Parcel scale population prediction method based on improved queue element prediction method

Abstract

The invention discloses a plot scale population prediction method based on an improved queue element prediction method, and relates to the field of urban and rural planning and design technology and demography, and the method comprises the following steps: collecting and processing plot scale population data to form a plot scale population database; constructing a plot scale population prediction model based on a traditional queue element prediction method, and correcting fertility data and death data input; on the basis of calculating the normalized population mobility, considering the influence of major industrial projects, and correcting the plot scale population mobility; the aspects of data input, operation speed, output result and result precision of the plot scale population prediction model are improved, the calculation precision and speed of a traditional queue element prediction method are optimized, and plot scale population fine prediction is achieved. According to the method, mastering of the population development situation by government departments can be improved, development policy making and space resource allocation are assisted, and objectivity and scientificity of urban government are improved.

Description

Technical Field

[0001] The present invention relates to the fields of demography, urban and rural planning and design, and particularly relates to a plot-scale population prediction method based on an improved cohort-component projection method. Background Art

[0002] In the fields of urban and rural planning and design, policy formulation, etc., population prediction is a basic and important content. The population size, growth rate and population structure determine the scale and timing of the supply of urban infrastructure and public services. The function of urban population prediction is mainly to meet the development needs of infrastructure and public services from the supply perspective, as well as the land supply required for urban development.

[0003] Traditionally, urban population prediction methods include statistical regression method, growth extrapolation method, ratio method, etc. However, these methods have their own advantages and disadvantages, and the refinement degree is not high. From the perspective of refined prediction of population size and structure, the cohort-component method is more commonly used by academic circles, governments and market departments. Although the cohort-component method is a relatively classic population prediction method, it also has certain limitations. For example, it has high requirements for data, and the change or deviation of a certain data will have a greater impact on the entire result; economic factors and policy factors that have a greater impact on population quantity change are not considered.

[0004] Therefore, in order to improve the reliability of urban population prediction methods, it is necessary to improve the existing prediction methods: one is to obtain more stable and reliable results by acquiring multi-source big data with higher accuracy; the second is to improve and simplify the input of existing fertility data, death data and migration data; the third is to add policy factor variables that have a greater impact on population quantity change and correct the population migration rate. Summary of the Invention

[0005] The problem to be solved by the present invention is: to provide a plot-scale population prediction method based on an improved cohort-component projection method, which improves the traditional cohort-component projection method in terms of model input, operation speed, output result and result accuracy, realizes fine prediction of plot-scale population, and improves the scale and accuracy of population prediction.

[0006] The present invention adopts the following technical solutions: a plot-scale population prediction method based on an improved cohort-component projection method, comprising the following steps:

[0007] S1. Collect plot-scale population data, and conduct data verification and correction to form a plot-scale population database;

[0008] S2. Based on the traditional cohort-component projection method, construct a plot-scale population prediction model, correct the input of fertility data and death data, fuse the fertility data into a single vector, and generate a population fertility data matrix in sequence, and convert the population mortality rate into a population death data matrix:

[0009] S3. Calculate the population migration rate at the plot scale based on the corrected population fertility data and death data;

[0010] S4. Construct an industrial population migration model based on the impact of major industrial projects on population migration, and correct the population migration rate at the plot scale based on deep learning methods;

[0011] S5. Optimize the population prediction model at the plot scale, screen and sort out the input data of the population prediction model at the plot scale, including: population base, fertility rate of women of childbearing age, sex ratio at birth, population mortality rate, migrant population; input the optimized population prediction model at the plot scale, and conduct population prediction at the plot scale through the cohort component method to obtain the prediction results of total quantity, fertility, death, and migration data.

[0012] Preferably, in step S1, collect population data at the plot scale to form a population database at the plot scale, including the following sub-steps:

[0013] S1.1. Collect population data at the plot scale. The data sources include: official statistical data, LBS data, and basic geographic spatial data, and construct an analysis base map based on the basic geographic spatial data;

[0014] S1.2. Take the official statistical data as the reference standard to check and correct the collected LBS data, including: data cleaning, data association, and data correction;

[0015] S1.2. Store the checked and corrected data in the database, and perform standardization processing on the data with different sources and structures, unify the encoding, data type, storage format, and input / output interface to form a population database at the plot scale.

[0016] Preferably, in step S1, the official statistical data includes national population census data and local statistical yearbook data for each year, and the LBS (Location Based Services) data includes the population quantity by sex and age at the plot scale;

[0017] Among them, data cleaning is used to remove duplicate, incorrect data and repair incomplete data in the LBS data;

[0018] Data association is used to match the LBS data with the corresponding basic geographic spatial data and geographic units in the official statistical data to the same spatial scale;

[0019] Data correction is to correct the data through the ratio adjustment method and regression analysis method, and compare the corrected LBS data with the official statistical data in different regions and different time periods;

[0020] The ratio adjustment method adjusts the LBS data by calculating the population ratio of official statistical data to the original LBS data in each geographical unit; the regression analysis method corrects the LBS data by establishing a regression model between the official statistical data and the LBS data and using the predicted values of the regression model.

[0021] Preferably, in step S2, based on the traditional cohort element prediction method, a population prediction model at the plot scale is constructed. The population of each age in each year is regarded as a different birth cohort, and the prediction is made with an age group interval of 1 year and a time interval of 1 year.

[0022] The input of the fertility data and the death data is corrected, including the following sub-steps:

[0023] S2.1. Correct the input fertility data. Based on the census data, combined with the statistical yearbook data and the LBS data, establish a plot-scale population fertility data matrix in a multi-threaded input mode.

[0024] The plot-scale population fertility data matrix: Input the data of age, fertility rate, and sex ratio at birth for each year from the starting year to the predicted year, and construct them into a single vector. Combine multiple single vectors in order, and convert them into a plot-scale population fertility data matrix through programming language and store it in tabular form.

[0025] S2.2. Correct the input death data. Reconstruct the life table based on the actual mortality rate, and convert the life table into a plot-scale death population data matrix.

[0026] The plot-scale death population data matrix: Input the mortality rates by sex and age for each year from the starting year to the predicted year, and construct them into a single vector. Combine multiple single vectors in order, and convert them into a plot-scale population death data matrix through programming language and store it in tabular form.

[0027] Preferably, in step S3, calculate the plot-scale population migration rate, including the following sub-steps:

[0028] S3.1. Migration-free prediction: Assume that there are no migrant populations within the plot, including in-migrant populations and out-migrant populations. Input the original data without the number of migrant populations, including the base-year population data, age structure, fertility rate, and mortality rate data, into the plot-scale population prediction model for prediction to obtain the total population data. Compare the obtained total population data with the actual population data, and obtain the number of migrant populations each year by calculating the difference.

[0029] The original data includes: population base, fertility rate of women of childbearing age, sex ratio at birth, and population mortality rate data.

[0030] S3.2. With migration prediction, the number of migrant population obtained in step S3.1 and the original data are input into the population prediction model at the plot scale again for re - operation to conduct migration prediction and obtain an accurate migration rate prediction result.

[0031] Preferably, in step S4, based on the impact of major industrial projects on population migration, analyze the spatio - temporal distribution law of industrial migrant population, construct an industrial population migration model, predict the distribution quantity, distance, and direction of industrial migrant population in different periods and of different scales in the future, and feedback the prediction result of industrial migrant population to the plot scale, which specifically includes the following sub - steps:

[0032] S4.1. Define the analysis unit of the industrial population migration model;

[0033] The analysis unit is surrounded by geographical boundary lines, including industrial units and residential units, and the boundary is corrected based on the land use right boundary, river water system boundary, road red line, combined with the connected regulatory planning unit boundary and administrative village boundary;

[0034] The industrial unit is the area where industrial land is concentratedly distributed, and the residential unit is the area where residential land is concentratedly distributed;

[0035] S4.2. Construct an index system of influencing factors for the industrial population migration model and conduct index calculation;

[0036] The index system of influencing factors includes: supporting facilities, residential characteristics, transportation costs, and industrial characteristics;

[0037] The index calculation includes: calculation of the land area of industrial units, classification of industrial unit categories, calculation of the industrial population quantity within each industrial unit, calculation of residential location, calculation of the land area of residential units, calculation of residents' income, acquisition of the quantity of transportation supporting facilities, statistics of the quantity of public supporting facilities within the unit, calculation of motor vehicle travel time, and calculation of non - motor vehicle travel time;

[0038] S4.3. Construct a deep learning model and conduct model training;

[0039] Use the BPNN deep learning method to simulate the spatio - temporal distribution law of industrial migrant population. The BPNN neural network consists of an input layer, several hidden layers, and an output layer. Each layer contains several neuron nodes, and the output of each node is the result after the weighted sum of the outputs of all nodes in the previous layer is processed by an activation function;

[0040] S4.4. Training and data verification of the industrial population migration model;

[0041] Construct the weight matrix of the industrial population migration distribution model by calculating the error and backpropagating the error; calculate the connection strength between each influencing factor index and the population, clean the abnormal data and missing data, take the residential unit - industrial unit pair as a data sample, make a data set, and divide the data set into a training set and a validation set;

[0042] Use Huber Loss as the loss function and Mae (Mean Absolute Error) as the model accuracy observation value. After several rounds of training, observe the change curves of Train_Loss and Val_Loss with the number of training rounds to obtain the model training effect;

[0043] S4.5. Prediction of the migrated population and correction of the population migration rate of the plot;

[0044] Combine the predicted residential population of the newly added industrial plot to correct the population migration rate of the plot, supplement the impact of population migration caused by major industrial projects, and correct the population migration rate of the plot.

[0045] Preferably, in step S4.5, the prediction of the migrated population is as follows:

[0046] S4.5.1. Calculate the industrial characteristics of the industrial unit, the supporting facilities and residential characteristics indicators of the residential unit, and the transportation cost of the unit pair;

[0047] S4.5.2. Put the data obtained in step S4.5.1 into the input end of the industrial population migration distribution model to calculate the predicted value of the population connection strength between the unit pairs;

[0048] S4.5.3. Add up the predicted values of the population connection strength of each residential unit to obtain the industrial population migration value of all residential units in the whole region;

[0049] S4.5.4. Decompose the industrial population migration value to the residential plot according to the proportion of the residential land area of each residential plot in the residential unit.

[0050] Preferably, in step S5, input the input data into the plot - scale population prediction model constructed in step S2, and through the cohort component forecasting method, perform iterative calculations to obtain the prediction results of the total quantity category, fertility category, death category, and migration category data,

[0051] The total quantity category results reflect the overall scale and structural changes of the population, including: the number of people by age and gender, the natural population growth rate, the total population change rate, the number of people by gender, the percentage of the population by age and gender;

[0052] The fertility category results reflect the information on population fertility, including: the number of fertility people by age, the total number of fertility people by gender, the crude birth rate, the fertility rate by age, the total fertility rate;

[0053] Mortality-related results reflect the population mortality situation, including: the number of deceased population by age and gender, the total number of deceased population by gender, the crude mortality rate, the complete life table by age and gender, etc.;

[0054] Migration-related results reflect the population mobility situation, including: the number of net migrant population by age and gender, the total number of net migrant population by gender, the crude net migration rate.

[0055] Preferably, in step S5, the population prediction model at the plot scale is optimized from four aspects: input data, operation speed, output results, and result accuracy. The methods are as follows:

[0056] S5.1. Optimization of input data: The optimized input data are: the population base number in the predicted starting year, the fertility rate of women of childbearing age, the sex ratio at birth, the population mortality rate, and the total migrant population;

[0057] Fuse the input fertility rate of women of childbearing age and sex ratio at birth data into a single vector, generate a population fertility data matrix in order, convert the input population mortality rate into a population mortality data matrix, use the plot-scale population migration rate corrected by deep learning as the migration data input, and regard the input data of each year as an independent thread;

[0058] S5.2. Optimization of operation speed: Develop a kernel model using the R language, use vectors and matrices as basic operation units, perform operations on the input data by establishing multi-threads according to years, supplement the missing data of subsequent threads according to the thread order, and perform predictions in the form of data preloading;

[0059] S5.3. Optimization of output results: Output several types of core data for each plot and extend them;

[0060] S5.4. Optimization of result accuracy: Check the population prediction data results, predict from the starting year to the ending year, combine the existing statistical yearbook data and population census data, and check the prediction data according to the known year data.

[0061] Compared with the prior art, the present invention adopts the above technical solutions and has the following technical effects:

[0062] 1. The population prediction method at the plot scale of the present invention independently collects population data at the plot scale, checks it with statistical data, constructs a population database at the plot scale, conducts population prediction, and advances the prediction scale of the cohort component method to the plot scale.

[0063] 2. The population prediction method at the plot scale of the present invention improves the input data of the population prediction model at the plot scale, fuses the fertility data into a single vector and generates a population fertility data matrix in order, converts the population mortality rate into a population mortality data matrix, and reduces the complexity of data input.

[0064] 3. The population prediction method at the plot scale of the present invention constructs a population migration distribution model based on the traditional calculation of the population migration rate, aiming at the impact of the implementation of major projects on the large-scale population migration, and corrects the population prediction results.

[0065] 4. The population prediction method at the plot scale of the present invention optimizes the operation speed, output results, result accuracy, etc. of the population prediction model at the plot scale. The operation speed shows an exponential increase, and the output results show diversity.

[0066] 5. The population prediction method at the plot scale of the present invention improves the factors such as the high data requirements of traditional population prediction, the too large prediction scale, the low prediction accuracy, and the failure to consider the impact of major policies on population migration. It improves the fineness and accuracy of population prediction and provides support for the formulation of population policies and the allocation of spatial resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a schematic diagram for constructing a single vector of fertility data of the present invention;

[0068] Figure 2 It is a schematic diagram of the population fertility data matrix of the present invention;

[0069] Figure 3 It is a schematic diagram for constructing a single vector of death data of the present invention;

[0070] Figure 4 It is a schematic diagram of the death data matrix of the present invention;

[0071] Figure 5 It is a schematic diagram for calculating population migration data of the present invention;

[0072] Figure 6 It is a flowchart of the method for correcting the population migration rate at the plot scale of the present invention;

[0073] Figure 7 It is a schematic diagram of multi-threaded operation and data preloading of the present invention;

[0074] Figure 8 It is a flowchart of the population prediction method at the plot scale of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0075] To make the objectives, technical solutions, and advantages of the present invention clearer, the following further elaborates on the technical solutions of the application in conjunction with the accompanying drawings. The described embodiments are only a part of the embodiments related to the present invention. All non-innovative embodiments made by other researchers in the field based on this embodiment fall within the protection scope of the present invention. At the same time, for the step numbers in the embodiments of the present invention, they are only set for the convenience of elaboration and explanation, and no limitation is placed on the order between steps. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0076] In one embodiment of the present invention, a plot-scale population prediction method based on an improved queue element prediction method is as Figure 8 shown, and specifically includes the following steps:

[0077] The first step: Collect plot-scale population data to form a plot-scale population database.

[0078] (1) Conduct extensive data collection. The required data mainly comes from three aspects, namely official statistical data, LBS data, and basic geospatial data, etc.

[0079] The official statistical data mainly comes from the national population census data and local statistical yearbook data for each year. Among them, the population census includes the permanent population, the population by age and gender, the death population status by age and gender, and the fertility status of women of childbearing age by age and birth order. The statistical yearbook data includes the permanent population, the total fertility rate, the average life expectancy (male, female), the number of births, and the number of deaths.

[0080] The LBS data is location-based service data. Use the basic geospatial data to construct an analysis base map, and use the plot-level scale population LBS location data as the core component of local analysis. The LBS data is mainly the population quantity of plot-scale population by gender and age.

[0081] (2) Check and correct the collected LBS population data and population statistical data to ensure the accuracy and reliability of the data. Taking the official statistical data as the reference standard, through processing the LBS data, make it accurately reflect the actual population situation and be consistent with the official statistical data in terms of trend and total amount.

[0082] The processing steps mainly include: data cleaning, data association, and data correction, etc.

[0083] Data cleaning mainly removes duplicate, incorrect data, and repairs incomplete data, etc. from the LBS population data.

[0084] Data association is to match the LBS data with the corresponding geospatial information and the geographical units in the official statistical data to ensure data comparison and analysis at the same spatial scale.

[0085] Data correction adopts the ratio adjustment method and the regression analysis method. Among them, the ratio adjustment method calculates the population ratio of official statistical data and LBS original data in each geographical unit, and then adjusts the LBS data according to this ratio. The regression analysis method establishes a regression model between official statistical data and LBS data, and uses the predicted values of the model to correct the LBS data, so that the LBS data can better fit the trend presented by the statistical data.

[0086] The processed LBS data is compared with the official statistical data in multiple dimensions, such as comparing the data distribution in different regions and different time periods, to ensure the consistency and rationality of the data.

[0087] (3) Store the processed data to form a population prediction database at the plot scale.

[0088] Standardize the data from different sources and with different structures, formulate data processing flow specifications, ensure the integrity and standardization of data processing, unify coding, data types, storage formats, input and output interfaces, etc., and follow the principles of unified coding and unified specifications to form databases such as basic geography and population development.

[0089] The second step: Based on the traditional cohort component forecasting method, construct a population prediction model at the plot scale and correct the input of fertility data and death data.

[0090] The traditional cohort component forecasting method predicts the future population quantity and structure by considering the changes of each age group of the population over time. The population of each age in each year is regarded as a different birth cohort, and according to the internal laws of population development, based on the existing population quantity and structure, fertility, death and migration parameters are set for prediction.

[0091] In this embodiment, by inputting the base-year population data, age structure, fertility rate, mortality rate, and migration rate, iterative calculations are performed to predict indicators such as future population quantity, age structure, gender structure, birth population, and death population.

[0092] Specifically, with the age grouped by 1-year intervals (the highest age group is divided into those over 100 years old) and the time predicted at 1-year intervals, the calculation formula used is:

[0093]

[0094] Among them, M x,t is the population of the age group x~x + 1 years old in year t, where x = 0, 1, …, 100, and the population over 100 years old is classified into the 100-year-old group; is the survival probability from the age group x - 1~x years old in year t to the age group x~x + 1 years old in year t + 1; is the net migration population from the age group of x - 1 to x years old in year t to the age group of x to x + 1 years old in year t + 1.

[0095] Specifically, in this embodiment, innovative improvements are made to the three parts of input data, namely fertility, mortality, and migration, required for the cohort element prediction method using the advanced data processing algorithm.

[0096] In terms of fertility data, based on the census data, combined with the statistical yearbook data and LBS data, a population fertility data matrix at the plot scale that conforms to the multi-threaded input mode is established to facilitate the subsequent transformation of the input port of the model.

[0097] In terms of mortality data, different from the traditional mortality mode and the input method of the life table, using the public data and taking the mortality rate (survival rate) as the data base, the death population data matrix at the plot scale is reconstructed to facilitate the optimization and adjustment of the input port.

[0098] In terms of migration data, in addition to the migration mode used in the traditional method, the population changes caused by major policies are additionally considered for the migration rate, and the prediction results of the migration rate are corrected and optimized to improve the model accuracy.

[0099] (1) Correction of the population fertility data matrix

[0100] Fertility-related data includes two parts: age-specific fertility rate and sex ratio at birth, and the existing fertility data and pre-input data from the starting year of prediction to the ending year of prediction are required.

[0101] Traditional population prediction methods only use the data input of the age-specific fertility rate of women aged 15 - 49 in the starting year. In this embodiment, the input of fertility data is improved, and the age-specific, fertility rate, and sex ratio at birth data of each year of the prediction year are input in the cohort element prediction method.

[0102] The specific method is: construct the age, fertility rate, and sex ratio at birth of each year as a single vector, and then combine the single vectors of multiple years in order. Finally, a population fertility data matrix is formed. From the starting year to the prediction year, the fertility data single vector and matrix input are as Figure 1 and Figure 2 shown. For the convenience of program language conversion, both the single vector and the matrix exist in the form of a table. The vector retains the numerical value, and the unified format of scientific notation is used when constructing the matrix to facilitate floating-point calculation. Finally, it is converted into a matrix by the program language and then input into the model.

[0103] In terms of method, define the population fertility data matrix as A, the fertility data single vector as B, the year as y, the age as a, the fertility rate as f, and the sex ratio at birth as b. Then the corresponding year, age, fertility rate, and sex ratio at birth vectors can be expressed as B y 、B a 、B f, B b , the four vectors are merged into a matrix in sequence, denoted as the fertility matrix A of year y y :

[0104] A y = [B y , B a , B f , B b

[0105] On this basis, assuming the prediction cut-off year is t, and combining the matrices by year, the population fertility data matrix is as follows:

[0106] A 总 = [A y , A y+1 , A y+2 ... A t .

[0107] (2) Population death data matrix based on real mortality rate

[0108] Traditional population prediction methods often use fixed death models to obtain death data, which is not applicable to different regions. With the improvement of medical level and the increase of average life expectancy, using a fixed life table as the death model for calculation will lead to large errors.

[0109] In this embodiment, through the collation of existing data, based on the real mortality rate, a life table is reconstructed, and the life table is converted into a population death data matrix.

[0110] The specific method is: using the mortality rate by gender and age for each year to generate a single vector, and then constructing the population death data matrix in sequence with the single vector. Both the single vector and the matrix exist in tabular form, the vector retains the numerical value, and the matrix is unified in scientific notation format during construction for convenient floating-point calculation. Finally, it is converted into a matrix by programming language and input into the model. From the starting year to the prediction year, the single vector and matrix of death data are input as shown in Figure 3 and Figure 4 shown, and the calculation method refers to the population fertility data matrix.

[0111] Step 3: Calculate the population migration rate at the plot scale based on the existing population birth and death data.

[0112] Given the difficulty in obtaining population migration data at the plot scale, this embodiment divides the calculation of population migration data at the plot scale into two steps, as shown in Figure 5 shown, to cope with the situation of missing migration data.

[0113] ​First step, migration-free prediction. Assume that there is no migrant population (including in-migration and out-migration) within the plot area, and directly input relevant data such as the population base, fertility rate of women of childbearing age, sex ratio at birth, and population mortality rate into the model for preliminary calculation.

[0114] Then, compare the population total data obtained in the first step with the actual population data in detail. The difference between the two is the number of migrant population each year.

[0115] Second step, migration-included prediction. Input the migrant population data calculated in the first step and other original input data into the model again. After the model recalculates, finally obtain a more accurate prediction result of the population migration rate at the block scale.

[0116] Fourth step: Considering the impact of major industrial projects on population migration, correct the population migration rate at the block scale.

[0117] In this embodiment, aiming at the impact of the implementation of major projects on the large-scale migration of the population, a population migration distribution model is constructed to correct the population prediction results. By analyzing the spatio-temporal distribution law of industrial migrant population, an industrial population migration distribution model is formed to predict the distribution quantity, distance, and direction of industrial migrant population in different periods and of different scales in the future, and feedback the prediction results of industrial migrant population to the block scale. The key technical steps, as Figure 6 shown, include: analysis unit delineation, index system construction, deep learning model construction, and industrial migrant population distribution prediction.

[0118] (1) Analysis unit delineation

[0119] First, delineate the analysis unit of the population migration model. The analysis unit is the basic unit for calculating various indicators, including industrial units and residential units. Among them, the industrial unit is the area where industrial land is concentratedly distributed, surrounded by geographical boundary lines such as roads and rivers. The residential unit is the area where residential land is concentratedly distributed, surrounded by geographical boundary lines such as roads and rivers. The analysis unit should be connected to the boundary of the regulatory planning unit and the administrative village boundary as much as possible.

[0120] The method for delineating the analysis unit is as follows:

[0121] 1) According to the distribution of industrial land and residential land, delineate industrial units and residential units respectively;

[0122] 2) Each unit does not overlap or cross to avoid double counting of population and industrial land;

[0123] 3) The geographical center of the residential unit overlaps with the population distribution center of gravity, and the geographical center of the industrial unit should overlap with the center of gravity of the industrial land;

[0124] 4) The analysis unit is associated with the planning and management unit and the natural geographical demarcation line, facilitating demarcation and identification.

[0125] 5) Scattered industrial and residential land uses do not need to be included in the analysis unit.

[0126] (2) Construction of the indicator system

[0127] Identify the influencing factors of the distribution of industrial migration population, and construct an indicator system for model influence, including four major categories: supporting facilities, residential characteristics, transportation costs, and industrial characteristics.

[0128] Among them, the sub-indicators of supporting facilities include: the number of public supporting facilities within the unit, the number of public supporting facilities within 1 km around the unit, and the number of transportation facilities within the unit. The sub-indicators of residential characteristics include: residential location, residential land area, and residents' income. The sub-indicators of transportation costs include: motor vehicle travel time and non-motor vehicle travel time. The sub-indicators of industrial characteristics include: industrial categories, industrial land area, and industrial population quantity.

[0129] The specific indicator calculations are as follows:

[0130] Industrial land area: Calculate the total sum of the industrial land areas within each industrial unit.

[0131] Industrial categories: According to the "Classification of National Economic Industries", classify the industrial categories of the leading enterprises within each unit, and identify the national economic industry classification where the enterprises are located.

[0132] Industrial population quantity: Overlay the employment population data in the Internet LBS population portrait data with the industrial land map patches, and count the industrial population quantity within each industrial unit.

[0133] Residential location: Calculate the distance from each residential unit to the urban functional center based on the locations of the urban functional centers.

[0134] Residential land area: Based on the land use map patches, count the total sum of urban residential land and rural homesteads within each residential unit.

[0135] Residents' income: Based on the income attributes of the permanent population in the residential plots in the Internet LBS population portrait data, count the average residents' income within each residential unit.

[0136] Number of transportation supporting facilities: Based on the obtained Internet map POI data, count the number of public transportation stations such as bus stops, subway stations, and light rail stations within each residential unit.

[0137] Number of public supporting facilities within the unit: Based on the Internet map POI data, count the number of public facilities such as education, medical care, social welfare, sports, and culture within each residential unit.

[0138] Number of public supporting facilities within 1 km around the unit: Based on the POI data of the Internet map, count the number of public facilities such as education, medical care, social welfare, sports, and culture outside each residential unit and within the 1 km buffer range.

[0139] Motor vehicle travel time: Based on the path planning interface of the Internet map platform, calculate the average value of the shortest commuting times of walking and cycling between residential-industrial unit pairs respectively.

[0140] Non-motor vehicle travel time: Based on the path planning interface of the Internet map platform, calculate the average value of the shortest commuting times of driving and public transportation (such as buses, subways, light rails, etc., comprehensive public transportation) between residential-industrial unit pairs respectively.

[0141] (3) Construction and training of the deep learning model

[0142] Use the BPNN deep learning algorithm to simulate the spatio-temporal distribution law of the population migrating due to industrial relocation. The BPNN neural network consists of an input layer, a hidden layer (one or more), and an output layer, and each layer contains several neuron nodes. The output of each node is the result after the weighted sum of the outputs of all nodes in the previous layer is processed by the activation function, and it can handle complex non-linear problems.

[0143] The specific formula is as follows:

[0144] Y hat = ∑w * σ(∑w * σ(∑w * x + b) + b) + b

[0145] In the formula: Y hat is the predicted value of the population connection strength, indicating the strength of the industrial population connection between a certain industrial unit and a residential unit; x is the index of each influencing factor, where the supporting facilities and residential characteristics are the location of the residential unit, the area of residential land, the income of residents, the number of transportation supporting facilities, the number of public supporting facilities, etc. The industrial characteristics are the area of industrial land, the industrial categories, the number of industrial population, etc. of the industrial unit, and the transportation cost is the motor vehicle and non-motor vehicle transportation costs between the residential unit and the industrial unit; w is the weight corresponding to the index, which is a trainable parameter; σ is the activation function, adding non-linear transformation to the model; b is the bias term.

[0146] (4) Model training and data verification

[0147] Construct the weight matrix of the distribution law model of the migrating population by calculating the error and backpropagating the error. First, calculate the above four types of indicators and the population connection strength, and clean the abnormal data and missing data. Make a data set with the residential-industrial unit pair as a data sample. The total number of samples in the data set should not be too small. If the data volume is too small, it is easy to overfit. The data set is divided into a training set and a validation set, and the training set should account for more than 70% of the total number of data samples.

[0148] Then, use the Huber Loss as the loss function and the Mae as the model accuracy observation value. After several rounds of training, observe the change curves of Train_Loss and Val_Loss with the number of training rounds to observe the training effect of the model.

[0149] If both of them decrease with the number of training rounds, it indicates that the training effect is good; if only Train_Loss decreases, it means that the model is overfitting; if neither of the two observation indicators decreases, it means that the indicator system is not suitable for this area, or there are problems with data collection, and it is necessary to check the data quality and optimize the indicator system.

[0150] (5) Prediction of migrant population and correction of population migration rate of plots

[0151] Predict the industrial migrant population of each unit based on the industrial population migration model, and then decompose it to each plot. The specific method is as follows:

[0152] First, calculate the industrial characteristics of industrial units, the supporting facilities and residential characteristics indicators of residential units, and the transportation costs of unit pairs.

[0153] Secondly, put the data into the model as the input end, and calculate the predicted value of the population connection strength between unit pairs through the model;

[0154] Then, add up the predicted values of the population connection strength of each residential unit to calculate the industrial population migration value of all residential units in the whole region;

[0155] Finally, decompose the industrial population migration value to the residential plots according to the proportion of the residential land area of each residential plot in the residential unit.

[0156] Furthermore, combine the predicted residential population of the new industrial plot, correct the population migration rate of this plot, supplement the population migration impact caused by major industrial projects, and further improve the correction of the population migration rate of the plot.

[0157] Step 5: Operation and analysis of population prediction data at the plot scale.

[0158] According to the specific requirements of the improved queue element prediction method above, screen and sort out the input data applicable to population prediction at the plot scale. Specifically include: data such as population base, fertility rate of women of childbearing age, sex ratio at birth, population mortality rate, migrant population, etc. In this embodiment, the specific calculation method is as follows:

[0159] Prediction of total amount data:

[0160] P t = P t-1 + B t-1 - Dt-1 +M t-1

[0161] Among them, P t represents the total population in year t, and P t-1 represents the total population in year t - 1, B t-1 represents the number of births in year t - 1, D t-1 represents the number of deaths in year t - 1, M t-1 represents the number of migrants in year t - 1;

[0162] Fertility data prediction:

[0163]

[0164] Among them, B t represents the number of births in year t, W t,x represents the number of women aged x (15 ≤ x ≤ 49) in year t, f t,x represents the fertility rate of women aged x in year t;

[0165] Mortality data prediction:

[0166]

[0167] Among them, D t represents the number of deaths in year t, P y,i represents the number of people aged x in year t, n is the maximum age of the population statistics, m t,x represents the mortality rate of people aged x in year t;

[0168] Migration data prediction:

[0169] M t = kM t-1 + c

[0170] Among them, M t represents the number of migrants in year t, M t-1 represents the number of population migrations in year t - 1, and k and c are the proportionality coefficient and correction constant deduced by the artificial intelligence algorithm.

[0171] In the prediction and analysis stage, the above five types of data that have been sorted out are accurately input into the plot-scale population prediction model constructed in the second step. After the iterative operation of the model, more than thirty detailed results in four major categories, namely total quantity, fertility, mortality, and migration, are output.

[0172] The total quantity results comprehensively reflect the overall scale and structural changes of the population, including the number of people by age and gender, the natural population growth rate, the total population change rate, the number of people by gender, the percentage of the population by age and gender, etc.

[0173] Fertility-related results focus on information regarding population fertility, including the number of people giving birth by age, the total number of people giving birth by gender, the crude birth rate, the age-specific fertility rate, the total fertility rate, etc.

[0174] Mortality-related results focus on population mortality, including the number of deaths by age and gender, the total number of deaths by gender, the crude death rate, the complete life table by age and gender, etc.

[0175] Migration-related results mainly reflect the population flow, including the number of net migrant population by age and gender, the total number of net migrant population by gender, the crude net migration rate, etc.

[0176] Specifically, in this embodiment, the population prediction model at the plot scale is optimized. Based on the traditional cohort element prediction method, the optimization of the population prediction model at the plot scale is carried out from four parts: input data, operation speed, output results, and result accuracy. According to the population fertility data matrix, the population mortality data matrix based on the real mortality rate, and the correction of the population migration rate based on deep learning, the comprehensive innovation of the model construction method is completed around the input, operation speed, output results, and result accuracy of the model.

[0177] (1) Input data

[0178] Existing population prediction methods require too many input data, including 7 types of data such as the initial population, fertility level, fertility pattern, mortality level, mortality pattern, migration level, and migration pattern. Through the improvement of the input data, the population prediction model at the plot scale proposed in this embodiment only needs to input five types of values, namely: the population base number in the predicted starting year, the fertility rate of women of childbearing age, the sex ratio at birth, the population mortality rate, and the total migration population.

[0179] Among them, after the fertility rate of women of childbearing age and the sex ratio at birth are input, they will be fused into a single vector and the population fertility data matrix will be generated in sequence. The population mortality rate will be converted into a population mortality data matrix, and the correction of the migrant population based on deep learning will be input as high-precision migration data. The comprehensive data for each year will exist independently as a thread, providing support for the design to improve the operation speed of the model.

[0180] (2) Operation speed

[0181] The kernel model of this embodiment is developed using the R language. Since the R language is based on vectors and matrices as basic operation units and is very efficient in performing vector and matrix operations, data is loaded in the form of constructing vectors and matrices, and at the same time, the input data is processed in multiple threads by year for operation.

[0182] During the prediction analysis process, all years exist independently as a single thread. The prediction will continuously supplement the missing data of subsequent threads in the thread order. Through the form of data preloading, the operation response can finally reach the second level.

[0183] In this embodiment, as Figure 7 shown, taking 2010 as the starting year and 2035 as the prediction end year as an example, the analysis logic of multi-threaded data preloading is demonstrated.

[0184] (3) Output results

[0185] Traditional population prediction methods can only output 4-6 types of data results, including the total population, gender-segregated population, age-segregated population in future years, and relatively simple output data such as gender-segregated births, deaths, and migrations in future years, and cannot customize the specific output content.

[0186] The method of this embodiment can output 16 types of core data for each plot, extend the core data, and finally a total of 38 types of data can be output, and the content to be output can be specified.

[0187] (4) Result verification

[0188] Verify the population prediction data. From the starting year to the end year of the prediction, combined with the existing statistical yearbook data and population census data, verify the prediction data according to the data of known years. If the error is small, the prediction is considered reasonable and within the reasonable range, and subsequent prediction work can be carried out.

[0189] In summary, the plot-scale population prediction method of the present invention based on the improved cohort-component method independently collects plot-scale population data, improves the input of fertility data and death data in the cohort-component method, and corrects the impact of population migration for the implementation of major projects, optimizing the population prediction model. Further, only relevant population data needs to be collected during the analysis process and input according to fertility, death, and migration data, realizing the optimization of input data, operation speed, output results, and result accuracy, improving the factors such as high data requirements, large prediction scale, low prediction accuracy, and failure to consider the impact of major policies on population migration in traditional population prediction methods, improving the fineness and accuracy of population prediction, and providing support for population policy formulation and spatial resource allocation.

[0190] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A plot-scale population prediction method based on an improved queue element prediction method, characterized in that: The steps include: S1. Collecting plot-scale population data, and verifying and correcting the data to form a plot-scale population database; S2. Based on the traditional cohort element prediction method, a plot-scale population prediction model is constructed to modify the input of fertility data and death data, merge the fertility data into a single vector, and sequentially generate a population fertility data matrix, and convert the population mortality rate into a population death data matrix: S3. Calculate the population migration rate at the plot scale based on the revised population fertility and death data; S4. Based on the impact of major industrial projects on population migration, an industrial population migration model is constructed, and the population migration rate at the plot scale is corrected based on the deep learning method; S5. Optimize the plot-scale population prediction model, screen and organize the input data of the plot-scale population prediction model, including: population base, fertility rate of women of childbearing age, sex ratio at birth, population mortality rate, and migrant population; The optimized plot-scale population prediction model was input, and the plot-scale population prediction was performed using the cohort element prediction method to obtain prediction results for total population, fertility, death, and migration data.

2. The plot-scale population prediction method based on the improved queue element prediction method according to claim 1, characterized in that: In step S1, the plot-scale population data is collected to form a plot-scale population database, which includes the following sub-steps: S1.

1. Collect plot-scale population data from official statistical data, LBS data, and basic geographic spatial data, and build an analysis base map based on the basic geographic spatial data; S1.

2. Using official statistical data as a reference standard, the collected LBS data are checked and corrected, including data cleaning, data association and data correction; S1.

2. Put the verified and corrected data into the database, standardize the data from different sources and structures, unify the coding, data type, storage format, input and output interface, and form a plot-scale population database.

3. The plot-scale population prediction method based on the improved queue element prediction method according to claim 2, characterized in that: In step S1, the official statistical data includes national census data and local statistical yearbook data of each year, and the LBS data includes the population number by gender and age at the plot scale; The data cleaning is used to remove duplicate and erroneous data and repair incomplete data in the LBS data; The data association matches the LBS data with the corresponding basic geospatial data and the geographic units in the official statistics to the same spatial scale; The data correction is carried out by correcting the data by using the ratio adjustment method and regression analysis method, and comparing the corrected LBS data with the official statistical data in different regions and different time periods; The ratio adjustment method adjusts the LBS data by calculating the population ratio between the official statistical data and the original LBS data in each geographical unit; the regression analysis method establishes a regression model between the official statistical data and the LBS data, and uses the predicted value of the regression model to correct the LBS data.

4. The plot-scale population prediction method based on the improved queue element prediction method according to claim 2, characterized in that: In step S2, based on the traditional cohort element prediction method, a plot-scale population prediction model is constructed. The population of each age in each year is regarded as a different birth cohort, and the age is grouped by 1-year intervals and the time is predicted by 1-year intervals. The calculation formula is: Among them, M x,t is the population in the age group x to x+1 in year t, x = 0, 1, …, 100, and the population over 100 years old is classified as the 100-year-old group; is the survival probability of the age group of x-1 to x years old in year t to the age group of x to x+1 years old in year t+1; It is the net number of migrants from the age group x-1 to x years old in year t to the age group x to x+1 years old in year t+1.

5. The plot-scale population prediction method based on the improved queue element prediction method according to claim 4, characterized in that: In step S2, the input of birth data and death data is corrected, including the following sub-steps: S2.

1. Modify the input fertility data, based on the population census data, combined with statistical yearbook data and LBS data, to establish a plot-scale population fertility data matrix with a multi-threaded input mode; The plot-scale population fertility data matrix: input the age, fertility rate, and sex ratio at birth data for each year from the starting year to the predicted year, and construct it into a single vector; combine multiple single vectors in sequence, convert them into a plot-scale population fertility data matrix through a programming language, and store them in a table form. The specific method is as follows: Define the population fertility data matrix as A, the fertility data single vector as B, the year as y, the age as a, the fertility rate as f, and the sex ratio at birth as b. The corresponding year, age, fertility rate, and sex ratio at birth vectors are represented as B y , B a , B f , B b , merge the four vectors into a matrix in order, and record it as the birth matrix A in year y y : A y =[B y ,B a ,B f ,B b ] Assume that the prediction end year is t, and combine the matrices by year to obtain the population fertility data matrix A 总 : A 总 =[A y ,A y+1 ,A y+2 ……A t ] S2.

2. Correct the input death data, reconstruct the life table based on the real mortality rate, and convert the life table into a death population data matrix at the plot scale; The plot-scale death population data matrix: input the sex- and age-specific mortality rates for each year from the starting year to the predicted year, and construct it into a single vector. Multiple single vectors are combined in sequence, converted into a plot-scale population death data matrix through a programming language, and stored in a table form. The method is the same as the plot-scale population fertility data matrix method in step S2.

1.

6. The plot-scale population prediction method based on the improved queue element prediction method according to claim 5, characterized in that: In step S3, the population migration rate at the plot scale is calculated, including the following sub-steps: S3.

1. Prediction without migration: Assume that there is no migrant population within the plot, including in-migrants and out-migrants; input the original data excluding the number of migrants, including base year population data, age structure, fertility rate, and mortality rate data, into the plot-scale population prediction model, perform prediction, and obtain the total population data; Compare the total population data obtained with the actual population data, and calculate the difference to get the number of migrants each year; The original data include: population base, fertility rate of women of childbearing age, sex ratio at birth, and population mortality rate data; S3.2, with migration prediction, the number of migrant population obtained in step S3.1 is input into the plot-scale population prediction model together with the original data to recalculate and perform migration prediction to obtain accurate migration rate prediction results.

7. The plot-scale population prediction method based on the improved queue element prediction method according to claim 6, characterized in that: In step S4, based on the impact of major industrial projects on population migration, the temporal and spatial distribution of industrial migrant population is analyzed, an industrial population migration model is constructed, the number, distance and direction of industrial migrant population distribution at different periods and scales in the future are predicted, and the prediction results of industrial migrant population are fed back to the plot scale, which specifically includes the following sub-steps: S4.

1. Define the analytical unit of the industrial population migration model; The analysis unit is surrounded by geographical boundaries, including industrial units and residential units, and the boundary correction is carried out based on the land ownership boundary, river system boundary, road red line, combined with the connected control planning unit boundary and administrative village boundary; The industrial unit is the area where industrial land is concentrated, and the residential unit is the area where residential land is concentrated; S4.

2. Construct an index system of influencing factors of the industrial population migration model and calculate the indexes; The influencing factor index system includes: supporting facilities, residential characteristics, transportation costs, and industrial characteristics; The calculation of the indicators includes: calculation of industrial unit land area, classification of industrial units, calculation of industrial population in each industrial unit, calculation of residential location, calculation of residential unit land area, calculation of residents' income, acquisition of transportation supporting facilities, statistics of public supporting facilities in the unit, calculation of motor vehicle travel time, and calculation of non-motor vehicle travel time; S4.

3. Build a deep learning model and perform model training; The BPNN deep learning method is used to simulate the temporal and spatial distribution of industrial migration population. The BPNN neural network consists of an input layer, several hidden layers and an output layer. Each layer contains several neuron nodes. The output of each node is the weighted sum of the outputs of all nodes in the previous layer after being processed by the activation function. The formula is as follows: Y hat =∑w*σ(∑w*σ(∑w*x+b)+b)+b Where: Y hat is the predicted value of population connection strength, indicating the industrial population connection strength between a certain industrial unit and a residential unit; x is the index of each influencing factor; w is the weight corresponding to the influencing factor index, which is a trainable parameter; σ is the activation function, which adds nonlinear transformation to the model; b is the bias term; S4.4, training and data verification of industrial population migration model; By calculating the error and back-propagating the error, the weight matrix of the industrial population migration distribution model is constructed; the connection strength between each influencing factor index and the population is calculated, and abnormal data and missing data are cleaned, and a data set is made with a residential unit-industrial unit pair as a data sample, and the data set is divided into a training set and a validation set; Use Huber Loss as the loss function and Mae as the model accuracy observation value. After several rounds of training, observe the curves of Train_Loss and Val_Loss changing with the number of training rounds to obtain the model training effect. S4.5, prediction of migrant population and correction of land-based population migration rate; Based on the predicted resident population of the newly added industrial plots, the population migration rate of the newly added industrial plots is revised, the impact of population migration caused by major industrial projects is supplemented, and the population migration rate of the plots is revised.

8. The plot-scale population prediction method based on the improved queue element prediction method according to claim 7, characterized in that: In step S4.5, the prediction of the migrant population is as follows: S4.5.

1. Calculate the industrial characteristics of industrial units, supporting facilities and residential characteristics indicators of residential units, and transportation costs of unit pairs; S4.5.2, put the data obtained in step S4.5.1 into the input of the industrial population migration distribution model, and calculate the predicted value of the population connection intensity between the unit pairs; S4.5.

3. Add the predicted values ​​of population connection intensity of each residential unit to obtain the industrial population migration value of all residential units in the entire region; S4.5.

4. Decompose the industrial population migration value to residential plots based on the proportion of the residential land area of ​​each residential plot to the residential unit.

9. The plot-scale population prediction method based on the improved queue element prediction method according to claim 7, characterized in that: In step S5, the input data is input into the plot-scale population prediction model constructed in step S2, and the prediction results of the total population, fertility, death and migration data are obtained by iterative calculation through the cohort element prediction method. The calculation method is as follows: Total data forecast: P t =P t-1 +B t-1 -D t-1 +M t-1 Among them, P t represents the total population in year t, P t-1 represents the total population in year t-1, B t-1 represents the number of births in year t-1, D t-1 represents the number of deaths in year t-1, M t-1 represents the number of migrant population in year t-1; The total results reflect the overall size and structural changes of the population, including: the number of people by age and sex, the natural population growth rate, the total population change rate, the number of people by sex, and the percentage of people by age and sex; Fertility data forecast: Among them, B t represents the number of births in year t, W t,x represents the number of women aged x in year t, with age 15≤x≤49, f t,x represents the fertility rate of women aged x in year t; Fertility results reflect information on population fertility, including: number of births by age, total number of births by sex, original birth rate, fertility rate by age, and total fertility rate; Death data prediction: Among them, D t represents the number of deaths in year t, P y,i represents the number of people aged x in year t, n is the maximum age of the population, m is t,x represents the mortality rate of the population aged x in year t; Death results reflect the death of the population, including: the number of deaths by age and sex, the total number of deaths by sex, the original mortality rate, the full life table by age and sex, etc. Migration data prediction: M t =kM t-1 +c Among them, M t represents the number of migrants in year t, M t-1 represents the number of population migration in year t-1, k and c are the proportional coefficient and correction constant derived by the artificial intelligence algorithm; Migration results reflect population mobility, including: net migrant population by age and sex, total net migrant population by sex, and raw net migration rate.

10. The plot-scale population prediction method based on the improved queue element prediction method according to claim 9, characterized in that: In step S5, the plot-scale population prediction model is optimized from four aspects: input data, operation speed, output result, and result accuracy. The method is as follows: S5.

1. Optimization of input data: The optimized input data are: population base in the forecast starting year, fertility rate of women of childbearing age, sex ratio at birth, mortality rate, and total migrant population; The input data of fertility rate and sex at birth of women of childbearing age are merged into a single vector, and the population fertility data matrix is ​​generated in sequence. The input population mortality rate is converted into a population death data matrix. The plot-scale population migration rate corrected by deep learning is used as the migration data input, and the input data of each year exists independently as a thread. S5.

2. Optimization of operation speed: Use R language to develop the kernel model, use vectors and matrices as basic operation units, establish multi-threaded operations for input data by year, supplement the missing data of subsequent threads in thread order, and make predictions by preloading data; S5.

3. Optimization of output results: Output several types of core data for each plot and extend them; S5.

4. Optimization of result accuracy: Verify the population forecast data results from the starting year to the ending year, combine the existing statistical yearbook data and population census data, and verify the forecast data based on the known year data.

Images