A method and system for predicting population dynamics in urban agglomerations using multi-spatial-scale planning
By constructing a multi-spatial-scale planning population dynamic prediction method for urban agglomerations, and utilizing multi-source data and dynamic model adjustments, the accuracy problem of large-scale cross-regional population prediction in urban agglomerations has been solved, achieving multi-level accurate population prediction and improved model adaptability.
Patent Information
- Application Number
- CN202611572381.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-12
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies are ill-suited to the large-scale, cross-regional, and multi-level planning needs of urban clusters, and population prediction models lack dynamic calibration and iterative update mechanisms, resulting in insufficient accuracy of prediction results.
A method for dynamic population prediction in urban agglomerations with multiple spatial scales is constructed. By collecting multi-source data, exponential smoothing model, moving average model, grey model and growth rate model are adopted, and the model parameters are dynamically adjusted in combination with the population proportion coefficient to achieve accurate population prediction at multiple spatial levels.
It enables accurate population prediction across multiple spatial levels within urban agglomerations, improves model adaptability and prediction accuracy, can adapt to dynamic data updates, and meets the needs of regional spatial planning and transportation model construction.
Smart Images

Figure CN123070769A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of smart city data analysis technology, specifically to a method and system for predicting population dynamics in urban agglomerations across multiple spatial scales. Background Technology
[0002] As the construction of urban agglomerations in my country continues to accelerate and regional integration progresses rapidly, regional spatial planning and the establishment of an integrated transportation system have become core tasks in territorial spatial planning and regional governance. Population size and spatial distribution characteristics are crucial supporting data in core tasks such as establishing urban agglomeration planning systems, constructing regional transportation models, and allocating territorial spatial resources.
[0003] Currently, the industry has conducted extensive research on population forecasting, with mainstream technical solutions primarily relying on traditional data statistical models and classical demographic methods for population calculation. However, existing conventional population forecasting technologies mostly focus on static calculations of the total population of a single city or independent area, and are limited to small-scale population data analysis of independent administrative units, making it difficult to meet the needs of large-scale, cross-regional, and multi-level planning and assessment of urban clusters.
[0004] Chinese patent application number 202111015024.0 discloses a method for predicting the refined spatial distribution of future population. By identifying land use types in the area to be predicted and constructing a spatial population model, it achieves the prediction of the future spatial distribution of the population in the region. In this patent, land use type is a crucial input data for the model. However, urban agglomerations cover a wide area, making it difficult to obtain basic data. Land use data varies significantly, making it difficult to obtain accurate basic land use data for the planning year, thus reducing the accuracy of the prediction results. Furthermore, this spatial population model is based on data from a single year and lacks a dynamic calibration and iterative update mechanism, posing a risk of gradually increasing model prediction errors.
[0005] To meet the demand for efficient and accurate population forecasting in regional spatial planning and governance, it is urgent to explore and construct a population dynamic forecasting technology solution that is adaptable to the entire urban agglomeration scenario, covers multiple spatial levels, and relies on multi-source data for dynamic updates. Summary of the Invention
[0006] Based on the above description, the present invention provides a method and system for dynamic population prediction in urban agglomeration planning at multiple spatial scales. This method can efficiently adapt to the entire urban agglomeration scenario, cover multiple spatial levels, and dynamically update population prediction based on multi-source data.
[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A method for predicting population dynamics in urban agglomerations across multiple spatial scales includes the following steps: S1. Collect the minimum spatial scale resident population data of each city in the urban agglomeration from the statistical yearbooks of the past n years, as well as the national population census data of the most recent year; take each city in the urban agglomeration as a major transportation region, take the district as a central transportation region, and take the streets and towns as minor transportation areas to form a spatial grid with full coverage across multiple spatial scales. S2. Divide the collected resident population data from the statistical yearbooks of the past n years into training set and test set. For each city's smallest spatial scale unit, use exponential smoothing model, moving average model, grey model, and growth rate model to fit the training set data, and use these models to predict the test set data. Select the model with the smallest root mean square error as the optimal model, thereby calibrating a series of differentiated population prediction models for each spatial scale unit in each city. S3. Based on the latest national census data, calculate the population proportion coefficient of each traffic zone in each city within its respective major and central traffic zones; input the planning year, run the population prediction series model calibrated for each city to obtain the predicted population data at the corresponding spatial scale; based on the corresponding population proportion coefficient, decompose the predicted population data at the scale of major or central traffic zones in each city to each traffic zone under its jurisdiction to obtain the predicted resident population data at the scale of all traffic zones in each city, thereby obtaining the predicted resident population data at multiple spatial scales within the urban agglomeration.
[0008] As a preferred option, the model recalibration step is also included: Obtain the actual resident population data for year n+1, calculate the average relative error between the predicted and actual population values for each city, and when the error is within a set range, use the ratio of the actual population value to the predicted population value to obtain an adjustment coefficient; by multiplying the adjustment coefficient by the predicted resident population for each year, dynamically adjust the resident population data for each year in the traffic zone cells contained in the spatial scale unit. When the error between the predicted value and the actual value exceeds the assumed range, the data from year n+1 is merged with the data from previous years, and the model is recalibrated.
[0009] As a preferred option, the method for selecting the optimal model in step S2 is as follows: The 11 consecutive years of resident population data collected from each city are divided into population data at three spatial scales: traffic sub-district, central traffic area, and major traffic area. Based on each city's own spatial scale, the first 8 years of historical resident population data for each spatial scale are used as the training set, and the last 3 years are used as the test set. The smoothing model, moving average model, grey model, and growth rate model are used to fit and adjust the parameters of the training data for each city. Each model independently calculates the predicted resident population value for the corresponding spatial scale in the last 3 years. The population prediction values output by each model are compared with the actual population in the yearbook at the same spatial scale. The root mean square error (RMSE) is used to evaluate the fitting accuracy, and the model with the smallest RMSE is selected as the optimal resident population prediction model.
[0010] As a preferred option, the calculation process for the population proportion coefficient in step S3 is as follows: Extract the actual resident population of the entire traffic central area from the census data; extract the actual resident population of each traffic sub-district within the traffic central area; calculate the population ratio of the traffic sub-district relative to the central area, as follows:
[0011] In the formula The population proportion coefficient of the traffic central area to which the i-th traffic zone belongs; The resident population of the i-th traffic area as measured in the population census; The total population of the Jiaotong Central District as determined by the census is the total population of this Jiaotong residential area. Similarly, calculate the ratio of traffic zones to traffic areas:
[0012] In the formula: is the population proportion coefficient of the transportation region to which the i-th transportation zone belongs; The resident population of the i-th traffic area as measured in the population census; This represents the total population of the transportation district to which this transportation community belongs in the census.
[0013] As a preferred option, the process of decomposing the predicted population at the scale of major and intermediate transportation zones into smaller transportation zones in step S3 is as follows: When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone:
[0014] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; This represents the projected total population of the central transportation zone to which this transportation community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic area; When the model outputs the predicted total population of a city's transportation region, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding transportation zone within the region:
[0015] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
[0016] A multi-spatial-scale planning population dynamics prediction system for urban agglomerations includes: The data acquisition unit is used to collect the minimum spatial scale resident population data of each city in the urban agglomeration from the statistical yearbooks of the past n years, as well as the national population census data of the most recent year. Spatial grid division units are used to divide each city in the urban agglomeration into a major transportation zone, a central transportation zone, and streets / townships into smaller transportation zones, based on the administrative divisions of each city within the urban agglomeration, thus forming a spatial grid that covers the entire area at multiple spatial scales. The model library includes exponential smoothing models, moving average models, grey models, and growth rate models. The model calibration unit is used to divide the resident population data of the statistical yearbooks of various cities over the past n years into training set and test set; the smoothing model, moving average model, grey model and growth rate model are used to predict the test set data respectively, and the model with the smallest root mean square error is selected as the optimal model, thereby calibrating a series of differentiated population prediction models for each spatial scale unit in each city. The calculation unit is used to calculate the population proportion coefficient of each traffic zone in each city within its respective traffic superzone and traffic central zone; and to calculate and decompose the predicted population data at the traffic superzone or traffic central zone scale of each city to the respective traffic zones based on the corresponding population proportion coefficient.
[0017] As a preferred embodiment, the model calibration unit divides the 11-year resident population data collected from each city into population data at three spatial scales: traffic sub-district, traffic central area, and traffic large area. Based on the spatial scale of each city, the first 8 years of historical resident population data for each spatial scale are used as the training set, and the last 3 years are used as the test set. The smoothing model, moving average model, grey model, and growth rate model are used to fit and adjust the parameters of the training data for each city. Each model independently calculates the predicted resident population value for the corresponding spatial scale in the last 3 years. The population prediction value output by each model is compared with the actual population of the yearbook at the same spatial scale. The root mean square error (RMSE) is used to evaluate the fitting accuracy, and the model with the smallest RMSE is selected as the optimal prediction model for resident population.
[0018] As a preferred embodiment, the model calibration unit also recalibrates the model. The process is as follows: obtain the actual resident population data for year n+1, calculate the average relative error between the predicted and actual population values for each city, and when the error is within a set range, obtain an adjustment coefficient by comparing the actual population value with the predicted population value; dynamically adjust the resident population data for each year in the traffic zone cells contained in the spatial scale unit by multiplying the adjustment coefficient with the predicted resident population for each year; when the error between the predicted and actual values exceeds the assumed range, merge the data for year n+1 with the data from previous years and recalibrate the model.
[0019] As a preferred embodiment, the process by which the calculation unit calculates the population proportion coefficient is as follows: Extract the actual resident population of the entire traffic central area from the census data; extract the actual resident population of each traffic sub-district within the traffic central area; calculate the population ratio of the traffic sub-district relative to the central area, as follows:
[0020] In the formula The population proportion coefficient of the traffic central area to which the i-th traffic zone belongs; The resident population of the i-th traffic area as measured in the population census; The total population of the Jiaotong Central District as determined by the census is the total population of this Jiaotong residential area. Similarly, calculate the ratio of traffic zones to traffic areas:
[0021] In the formula: is the population proportion coefficient of the transportation region to which the i-th transportation zone belongs; The resident population of the i-th traffic area as measured in the population census; This represents the total population of the transportation district to which this transportation community belongs in the census.
[0022] As a preferred embodiment, the process by which the calculation unit decomposes the predicted population at the scale of major and intermediate transportation zones into smaller transportation zones is as follows: When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone:
[0023] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; This represents the projected total population of the central transportation zone to which this transportation community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic area; When the model outputs the predicted total population of a city's transportation region, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding transportation zone within the region:
[0024] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
[0025] Compared with the prior art, the technical solution of this application has the following beneficial technical effects: 1. The model is built based on readily available statistical yearbook data. The data requirements and model methods are simple. It can predict the permanent population of cities, districts, streets / townships in different planning years at multiple spatial scales within the urban agglomeration. It meets the application needs of regional spatial planning and transportation model construction, and can better adapt to the whole-domain scenario and cover multiple spatial levels.
[0026] 2. Based on the spatial scale characteristics of the resident population data of each city, a differentiated model parameter system is constructed to effectively improve the model's adaptability and prediction accuracy.
[0027] 3. It can dynamically adjust the existing model parameter system and traffic zone proportion coefficient based on multi-source data to ensure that the model has good adaptability in the long term. Attached Figure Description
[0028] Figure 1 This is a flowchart of the method in Example 1; Figure 2 This is a schematic diagram of the traffic zone scale adjustment coefficient construction in Example 1. Detailed Implementation
[0029] Example 1: Reference Figure 1 A method for predicting population dynamics in urban agglomerations using multi-spatial-scale planning includes the following steps: S1. Collect the minimum spatial scale resident population data of each city in the urban agglomeration from the statistical yearbooks of the past n years, as well as the national population census data of the most recent year; take each city in the urban agglomeration as a major transportation region, take the district as a central transportation region, and take the streets and towns as minor transportation areas to form a spatial grid that covers the entire region at multiple spatial scales.
[0030] S2. Divide the collected resident population data from the statistical yearbooks of the past n years into training set and test set. For each city's smallest spatial scale unit, use time series models such as exponential smoothing model, moving average model, grey model, and growth rate model to fit the training set data, and apply these models to predict the test set data. Select the model with the smallest root mean square error as the optimal model, thereby calibrating a series of differentiated population prediction models for each spatial scale unit in each city.
[0031] S3. Based on the latest national census data, calculate the population proportion coefficient of each traffic zone in each city within its respective major and central traffic zones; input the planning year, run the population prediction series model calibrated for each city to obtain the predicted population data at the corresponding spatial scale; based on the corresponding population proportion coefficient, decompose the predicted population data at the scale of major or central traffic zones in each city to each traffic zone under its jurisdiction to obtain the predicted resident population data at the scale of all traffic zones in each city, and aggregate them upwards to obtain the resident population at the district and city scales, thus obtaining the predicted resident population data at multiple spatial scales within the urban agglomeration.
[0032] This embodiment also includes a model recalibration step: Obtain the actual resident population data for year n+1, calculate the average relative error between the predicted and actual population values for each city, and when the error is within a set range, use the ratio of the actual population value to the predicted population value to obtain an adjustment coefficient; by multiplying the adjustment coefficient by the predicted resident population for each year, dynamically adjust the resident population data for each year in the traffic zone cells contained in the spatial scale unit. When the error between the predicted value and the actual value exceeds the assumed range, the data from year n+1 is merged with the data from previous years, the model is recalibrated, the model algorithm and parameters are updated, and the population is predicted again.
[0033] Adjusted population forecast for the planning year: Input the planning year, calculate the predicted population at the corresponding spatial scale again using the dynamically adjusted model algorithm above, and obtain the adjusted resident population forecast data for the urban agglomeration at multiple spatial scales through proportion coefficient decomposition and upward aggregation, thereby realizing the rolling update of the resident population forecast model and results.
[0034] In this embodiment, the method for selecting the optimal model in step S2 is as follows: the 11 consecutive years of permanent resident population data collected from each city are divided into population data at three spatial scales: traffic sub-district, traffic central area, and traffic large area; based on the spatial scale of each city, the data of the first 8 years of the 11-year historical permanent resident population data at each spatial scale are used as the training set and the data of the last 3 years are used as the test set; the grey model, exponential smoothing model, moving average model, and growth rate model are used to fit and adjust the parameters of the training data of each city, and each model independently calculates the predicted value of the permanent resident population at the corresponding spatial scale for the last 3 years; the population prediction value output by each model is compared with the actual population of the yearbook at the same spatial scale, and the fitting accuracy is evaluated by the root mean square error (RMSE), and the model with the smallest RMSE is selected as the optimal prediction model for the permanent resident population.
[0035] In this embodiment, the calculation process of the population proportion coefficient in step S3 is as follows: Extract the actual resident population of the entire traffic central area from the census data; extract the actual resident population of each traffic sub-district within the central area; calculate the population ratio of the traffic sub-district relative to the central area, as follows:
[0036] In the formula The population proportion coefficient of the traffic central area to which the i-th traffic zone belongs; The resident population of the i-th traffic area as measured in the population census; The total population of the Jiaotong Central District as determined by the census is the total population of this Jiaotong residential area. Similarly, calculate the ratio of traffic zones to traffic areas:
[0037] In the formula: is the population proportion coefficient of the transportation region to which the i-th transportation zone belongs; The resident population of the i-th traffic area as measured in the population census; This represents the total population of the transportation district to which this transportation community belongs in the census.
[0038] In this embodiment, the process of decomposing the predicted population at the scale of major transportation areas and intermediate transportation areas into smaller transportation areas in step S3 is as follows: When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone:
[0039] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; This represents the projected total population of the central transportation zone to which this transportation community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic area; When the model outputs the predicted total population of a city's transportation region, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding transportation zone within the region:
[0040] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
[0041] To ensure the accuracy and operability of population forecast results, this invention mainly constructs a multi-spatial-scale differentiated model parameter system based on statistical yearbook data published by various cities and national population census data. Through decomposition and aggregation, as well as multi-source data calibration based on mobile phone signaling, it realizes dynamic prediction of the spatial distribution of permanent residents at multiple spatial scales in the macro and micro aspects of the planning year.
[0042] The following uses a study of nine cities in a certain urban cluster as an example to illustrate the specific implementation method. The nine cities are A, B, C, D, E, F, G, H, and I.
[0043] Basic data processing: This study collected statistical yearbooks from 2010 to 2020 for nine cities within the study's urban agglomeration. Analysis revealed inconsistencies in the spatial scale and minimum statistical units of publicly available resident population data, making it difficult to directly generate consistent time-series datasets. For some cities, only district-level data on the historical evolution of resident population at the smallest spatial scale was available; for others, without district boundaries, more precise street / township-level data on the historical evolution of resident population could be directly obtained; and for some cities, multiple administrative division adjustments occurred during the study period, altering population statistical boundaries and resulting in inconsistencies in the time-series resident population data from the original yearbooks. To unify the spatial statistical scale of the population across the entire region, correct data biases caused by administrative division changes, and supplement high-precision micro-level population baseline data, the study further collected official national census data for the two most recent years from the nine cities within the study's urban agglomeration. This yielded unified street / township-level resident population census data for the entire region, providing accurate benchmark data for subsequent spatial scale population conversion, proportion coefficient calculation, and multi-scale population forecasting.
[0044] By dividing cities into major transportation zones, districts into intermediate transportation zones, and streets / townships into smaller transportation zones, the nine cities within the study urban cluster are ultimately divided into multiple smaller transportation zones, forming a spatial grid that covers the entire region at multiple spatial scales. This serves as an important basis for subsequent predictions of the resident population.
[0045] Comparison of prediction models: The available resident population time-series data for each city from 2010 to 2020 were divided into three spatial scales: traffic sub-regions, central traffic areas, and major traffic areas. Each city conducted "8+3" grouped modeling and prediction based on its own spatial scale: the 11-period historical time series at a single scale were split into a training set of the first 8 years and a test set of the last 3 years. Grey model, exponential smoothing model, moving average model, and growth rate model were used to fit and tune the training data for each city, and each model independently calculated the predicted resident population value for the corresponding scale in the last 3 years. The predicted values of each model were compared with the actual population in the yearbook at the same scale, and the root mean square error (RMSE) was used to evaluate the fitting accuracy, as shown in Equation 1. The model with the smallest RMSE was selected as the optimal resident population prediction model.
[0046] Extrapolation calculations were performed using the optimal model to obtain the predicted permanent resident population for each city at the corresponding spatial scale in 2021 and beyond.
[0047] Taking City A as an example, only district-level data on its historical resident population is available, allowing for the construction of separate time-series forecasting models for each district. Error comparison revealed that the root mean square error (RMSE) calculated using the exponential smoothing model for multiple target spatial scale units was the lowest. Therefore, the exponential smoothing model was selected for this type of spatial scale unit, and its corresponding parameters were matched as the optimal forecasting model. This process was repeated to obtain a series of differentiated model parameter systems covering different spatial scales across nine cities.
[0048] (1) In the formula, Let n be the root mean square error, and n be the number of samples. This is the projected value of the permanent resident population. This represents the actual number of permanent residents.
[0049] Calculate the population ratio of the traffic community: The spatial scale of the resident population prediction output by each city model is different. Some cities output a large transportation area (city level) and some output a medium transportation area (district level). It is necessary to use the population proportion coefficient to unify the predicted population at the large and medium scales to the transportation micro-area to obtain resident population data with a consistent scope across the entire urban agglomeration of this study, all at the transportation micro-area scale.
[0050] (1) Calculation of population proportion coefficient Based on the national census data of the most recent year, the population proportion coefficient of each traffic subdistrict within its respective traffic central area is calculated layer by layer: 1) Extract the actual permanent resident population of the entire transportation district under the Seventh National Population Census; 2) Extract the actual permanent resident population of each traffic subdistrict within the central area; 3) Calculate the population ratio of the traffic-adjacent area relative to the central area, as follows:
[0051] In the formula: The population proportion coefficient of the central district to which the i-th traffic zone belongs; The actual resident population of the community as measured by the Seventh National Population Census; The population of this community is determined by the total population of the 7th National Population Census in the Central Transportation District.
[0052] Similarly, calculate the proportion coefficient of the neighborhood relative to the larger area:
[0053] In the formula: The population proportion coefficient of the region to which the i-th traffic zone belongs; The actual resident population of the community as measured by the Seventh National Population Census; The total population of the seventh national census of the transportation region to which this community belongs.
[0054] Based on the stability verification of census data, a comparison of population census data from two baseline years shows that the population distribution ratio of neighborhoods within central and large districts fluctuates very little, indicating that the two sets of proportion coefficients can be used stably in the long term. A complete database of the two sets of population proportion coefficients is generated in batches by traversing all combinations of central, large, and neighborhood districts in all cities.
[0055] (2) Based on the proportion coefficient, the population is predicted and the permanent population of the transportation community is generated.
[0056] Time series models output population predictions at three scales: large transportation areas, medium transportation areas, and small transportation areas. The small area is the smallest spatial unit, and the predicted values for that area can be used directly. Population predictions for medium and large transportation areas need to be broken down to the small area level step by step. ① When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone:
[0057] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the central transportation zone to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic central area.
[0058] ② When the model outputs the predicted total population of a transportation region to which a city belongs, the predicted resident population of each community is calculated by combining the population proportion coefficient of the corresponding transportation community within the region:
[0059] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
[0060] By summing the projected populations of all traffic zones within the same traffic central area, we can obtain the projected total population of that central area:
[0061] In the formula, Projected total population for the central transportation area; Let be the predicted permanent resident population of the i-th traffic zone.
[0062] By summing up the population of all central districts within the same transportation region, the overall predicted permanent resident population of the region (city) can be restored. This enables two-way conversion of population at three spatial scales: region, central district, and sub-region, ensuring that the population values at each level are consistent and uniform.
[0063] Population forecast for the planning year: Input the planning year, calculate the predicted population at the corresponding spatial scale by operating the series of calibrated models for each city, and obtain the predicted resident population data at multiple spatial scales within the city cluster by decomposing the proportion coefficient and aggregating upwards.
[0064] Adjusting coefficients and dynamically constructing the model algorithm: To ensure that the model algorithm can adapt to the rolling updates of resident population data and realize the dynamic rolling prediction of the resident population in the planning year, the calibrated model is dynamically checked and adjusted by combining the real resident population data of subsequent years obtained from statistical yearbooks, mobile phone signaling and other data.
[0065] The above model algorithm is based on the resident population data from the Statistical Yearbook from 2010 to 2020. After obtaining the actual resident population data for 2021—for example, based on the 2021 resident population data for each district of City A—the actual resident population value is compared with the predicted value at the corresponding scale, and the average relative error of the city's resident population prediction is calculated. As shown in Equation 2.
[0066] (2) In the formula, The mean relative error is given by n, where n is the sample size. This is the projected value of the permanent resident population. This represents the actual number of permanent residents.
[0067] when When the percentage is ≤10%, the population prediction model method and parameter selection for the city are considered reasonable, and the prediction accuracy meets the requirements. To further improve data accuracy, an adjustment coefficient is introduced to correct the prediction results: the ratio of the actual value to the predicted value of each spatial grid is used as the adjustment coefficient for the corresponding spatial scale; this coefficient is uniformly applied to all traffic zones within the grid, thus obtaining the adjustment coefficients for the permanent residents of all traffic zones in the Greater Bay Area, such as... Figure 2 As shown, this coefficient is multiplied by the model's predicted permanent resident population data to achieve dynamic correction of the predicted permanent resident population data for each planning year.
[0068] For example, if the actual population of a certain municipal district is predicted to be 1 million, while the predicted value is 950,000, then the error for that municipal district is 5%. Similarly, the errors for other municipal districts can be obtained. The average error of all municipal districts is calculated. If it is less than 10%, the model prediction accuracy for that city is considered to meet the requirements, and the prediction result is adjusted using an adjustment coefficient method. For example, the adjustment coefficient for that municipal district = actual population / predicted population = 100 / 95 = 1.0526. The predicted population of each sub-spatial scale unit under its jurisdiction is also multiplied by 1.0526 for adjustment.
[0069] when When the percentage exceeds 10%, the city's population prediction model and parameters are deemed to require optimization and adjustment. The 2021 population data is merged with the data from 2010-2021, and the data is divided into training and test sets at a 70% to 30% ratio (2010-2017 data as the training set, 2018-2021 data as the test set). Model calibration is then performed, and the model algorithm and parameters are updated. If the resident population data is at the traffic zone scale, the proportion coefficient of each traffic zone at the central district level is recalculated, and the traffic zone proportion coefficient parameters are updated accordingly. For example, if the 2021 resident population data for a city is at the street scale, the proportion coefficient of each street needs to be recalculated. Then, using the updated model method and traffic zone proportion coefficients, the resident population for each planning year is re-predicted.
[0070] Adjusted population forecast for the planning year: Input the planning year, calculate the predicted population at the corresponding spatial scale again using the dynamically adjusted model algorithm above, and obtain the predicted resident population at multiple spatial scales of the nine cities in the study urban agglomeration through proportion coefficient decomposition and upward aggregation, thereby realizing the rolling update of the resident population forecast model and results.
[0071] Example 2: A multi-spatial-scale planning population dynamics prediction system for urban agglomerations includes: The data acquisition unit is used to collect the minimum spatial scale resident population data of each city in the urban agglomeration from the statistical yearbooks of the past n years, as well as the national population census data of the most recent year. Spatial grid division units are used to divide each city in the urban agglomeration into a major transportation zone, a central transportation zone, and streets / townships into smaller transportation zones, based on the administrative divisions of each city within the urban agglomeration, thus forming a spatial grid that covers the entire area at multiple spatial scales. The model library includes exponential smoothing models, moving average models, grey models, and growth rate models. The model calibration unit is used to divide the resident population data of the statistical yearbooks of various cities over the past n years into training set and test set; the smoothing model, moving average model, grey model and growth rate model are used to predict the test set data respectively, and the model with the smallest root mean square error is selected as the optimal model, thereby calibrating a series of differentiated population prediction models for each spatial scale unit in each city. The calculation unit is used to calculate the population proportion coefficient of each traffic zone in each city within its respective traffic superzone and traffic central zone; and to calculate and decompose the predicted population data at the traffic superzone or traffic central zone scale of each city to the respective traffic zones based on the corresponding population proportion coefficient.
[0072] In this embodiment, the model calibration unit divides the 11-year resident population data collected from each city into population data at three spatial scales: traffic sub-district, traffic central area, and traffic large area. Based on the spatial scale of each city, the first 8 years of historical resident population data for each spatial scale are used as the training set, and the last 3 years are used as the test set. The smoothing model, moving average model, grey model, and growth rate model are used to fit and adjust the parameters of the training data for each city. Each model independently calculates the predicted resident population value for the corresponding spatial scale in the last 3 years. The population prediction value output by each model is compared with the actual population of the yearbook at the same spatial scale. The root mean square error (RMSE) is used to evaluate the fitting accuracy, and the model with the smallest RMSE is selected as the optimal prediction model for resident population.
[0073] In this embodiment, the model calibration unit also recalibrates the model. The process is as follows: obtain the actual resident population data for year n+1, calculate the average relative error between the predicted and actual population values for each city, and when the error is within a set range, obtain an adjustment coefficient by comparing the actual population value with the predicted population value; dynamically adjust the resident population data for each year in the traffic zone cells contained in the spatial scale unit by multiplying the adjustment coefficient with the predicted resident population for each year; when the error between the predicted and actual values exceeds the assumed range, merge the data for year n+1 with the data from previous years and recalibrate the model.
[0074] In this embodiment, the process by which the calculation unit calculates the population proportion coefficient is as follows: Extract the actual resident population of the entire traffic central area from the census data; extract the actual resident population of each traffic sub-district within the central area; calculate the population ratio of the traffic sub-district relative to the central area, as follows:
[0075] In the formula The population proportion coefficient of the traffic central area to which the i-th traffic zone belongs; The resident population of the i-th traffic area as measured in the population census; The total population of the Jiaotong Central District as determined by the census is the total population of this Jiaotong residential area. Similarly, calculate the ratio of traffic zones to traffic areas:
[0076] In the formula: is the population proportion coefficient of the transportation region to which the i-th transportation zone belongs; The resident population of the i-th traffic area as measured in the population census; This represents the total population of the transportation district to which this transportation community belongs in the census.
[0077] In this embodiment, the process by which the computing unit decomposes the predicted population at the scale of major transportation regions and intermediate transportation regions into smaller transportation zones is as follows: When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone:
[0078] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; This represents the projected total population of the central transportation zone to which this transportation community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic area; When the model outputs the predicted total population of a city's transportation region, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding transportation zone within the region:
[0079] In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting population dynamics in urban agglomerations across multiple spatial scales, characterized by: Includes the following steps: S1. Collect the minimum spatial scale resident population data of each city in the urban agglomeration from the statistical yearbooks of the past n years, as well as the national population census data of the most recent year; take each city in the urban agglomeration as a major transportation region, take the district as a central transportation region, and take the streets and towns as small transportation areas to form a spatial grid with full coverage of multiple spatial scales. S2. Divide the collected resident population data from the statistical yearbooks of the past n years into training set and test set. For each city's smallest spatial scale unit, use exponential smoothing model, moving average model, grey model, and growth rate model to fit the training set data, and use these models to predict the test set data. Select the model with the smallest root mean square error as the optimal model, thereby calibrating a series of differentiated population prediction models for each spatial scale unit in each city. S3. Based on the latest national census data, calculate the population proportion coefficient of each traffic zone in each city within its respective major and central traffic zones; input the planning year, run the population prediction series model calibrated for each city to obtain the predicted population data at the corresponding spatial scale; based on the corresponding population proportion coefficient, decompose the predicted population data at the scale of major or central traffic zones in each city to each traffic zone under its jurisdiction to obtain the predicted resident population data at the scale of all traffic zones in each city, thereby obtaining the predicted resident population data at multiple spatial scales within the urban agglomeration.
2. The method for predicting population dynamics in urban agglomerations using multi-spatial-scale planning according to claim 1, characterized in that, It also includes the step of model recalibration: Obtain the actual resident population data for year n+1, calculate the average relative error between the predicted and actual population values for each city, and when the error is within a set range, use the ratio of the actual population value to the predicted population value to obtain an adjustment coefficient; by multiplying the adjustment coefficient by the predicted resident population for each year, dynamically adjust the resident population data for each year in the traffic zone cells contained in the spatial scale unit. When the error between the predicted value and the actual value exceeds the assumed range, the data from year n+1 is merged with the data from previous years, and the model is recalibrated.
3. The method for predicting population dynamics in urban agglomerations using multi-spatial-scale planning according to claim 1, characterized in that, The method for selecting the optimal model in step S2 is to divide the 11-year resident population data collected from each city into population data at three spatial scales: traffic sub-district, traffic central area, and traffic large area. Based on the spatial scale of each city, the first 8 years of historical resident population data for each spatial scale are used as the training set and the last 3 years are used as the test set. The training data for each city were fitted and the parameters were adjusted using a smoothing model, a moving average model, a grey model, and a growth rate model, respectively. Each model independently calculated the predicted value of the permanent resident population at the corresponding spatial scale for the next three years. The population predictions output by each model are compared with the actual population in the yearbook at the same spatial scale. The accuracy of the fit is evaluated by the root mean square error (RMSE). The model with the smallest RMSE is selected as the optimal prediction model for the permanent population.
4. The method for predicting population dynamics in urban agglomerations at multiple spatial scales according to claim 1, characterized in that, The calculation process for the population proportion coefficient in step S3 is as follows: Extract the actual resident population of the entire traffic central area from the census data; extract the actual resident population of each traffic sub-district within the traffic central area; calculate the population ratio of the traffic sub-district relative to the central area, as follows: ; In the formula The population proportion coefficient of the traffic central area to which the i-th traffic zone belongs; The resident population of the i-th traffic area as measured in the population census; The total population of the Jiaotong Central District as determined by the census is the total population of this Jiaotong residential area. Similarly, calculate the ratio of traffic zones to traffic areas: ; In the formula: is the population proportion coefficient of the transportation region to which the i-th transportation zone belongs; The resident population of the i-th traffic area as measured in the population census; This represents the total population of the transportation district to which this transportation community belongs in the census.
5. The method for predicting population dynamics in urban agglomerations using multi-spatial-scale planning according to claim 1, characterized in that, S3 The process of breaking down the predicted population at the level of major transportation regions and intermediate transportation regions into smaller transportation sub-regions is as follows: When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone: In the formula, To predict the permanent resident population of the i-th traffic area in a given year; This represents the projected total population of the central transportation zone to which this transportation community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic central area; When the model outputs the predicted total population of a city's transportation region, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding transportation zone within the region: ; In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
6. A multi-spatial-scale planning population dynamics prediction system for urban agglomerations, characterized in that, include: The data acquisition unit is used to collect the minimum spatial scale resident population data of each city in the urban agglomeration from the statistical yearbooks of the past n years, as well as the national population census data of the most recent year. Spatial grid division units are used to divide each city in the urban agglomeration into a major transportation zone, a central transportation zone, and streets / townships into smaller transportation zones, based on the administrative divisions of each city within the urban agglomeration, thus forming a spatial grid that covers the entire area at multiple spatial scales. The model library includes exponential smoothing models, moving average models, grey models, and growth rate models. The model calibration unit is used to divide the collected resident population data from the statistical yearbooks of various cities over the past n years into a training set and a test set. The test set data were predicted using smoothing model, moving average model, grey model and growth rate model respectively. The model with the smallest root mean square error was selected as the optimal model, thereby calibrating a series of differentiated population prediction models for each spatial scale unit in each city. The calculation unit is used to calculate the population proportion coefficient of each traffic zone in each city within its respective traffic superzone and traffic central zone; and to calculate and decompose the predicted population data at the traffic superzone or traffic central zone scale of each city to the respective traffic zones based on the corresponding population proportion coefficient.
7. The urban agglomeration multi-spatial-scale planning population dynamics prediction system according to claim 6, characterized in that, The model calibration unit breaks down the 11-year resident population data collected from each city into population data at three spatial scales: traffic sub-district, traffic central area, and traffic large area. Based on the spatial scale of each city, the first 8 years of historical resident population data for each spatial scale are used as the training set and the last 3 years are used as the test set. The training data for each city were fitted and the parameters were adjusted using a smoothing model, a moving average model, a grey model, and a growth rate model, respectively. Each model independently calculated the predicted value of the permanent resident population at the corresponding spatial scale for the next three years. The population predictions output by each model are compared with the actual population in the yearbook at the same spatial scale. The accuracy of the fit is evaluated by the root mean square error (RMSE). The model with the smallest RMSE is selected as the optimal prediction model for the permanent population.
8. The urban agglomeration multi-spatial-scale planning population dynamics prediction system according to claim 6, characterized in that, The model calibration unit also recalibrates the model. The process is as follows: obtain the actual resident population data for year n+1, calculate the average relative error between the predicted population value and the actual population value for each city, and when the error is within a set range, obtain the adjustment coefficient by comparing the actual population value with the predicted population value; and dynamically adjust the resident population data for each year of the traffic zone cell contained in the spatial scale unit by multiplying the adjustment coefficient with the predicted resident population for each year. When the error between the predicted value and the actual value exceeds the assumed range, the data from year n+1 is merged with the data from previous years, and the model is recalibrated.
9. The urban agglomeration multi-spatial-scale planning population dynamics prediction system according to claim 6, characterized in that, The process by which the calculation unit calculates the population proportion coefficient is as follows: Extract the actual resident population of the entire traffic central area from the census data; extract the actual resident population of each traffic sub-district within the traffic central area; calculate the population ratio of the traffic sub-district relative to the central area, as follows: ; In the formula The population proportion coefficient of the traffic central area to which the i-th traffic zone belongs; The resident population of the i-th traffic area as measured in the population census; The total population of the Jiaotong Central District as determined by the census is the total population of this Jiaotong residential area. Similarly, calculate the ratio of traffic zones to traffic areas: ; In the formula: is the population proportion coefficient of the transportation region to which the i-th transportation zone belongs; The resident population of the i-th traffic area as measured in the population census; This represents the total population of the transportation district to which this transportation community belongs in the census.
10. The urban agglomeration multi-spatial-scale planning population dynamic prediction system according to claim 9, characterized in that, The process by which the calculation unit decomposes the predicted population at the scale of major and medium transportation zones into smaller transportation zones is as follows: When the model outputs the predicted total population of a city's central transportation zone, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding central transportation zone: ; In the formula, To predict the permanent resident population of the i-th traffic area in a given year; This represents the projected total population of the central transportation zone to which this transportation community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic central area; When the model outputs the predicted total population of a city's transportation region, the predicted resident population of each transportation zone is calculated by combining the population proportion coefficient of the corresponding transportation zone within the region: ; In the formula, To predict the permanent resident population of the i-th traffic area in a given year; The projected total population of the transportation district to which this community belongs; is the population proportion coefficient of the i-th traffic zone within its respective traffic region.
Citation Information
Patent Citations
Method and system for predicting future population fine spatial distribution and computer readable storage medium
CN113689048A
