (such as the gravity model), it is necessary to give priority to whether the parameter signs are consistent with theoretical expectations (economic variables positively drive migration, and geographical distance has a negative impact). Even if the prediction accuracy is slightly lower, it may be retained due to the advantage of interpretability. The final model selection is based on MSE and R 2 The core basis is the test set performance; in terms of accuracy-complexity trade-off, if the accuracy difference between models is less than 5%, models with simpler structures and fewer parameters are preferred; in policy analysis scenarios, appropriate accuracy loss can be accepted in exchange for interpretability (such as choosing decision trees instead of neural networks) to increase business logic adaptability. In addition, the robustness of the model is verified through sensitivity analysis, such as adding noise to key features or introducing external data sets to ensure that the model can maintain stable prediction capabilities in data perturbations and cross-regional scenarios. In this embodiment, the MSE of the multivariate linear regression model performed poorly, and the neural network model was finally selected for prediction. The code example using python as an implementation tool is as follows:

[0104] #Model evaluation example (taking random forest and linear regression as examples)

[0105] fromsklearn.metrics import mean_squared_error,r2_score

[0106] fromsklearn.model_selection import cross_val_score

[0107] #Calculate the test set MSE and R 2

[0108] y_pred_rf=rf_model.predict(X_test)

[0109] mse_rf=mean_squared_error(y_test,y_pred_rf)

[0110] r2_rf=r2_score(y_test,y_pred_rf)

[0111] #Cross validation stability evaluation

[0112] cv_scores=cross_val_score(rf_model,X_train,y_train,cv=5,

[0113] scoring='neg_mean_squared_error')

[0114] cv_std = np.std(-cv_scores)

[0115] #Statistical model AIC / BIC (taking statsmodels as an example)

[0116] import statsmodels.api as sm

[0117] ols_model=sm.OLS(y_train,X_train).fit()

[0118] print(f"AIC:{ols_model.aic},BIC:{ols_model.bic}")

[0119] Step 5: Use the optimal simulation model to predict the migrant population within the preset planning period, set constraints, jointly revise the predicted values of the migrant population, and summarize the predicted population in- and out-migration.

[0120] During the preset planning period, predicted values for driving factors are set based on the future development conditions of each spatial unit. In this embodiment, a time series recursion method is used to predict the total population, population of each age group, total employed persons, number of re-employed urban unemployed persons, GDP, per capita disposable income, and the proportion of added value of the tertiary industry for each city one year from now. Based on the recent planning of each city, public transportation coverage, road network density, number of medical beds per thousand people, and number of schools per ten thousand people are predicted. Other driving factors remain unchanged.

[0121] The predicted values of driving factors are input into the simulation model of all the selected migrant populations to generate the predicted value T of the bidirectional migration of all the population between two spatial units i and j. ij , and the predicted value S of the population bidirectional migration of each small group k between i and j ij,k .

[0122] With the constraints that the total number of migrants between any two units is equal to the sum of the number of migrants in each small group and that the amount of each migrant is greater than or equal to 0, the predicted values of the total number of migrants and the number of migrants in a specific type of small group are jointly revised through mathematical programming, proportional allocation, weighted least squares adjustment, Bayesian adjustment and other methods to obtain the final predicted value T′. ij , S′ ij,k .

[0123] In this embodiment, taking the mathematical programming method as an example, an objective function related to the variance of the simulation model can be selected. For each migration amount ij, the objective function is:

[0124]

[0125] Among them, T′ ij 、T ij is the predicted migration amount of the entire population between the ij units before and after the correction; S′ ij,k 、S ij,k is the population migration prediction of small group k between ij units before and after correction; Var T is the predicted variance of the simulation model for the migration of the entire population; is the predicted variance of the migration simulation model for small group k; K is the total number of small group types. This objective function indicates that the larger the variance of the model, the more uncertain the prediction, and the larger the allowable correction range.

[0126] The constraints are:

[0127]

[0128] T′ ij ≥0

[0129]

[0130] In this example, the revised predicted migration population between each spatial unit is as follows, see Table 6-9:

[0131] Table 6 Summary of all population migration numbers (end of planning period)

[0132]

[0133] Table 7 Summary of Youth Migration (End of Planning Period)

[0134]

[0135] Table 8 Summary of Migrants in Middle Ages (End of Planning Period)

[0136]

[0137] Table 9 Summary of Migration of Elderly Population (End of Planning Period)

[0138]

[0139] Finally, the predicted inflow of all population in each spatial unit is summarized and counted as M′ in , outflow amount M' out , net inflow M′ net . Synchronously output the influx of specific small groups of each spatial unit Migration and the difference between the two as net inflow In this embodiment, the prediction results summary table is shown in Table 10, taking the net in-migration population of the five spatial units a, b, c, d, and e at the end of the planning period as an example.

[0140] Table 10 Summary of Net In-migration Population (End of Planning Period)

[0141]

[0142] An embodiment of the present invention further provides a system for implementing the above-mentioned method for predicting regional population migration, comprising:

[0143] The data acquisition and processing module is used to obtain the situation of smart terminal users in the study area, determine the migrant population and the non-migrant population, and further classify the migrant population or the non-migrant population according to the user portrait attributes;

[0144] The bidirectional migration network construction module is used to construct the migration vector of each user based on the acquired user information, and then construct the bidirectional migration network of the entire population and various types of groups in each time period, and count the in-migration and out-migration volumes;

[0145] The driving factor measurement module is used to obtain the independent characteristics and inter-related characteristics of each unit in the study area, and accordingly construct the driving factor system of population in-migration and out-migration;

[0146] The simulation model construction module is used to set the constraints of the population migration network. Based on the above driving factors, a series of alternative simulation models for preliminary simulation of the two-way population migration volume are constructed, trained and tested, and the optimal simulation model is selected.

[0147] The prediction module is used to use the optimal simulation model to predict the migrant population within the preset planning period, set constraints, revise the predicted value of the migrant population, and summarize the revised predicted population inflow and outflow.

[0148] The above are only preferred embodiments of the present invention and do not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of the present invention specification should be included in the protection scope of the present invention.

Claims

1. A method for predicting regional population migration, characterized in that: The following steps are involved: Step 1: Obtain the situation of smart terminal users in the study area, determine the migrant population and non-migrant population, and further classify the migrant population or non-migrant population according to user portrait attributes; Step 2: Based on the obtained user information, a migration vector is constructed for each user. Then, a bidirectional migration network is constructed for the entire population and each type of population in each time period, and the in-migration and out-migration volumes are counted. Step 3: Obtain the independent characteristics and inter-related characteristics of each unit in the study area, and construct a system of driving factors for population in-migration and out-migration based on them; Step 4: Set the constraints of the population migration network and, based on the above driving factors, construct a series of alternative simulation models for preliminary simulation of bidirectional population migration, train and test them, and select the optimal simulation model; Step 5: Use the optimal simulation model to predict the migrant population within the preset planning period, set constraints, revise the predicted value of the migrant population, and summarize the revised predicted population in-migration and out-migration situation.

2. The method for predicting regional population migration according to claim 1, characterized in that: In step 1, obtaining user information includes: based on the positioning data of the smart terminal, obtaining the spatial position of the user at the beginning and end of the period within multiple time periods with the same time interval, dividing the study area into different spatial units, and setting spatial unit attributes. The spatial unit includes multiple types of spatial scales.

3. The method for predicting regional population migration according to claim 1, wherein: The method for determining whether the population is migrating in step 1 is as follows: within multiple time periods with the same time interval, if the Euclidean distance between the user's initial and final spatial positions meets the distance threshold and the spatial unit to which the user belongs changes, the user is determined to be a migrant population across spatial units; otherwise, the user is determined to be a non-migrant population.

4. The method for predicting regional population migration according to claim 1, wherein: The specific implementation method of step 2 includes: Constructing user migration vectors based on the spatial positions of the migrating users at the beginning and end of each time period; For all pairs of spatial units A and B, the migration volume V of all population from A→B and B→A in each period is summarized separately. ij and the migration of various groups Where k is the number of each type of group, i and j are the numbers of spatial units; Using the bidirectional migration flows of two spatial units as edges and spatial units as nodes, a bidirectional migration network of all populations and all types of groups in each period is constructed; Count the total inflow of all migrant population in each spatial unit and each time period Migration and the influx of various groups Migration k is the number of each type of group, and i is the number of the spatial unit.

5. The method for predicting regional population migration according to claim 1, characterized in that: The independent and interrelated characteristics of the population obtained in step 3 include: Obtain the independent demographic characteristics P of the population in-migration and out-migration spatial units during each period, including the total population, the population base of each age group, the total number of employed persons, the number of newly employed persons in urban areas, the number of re-employed persons in urban areas, and the unemployment rate; Obtain the independent ecological and environmental characteristics E of the spatial units of population inflow and outflow during each period, including all independent ecological and environmental characteristics of the air quality index, water resource quality index, green space and ecological space index, biodiversity index, carbon emission and energy structure index; Obtain the independent characteristics L of the living environment of the spatial unit of population in-migration and out-migration during each period, including all independent characteristics of the living environment such as infrastructure index, living conditions index, public service index, public safety index, and social security index; Obtain the independent characteristics M of the production environment of the spatial unit of population inflow and outflow during each period, including all independent characteristics of the production environment, such as the number of employed people, the number of unemployed people, the economic base index, the employment and entrepreneurship index, the industrial supporting index, the business environment index, and the scientific and technological innovation index; Obtain the spatial connection correlation characteristics R between the population in-migration and out-migration spatial units during each time period, including all spatial connection correlation characteristics of geographical proximity, location, and transportation connection, as well as the population migration connection characteristics within the preset time period in step 2.

6. The method for predicting regional population migration according to claim 2, characterized in that: The method for constructing the simulation model in step 4 includes: The total population and the migration of various groups of people in all time periods or between two spatial units in a certain time period are used as explained variables, and the driving factors of population migration in and out of spatial units and related units are used as explanatory variables. Mathematical modeling methods are used to construct a series of alternative simulation models, among which mathematical modeling methods include regression analysis, machine learning, deep learning, neural networks, and graph neural networks. When constructing the above-mentioned alternative simulation models, according to the simulation requirements and accuracy, the number of samples, and the computing power of the computer, set the population migration network constraints and conduct constrained modeling, or first conduct modeling and then use the constraints to modify the established simulation model; After the simulation model is established, the performance of each alternative simulation model is compared through evaluation indicators, and the simulation model with the best performance is selected as the final bidirectional migration simulation model.

7. The method for predicting regional population migration according to claim 6, characterized in that: The constraints on population migration networks include: ① The sum of the simulated inflow of population from other units to any spatial unit is equal to the actual inflow of population to the unit. ② The sum of all simulated population outflows from any spatial unit to other units is equal to the actual population outflow from that unit. ③ The sum of the simulated influx of any type of population from other units into any spatial unit is equal to the actual influx of that specific type of small population into that unit. ④ The sum of the simulated outflow of any type of population from any spatial unit to other units is equal to the actual outflow of that specific type of small population in that unit ⑤ The sum of the simulated migration of each type of population between any two spatial units is equal to the actual in-migration of the entire population between those two units; ⑥ For the entire population, the difference between the simulated in-migration and out-migration of each unit between any two spatial units is equal to the actual net migration; ⑦ For each type of population, the difference between the simulated in-migration and out-migration of each unit between two spatial units is equal to the actual net migration; ⑧For the entire population and all types of groups, the simulated migration amount is greater than or equal to 0; Select one or more of the above constraints to perform constrained modeling or post-modeling correction.

8. The method for predicting regional population migration according to claim 1, characterized in that: Step 5 implementation methods include: During the preset planning period, the predicted values of driving factors are set according to the future development conditions of each spatial unit; The predicted values of driving factors are input into the above simulation models of the two-way migration of the entire population and all types of groups, and the predicted values of the two-way migration of the entire population between two spatial units i and j are generated. ij , and the predicted value S of the population bidirectional migration of each type of group k between i and j ij,k ; With the constraints that the total number of migrants between two units is equal to the sum of the number of migrants of each type of group and that the amount of each migration is greater than or equal to 0, a mathematical method is used to jointly correct the predicted values of the total number of migrants and the number of migrants of each type of group to obtain the final predicted value T′. ij , S′ ij,k ; The predicted inflow of all population in each spatial unit after statistical correction is summarized and calculated M′ in , outflow amount M' out , net inflow M′ net , synchronously output the immigration volume of each type of group in each spatial unit Migration and the difference between the two as net inflow 9. The method for predicting regional population migration according to claim 8, characterized in that: The mathematical methods used for correction include one of mathematical programming method, proportional distribution method, weighted least squares adjustment and Bayesian adjustment.

10. A system for implementing the method for predicting regional population migration according to any one of claims 1 to 9, characterized in that: include: The data acquisition and processing module is used to obtain the situation of smart terminal users in the study area, determine the migrant population and the non-migrant population, and further classify the migrant population or the non-migrant population according to the user portrait attributes; The bidirectional migration network construction module is used to construct the migration vector of each user based on the acquired user information, and then construct the bidirectional migration network of the entire population and various types of groups in each time period, and count the in-migration and out-migration volumes; The driving factor measurement module is used to obtain the independent characteristics and inter-related characteristics of each unit in the study area, and accordingly construct the driving factor system of population in-migration and out-migration; The simulation model construction module is used to set the constraints of the population migration network and, based on the above-mentioned driving factors, to construct a series of alternative simulation models for preliminary simulation of the two-way population migration volume, train and test them, and select the optimal simulation model; The prediction module is used to use the optimal simulation model to predict the migrant population within the preset planning period, set constraints, revise the predicted value of the migrant population, and summarize the revised predicted population inflow and outflow.

Images