CN110689055B - Cross-scale statistical index spatialization method considering grid unit attribute grading - Google Patents

Abstract

The invention discloses a cross-scale statistical index spatialization method considering the attribute classification of grid cells. First, the correlation between statistical indicators to be spatialized and multi-source data is analyzed at the scale of coarse-grained administrative units, and the statistical indicators to be spatialized are selected. The data with high correlation of the indicators is used as the modeling auxiliary data; then the grid statistics of various types of modeling auxiliary data are classified by the classification method, and the optimal number of classifications for each type of modeling auxiliary data is determined; then, At the administrative unit scale, construct the feature vector of grade proportion and input it into the regression model for training; then at the fine-grained grid unit scale, divide each grid unit according to the best grade of various auxiliary data to construct a feature vector and input it into the regression model The weights of the statistical indicators of each grid unit are obtained; finally, the total value of the statistical indicators to be spatialized in the administrative unit is distributed to each grid unit according to the weight to obtain the final grid statistical value. The method of the present invention can greatly improve the prediction accuracy.

Description

A Spatialization Method of Cross-scale Statistical Indicators Considering Grid Cell Attribute Classification

technical field

The invention relates to the field of geographic information science, including economic geography, population geography, environmental geography, etc., in particular to a cross-scale statistical index spatialization method that takes into account the classification of grid cell attributes.

Background technique

Spatialization of statistical indicators aims to reproduce the spatial distribution of statistical indicators in geographic grids or other divisions (such as hexagons, buildings, or communities), usually by converting the spatial expression of statistical indicators into coarse-grained administrative units. Convert to a fine-grained geographic grid. Spatialization of statistical indicators has been extensively studied on data such as population, GDP and grain output. It has important scientific significance and broad application prospects for finely depicting the spatial distribution of statistical indicators, rational allocation of auxiliary resources, and guidance for government decision-making.

In the process of implementing the present invention, the inventor of the present application found that the method of the prior art has at least the following technical problems:

Due to the lack of grid-scale training data for statistical indicators, traditional spatialization usually builds the correlation between administrative unit-level modeling factors and statistical indicators, and transfers this law from administrative units to grid units; There are scale differences across orders of magnitude between them, which leads to the downscaling problem of model migration, resulting in low spatialization accuracy of statistical indicators.

It can be seen from this that the method in the prior art has a technical problem of low precision.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a cross-scale statistical index spatialization method considering grid cell attribute classification, so as to solve or at least partially solve the technical problem of low precision existing in the methods in the prior art.

In order to solve the above technical problems, the present invention provides a cross-scale statistical index spatialization method that takes into account the classification of grid cell attributes, including:

Step S1: Analyze the correlation between the statistical indicators to be spatialized and the multi-source data at the coarse-grained administrative unit scale, and select data from the multi-source data that have a predetermined degree of correlation with the statistical indicators to be spatialized as modeling aids data;

Step S2: grading the grid statistics of various types of modeling auxiliary data by using a classification method, and determining the optimal classification result of each type of modeling auxiliary data through classification evaluation indicators;

Step S3: at the coarse-grained administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, construct a level proportion feature vector, and input the level proportion feature vector into the regression model for training, and obtain the training After the regression model;

S4: At the fine-grained grid unit scale, construct feature vectors for each grid unit according to the optimal classification results of various auxiliary data, and input the trained regression model to obtain the statistical index weight of each grid unit;

S5: Normalize the statistical index weights of the grid units included in each administrative unit, and distribute the total value of the statistical indicators to be spatialized in the administrative unit to each grid unit according to the weight, to obtain the final grid statistical value.

In one embodiment, step S1 specifically includes:

Step S1.1: Count m types of multi-source data attribute values that can be quantified at the grid unit scale on all n administrative units, where both n and m are positive integers greater than 0;

Step S1.2: Calculate the Pearson correlation coefficient between the statistical index to be spatialized and the attribute value of the multi-source data at the administrative unit level;

Step S1.3: Select multi-source data with a correlation coefficient greater than a threshold T as the final M-type modeling auxiliary data, where M is a positive integer greater than 0.

In one embodiment, step S2 specifically includes:

Step S2.1: Count the quantified values of M types of modeling auxiliary data on all N grid units, then each grid unit corresponds to M statistical values;

用分级方法将格网划分到不同等级，对每次分级的结果采取预设评价指标进行度量以确定最优的分级结果。Step S2.2: Grid statistics for the t-th type of modeling auxiliary data

The grid is divided into different grades by a grading method, and a preset evaluation index is used to measure the results of each classification to determine the optimal classification result.

In one embodiment, step S3 specifically includes:

Step S3.1: At the coarse-grained administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, and construct a level proportion feature vector β _i ,

表示属于第t类建模辅助数据第k个等级的格网数量，一个行政单元对应一个特征向量，特征向量中包含多种建模辅助数据的多个等级占比；Among them, Ni represents the total grid number of the _ith administrative unit,

Indicates the number of grids belonging to the k-th level of the t-th type of modeling auxiliary data, one administrative unit corresponds to one eigenvector, and the eigenvector contains the proportions of multiple levels of various modeling auxiliary data;

Step S3.2: Take an administrative unit as a sample, take the rank proportion feature vector of the administrative unit as the input, and the spatialized statistical index as the output, train the random forest regression model, and obtain the trained random forest regression model.

In one embodiment, step S4 specifically includes:

Step S4.1: For a grid unit, according to the optimal classification result, determine the level of the grid in various types of auxiliary modeling data, and construct a grid according to the level of the grid in various auxiliary modeling data. the eigenvectors of the net elements;

Step S4.2: Input the constructed feature vectors of all grid cells into the trained random forest regression model, and output the statistical index weights of the grid.

In one embodiment, in step S5, the total value of the statistical index to be spatialized in the administrative unit is allocated to each grid unit according to the weight, and the calculation method is as follows:

Among them, i represents the ith administrative unit, j represents the jth grid, SI _ij represents the final statistical index value of the jth grid of the ith administrative unit, and SI _i represents the statistics of the ith administrative unit to be spatialized The total value of the index, W _ij and W _iu are the j-th and u-th grid weights of the _ith administrative unit, respectively, and Ni represents the total number of grids of the ith administrative unit.

In one embodiment, the indicators to be spatialized include but are not limited to population and grain, and the total value of the statistical indicators to be spatialized in the administrative unit is allocated to each grid unit according to the weight to obtain the final grid statistical value, including:

The population and grain statistics of the administrative unit are distributed to each grid unit according to the weight, and the final grid population and grain output are obtained.

In one embodiment, the method further includes:

After the grid statistical indicators are obtained, the predicted grid values obtained are summarized in the next-level administrative unit of the model training scale, and compared with the actual statistical indicators of the administrative unit to verify the accuracy.

In one embodiment, the method for verifying the accuracy is specifically:

The indicators to measure the error are the mean absolute error MAE and the root mean square error RMSE, and the calculation formula is as follows:

表示第i个下一级行政单元的预测统计指标值，

表示第i个下一级行政单元的真实统计指标值,′表示下一级行政单元的数量。in,

represents the predicted statistical index value of the i-th next-level administrative unit,

Represents the true statistical index value of the i-th next-level administrative unit, and ′ represents the number of the next-level administrative unit.

The above-mentioned one or more technical solutions in the embodiments of the present application have at least one or more of the following technical effects:

The present invention provides a cross-scale statistical index spatialization method that takes into account the classification of grid cell attributes. First, at the coarse-grained administrative unit scale (eg, districts and counties), the statistical indicators to be spatialized and multi-source data (which can be located in grid cells) are analyzed. Scale and quantification) correlation, and select the data with high correlation with the statistical indicators to be spatialized as the modeling auxiliary data; and use the classification method to classify the grid statistical values of various modeling auxiliary data. The index determines the optimal grading result of each type of auxiliary modeling data; then, at the administrative unit scale, count the proportion of the number of grid cells of each grade in each type of auxiliary modeling data, construct the characteristic vector of grade proportion and input it into the regression model Then, at the fine-grained grid unit scale, according to the best level of various auxiliary data, the grid units are divided into each grid unit to construct feature vectors, and input the regression model, according to which we get

Statistical index weights of each grid unit; then take each administrative unit as a unit, normalize the statistical index weights of the grid units contained in it, and assign the total value of the statistical indexes to be spatialized in the administrative unit to each grid according to the weight. The final grid statistics are obtained in the grid cells.

Due to the method provided by the present invention, the regression modeling process of the spatialization of statistical indicators is unified on the grid scale, thereby avoiding that the existing spatialization method directly uses coarse-grained administrative unit data due to the lack of grid-scale training data for statistical indicators. Cross-scale issues caused when training and migrating to fine-grained grid cells. This method takes into account the hierarchical distribution characteristics of grid cell scale attributes, which makes the model migration smoother. Compared with the traditional spatialization method, the accuracy is higher, and it can provide a new solution for the spatialization of statistical indicators based on multi-source data fusion in the context of big data. , especially the spatialization of statistical indicators such as population, GDP, grain output, and meteorological and climatic factors.

Description of drawings

In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are For some embodiments of the present invention, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative efforts.

FIG. 1 is a schematic flowchart of a method for spatializing cross-scale statistical indicators in consideration of grid cell attribute classification provided by the present invention;

Fig. 2 is the technical roadmap of the method provided by the present invention;

Fig. 3 is the calculation flow schematic diagram of the method of the present invention;

Fig. 4 is the flow chart of the cross-scale feature extraction method in the present invention;

Figure 5 is a comparison chart of the average absolute error of the spatialization results of Wuhan population based on high, medium and low correlation POI data at the street level in a specific example (using three grid cell sizes of 1000 meters, 500 meters, and 200 meters);

Figure 6 is a comparison chart of the average absolute error at the street level of the spatialization results of Wuhan population determined by the DBI index and at the street level (three grid cell sizes of 1000 meters, 500 meters and 200 meters are used, and the errors are arranged from low to high );

Fig. 7 is a comparison chart of the mean absolute error at street level of the spatialization results of Wuhan population of the traditional method and the method of the present invention (using three grid cell sizes of 1000 meters, 500 meters and 200 meters);

FIG. 8 is a comparison diagram of the results and error distribution of the population spatialization of Wuhan City by the traditional method and the method of the present invention (200-meter grid unit size).

Detailed ways

The purpose of the present invention is to solve the problem that statistical indicators lack grid-scale training data, and directly transfer the laws learned by administrative units to grids, resulting in low spatialization accuracy of statistical indicators due to downscaling, and to provide a grid-based method. Spatialization of cross-scale statistical indicators for cell attribute classification.

For achieving the above object, main design of the present invention is as follows:

First, analyze the correlation between the statistical indicators to be spatialized and multi-source data (which can be quantified at the grid unit scale) at the coarse-grained administrative unit scale (such as districts and counties), and select the statistical indicators to be spatialized with high correlation. Then, use the classification method to classify the grid statistics of various types of modeling auxiliary data, and determine the optimal number of classifications for each type of modeling auxiliary data through the classification evaluation index; At the administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, construct the level proportion feature vector and input it into the regression model for training; then, at the fine-grained grid unit scale, according to various auxiliary data The optimal level is divided into each grid unit to construct a feature vector, and input the regression model to obtain the statistical index weight of each grid unit; Normalization is to assign the total value of the statistical indicators to be spatialized in the administrative unit to each grid unit according to the weight to obtain the final grid statistical value. The next-level administrative unit (eg, street) of the administrative unit scale used in the model training phase aggregates the grid prediction values and compares them with the actual statistical index values of the administrative unit to verify the accuracy.

The present invention unifies the regression modeling process of statistical index spatialization on the grid scale, thereby avoiding that the existing spatialization method directly uses coarse-grained administrative unit data for training and migrates to fine-grained administrative unit data due to lack of grid-scale training data for statistical indicators. Cross-scale problems caused by granular grid cells. This method takes into account the hierarchical distribution characteristics of grid cell scale attributes, which makes the model migration smoother. Compared with the traditional spatialization method, the accuracy is higher, and it can provide a new solution for the spatialization of statistical indicators based on multi-source data fusion in the context of big data. , especially the spatialization of statistical indicators such as population, GDP, grain output, and meteorological and climatic factors.

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

This embodiment provides a cross-scale statistical index spatialization method that considers grid cell attribute classification, see FIG. 1 , and the method includes:

Step S1: Analyze the correlation between the statistical indicators to be spatialized and the multi-source data at the coarse-grained administrative unit scale, and select data from the multi-source data that have a predetermined degree of correlation with the statistical indicators to be spatialized as modeling aids data.

Specifically, the scale of coarse-grained administrative units can be district, county, etc. Taking population spatialization as an example, multi-source data can include POI data, night light remote sensing data, land use type data, DEM data, and road network density data, etc. And multi-source data can be quantified at the grid cell scale. Different methods can be used to analyze the correlation between statistical indicators to be spatialized and multi-source data, such as graph correlation analysis, covariance and covariance matrix, correlation coefficient, etc.

In one embodiment, step S1 specifically includes:

Step S1.1: Count m types of multi-source data attribute values that can be quantified at the grid unit scale on all n administrative units, where both n and m are positive integers greater than 0;

Step S1.2: Calculate the Pearson correlation coefficient between the statistical index to be spatialized and the attribute value of the multi-source data at the administrative unit level;

Step S1.3: Select multi-source data with a correlation coefficient greater than a threshold T as the final M-type modeling auxiliary data, where M is a positive integer greater than 0.

In this embodiment, the correlation coefficient method is adopted. Specifically, calculate the Pearson correlation coefficient between the statistical indicators to be spatialized and the attribute values of multi-source data at the administrative unit level:

是所有行政单元统计指标的平均值；DV_i,t是第i个行政单元上第t类多源数据的量化数值，

是第t类多源数据在所有行政单元的平均量化数值。ρ的取值在-1与+1之间，若ρ>0，表明两个变量是正相关，ρ<0,则表明两个变量是负相关，皮尔森相关系数的绝对值越大则表示该两个变量间的相关性程度越大。Among them, ρ _t is the Pearson correlation coefficient between the t-th multi-source data and the statistical indicators to be spatialized at the administrative unit level, n is the total number of administrative units, and SI _i is the statistical indicator of the i-th administrative unit. ,

is the average of the statistical indicators of all administrative units; DV _i,t is the quantified value of the t-th multi-source data on the i-th administrative unit,

is the average quantitative value of the t-th multi-source data in all administrative units. The value of ρ is between -1 and +1. If ρ>0, it indicates that the two variables are positively correlated, and ρ<0, it indicates that the two variables are negatively correlated. The greater the degree of correlation between the two variables.

Then, the data with the correlation coefficient greater than a certain threshold T is selected as the final M-type modeling auxiliary data. It is generally believed that the two variables with the Pearson coefficient greater than 0.6 have strong correlation, so the multi-source data with the Pearson coefficient greater than 0.6 are used as Modeling auxiliary data. Therefore, the data with high correlation with the statistical index to be spatialized can be selected as the modeling auxiliary data.

Step S2: using a grading method to classify the grid statistical values of various types of modeling auxiliary data, and determining the optimal classification result of each type of modeling auxiliary data through the classification evaluation index.

Wherein, step S2 specifically includes:

Step S2.1: Count the quantified values of M types of modeling auxiliary data on all N grid units, then each grid unit corresponds to M statistical values;

用分级方法将格网划分到不同等级，对每次分级的结果采取预设评价指标进行度量以确定最优的分级结果。Step S2.2: Grid statistics for the t-th type of modeling auxiliary data

The grid is divided into different grades by a grading method, and a preset evaluation index is used to measure the results of each classification to determine the optimal classification result.

Specifically, the grading method can take different approaches. For example, the natural breakpoint method divides the grid into different levels. The number of classifications is set as k∈[2,K], and by changing the number of classifications, the results of each classification are measured by appropriate evaluation indicators to determine the optimal classification results, such as the DBI index:

是第t类建模辅助数据分级数量为k时的戴维森堡丁指数，

和

分别是分级结果中第x个和第y个等级的等级内平均距离，σ_x和σ_y分别是第x和第y两个等级中心间的距离。戴维森堡丁指数取值范围为[0,+∞)，DB越小意味着等级内距离越小，同时等级间距离越大。在划分等级后，选取戴维森堡丁指数最小的分级数量作为最优的分级数量。对所有M类数据重复步骤S2.2的步骤，为每一类建模辅助数据确定最优分级方案，从而得到最优分级数量C_t∈[2,K](t＝1,2,…,M)。in,

is the Davidson Pottinger index when the number of classifications of the t-th type of modeling auxiliary data is k,

and

are the average distances within the grades of the _xth and _yth grades, respectively, and σx and σy are the distances between the centers of the xth and yth grades, respectively. The value range of the Davidson Pottinger exponent is [0, +∞), and the smaller the DB is, the smaller the intra-level distance is, and the larger the inter-level distance is. After classifying the grades, the smallest number of grades with the Davidson Bodding Index is selected as the optimal number of grades. Repeat step S2.2 for all M types of data, and determine the optimal classification scheme for each type of modeling auxiliary data, so as to obtain the optimal classification number C _t ∈ [2, K] (t=1,2,…, M).

Step S3: at the coarse-grained administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, construct a level proportion feature vector, and input the level proportion feature vector into the regression model for training, and obtain the training The latter regression model.

Wherein, step S3 specifically includes:

Step S3.1: At the coarse-grained administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, and construct a level proportion feature vector β _i ,

表示属于第t类建模辅助数据第k个等级的格网数量，一个行政单元对应一个特征向量，特征向量中包含多种建模辅助数据的多个等级占比；Among them, Ni represents the total grid number of the _ith administrative unit,

Indicates the number of grids belonging to the k-th level of the t-th type of modeling auxiliary data, one administrative unit corresponds to one eigenvector, and the eigenvector contains the proportions of multiple levels of various modeling auxiliary data;

Step S3.2: Take an administrative unit as a sample, take the rank proportion feature vector of the administrative unit as the input, and the spatialized statistical index as the output, train the random forest regression model, and obtain the trained random forest regression model.

表示第1类建模辅助数据第1个等级的格网数量的占比，

表示第1类建模辅助数据第C₁个等级的格网数量的占比，通过步骤S3.1可以得到所有行政单元的特征向量，然后将其作为样本，将构建的等级占比特征向量作为输入对随机森林回归模型进行训练。Specifically, in the rank proportion feature vector β _i

represents the proportion of the grid number of the first level of the first type of modeling auxiliary data,

Indicates the proportion of the number of grids of the _first level of the first-class modeling auxiliary data. Through step S3.1, the feature vectors of all administrative units can be obtained, and then they are used as samples, and the constructed feature vector of level proportion is used as Input to train a random forest regression model.

In order to verify the effectiveness of the method, the method also includes:

After the grid statistical indicators are obtained, the predicted grid values obtained are summarized in the next-level administrative unit of the model training scale, and compared with the actual statistical indicators of the administrative unit to verify the accuracy.

Among them, the method of verifying the accuracy is as follows:

The indicators to measure the error are the mean absolute error MAE and the root mean square error RMSE, and the calculation formula is as follows:

表示第i个下一级行政单元的预测统计指标值，

表示第i个下一级行政单元的真实统计指标值,’表示下一级行政单元的数量。in,

represents the predicted statistical index value of the i-th next-level administrative unit,

Represents the true statistical indicator value of the i-th lower-level administrative unit, ' represents the number of lower-level administrative units.

S4: At the fine-grained grid unit scale, construct feature vectors for each grid unit according to the optimal classification results of various auxiliary data, and input the trained regression model to obtain the statistical index weight of each grid unit.

Specifically, the trained regression model is obtained in step S3. In this step, on the fine-grained grid unit scale, feature vectors are constructed for each grid unit according to the optimal classification result of the auxiliary data, so that the grid unit can be obtained. Statistical indicator weights.

In one embodiment, step S4 specifically includes:

Step S4.1: For a grid unit, according to the optimal classification result, determine the level of the grid in various types of auxiliary modeling data, and construct a grid according to the level of the grid in various auxiliary modeling data. the eigenvectors of the net elements;

Step S4.2: Input the constructed feature vectors of all grid cells into the trained random forest regression model, and output the statistical index weights of the grid.

Specifically, if the grid belongs to the k-th level of the t-th type of modeling auxiliary data, the eigenvalue is assigned to 1 at the feature vector code corresponding to the k-th level of the t-th type of data, and the other levels of the t-th type of data are coded is set to 0, so that the eigenvectors of all grid cells can be constructed.

When predicting through the regression model, the present invention does not directly take the predicted value as the final statistical value, but takes it as the weight of the statistical value, so as to provide a basis for the next index space conversion, thereby improving the prediction accuracy. Taking population spatialization as an example, the statistical index refers to the population, and the statistical population of the administrative unit needs to be finally converted into the grid population. The output value of the model can be regarded as the population weight of each grid, and finally the statistical population of the administrative unit is allocated to the grid according to the weight of the population in each administrative unit. The reason why the model output value is not directly regarded as the grid population value in the prediction stage is that by distributing the population according to the weight within the administrative unit, it can ensure that there is no error in the total grid population at the administrative unit level, and the accuracy is higher.

S5: Normalize the statistical index weights of the grid units included in each administrative unit, and distribute the total value of the statistical indicators to be spatialized in the administrative unit to each grid unit according to the weight, to obtain the final grid statistical value.

Specifically, after obtaining the weights of the statistical indicators of the grid cells, they are normalized, and then these statistical indicators can be allocated to the grid cells according to the weights, thus realizing the coarse-grained level (administrative unit) to Fine-grained level (grid cell) conversion.

In one embodiment, in step S5, the total value of the statistical index to be spatialized in the administrative unit is allocated to each grid unit according to the weight, and the calculation method is as follows:

Among them, i represents the ith administrative unit, j represents the jth grid, SI _ij represents the final statistical index value of the jth grid of the ith administrative unit, and SI _i represents the statistics of the ith administrative unit to be spatialized The total value of the index, W _ij and W _iu are the j-th and u-th grid weights of the _ith administrative unit, respectively, and Ni represents the total number of grids of the ith administrative unit.

Among them, the indicators to be spatialized include but are not limited to population and grain. The total value of the statistical indicators to be spatialized in the administrative unit is allocated to each grid unit according to the weight, and the final grid statistical value is obtained, including:

The population and grain statistics of the administrative unit are distributed to each grid unit according to the weight, and the final grid population and grain output are obtained.

Please refer to FIG. 2, which is a technical roadmap of the method provided by the present invention. First, select modeling data (including administrative unit scale data statistics, correlation analysis, and determine modeling auxiliary data), and then determine the optimal grid classification (grid Grid cell scale data statistics, grid grading, and optimal grading based on evaluation indicators), followed by feature modeling and migration (model modeling, model prediction, and grid statistical index weights), and finally grid allocation and accuracy. tests (distribution by weight, grid statistics, and accuracy tests).

Figure 3 shows the calculation flow of the method of the present invention. When performing the administrative unit level correlation analysis, it can be realized by calculating the Pearson correlation coefficient, the Spearman correlation coefficient, and the Kendall correlation coefficient. Equal interval grading, natural segment point method grading or equal quantile grading can be used. The evaluation and grading results can use the contour coefficient, DBI coefficient, elbow method and other indicators.

In order to more clearly illustrate the implementation process and beneficial effects of the method provided by the present invention, specific examples will be used below to describe in detail.

The existing Wuhan street administrative division data, street population data and 10 types of POI data need to be spatialized according to the POI data and allocated to a 200m grid to obtain a more refined spatial distribution of the population. Due to the lack of real grid population training data, traditional population spatialization modeling methods often directly apply the rules learned from administrative units to the grid, resulting in downscaling problems in the model migration process.

By adopting a cross-scale population spatialization method and grading grid cells according to statistical information, the present invention overcomes the feature cross-scale problem that occurs in the migration process of training and prediction in the traditional population spatialization method, and realizes a more refined Population Spatialization.

The algorithm process of the present invention will be described in detail below in conjunction with the accompanying drawings in the present invention, and the specific steps are as follows:

1) Count the number of 10 types of POIs and the number of population in all streets in Wuhan, calculate the Pearson coefficients of the 10 types of POIs and the population at the street level, and select 4 types of POIs with a Pearson coefficient greater than 0.6 as the final modeling auxiliary data ;

2) Map the four types of POI points to each grid cell, and count and record the number of various POIs in each grid cell. Use the natural breakpoint method and DBI index to determine the optimal number of grid classifications for each type of POI as follows:

①Organize the statistical values of all grids of a class of POIs into a one-dimensional vector, set the number of classifications to [2,10], and use the natural breakpoint method to classify them;

②Continuously change the number of classifications, repeat step ①, calculate the DBI index for each classification result, and select the classification result with the smallest DBI index as the optimal classification result;

③ Steps ① and ② are performed on all 4 types of POIs, and the optimal number of grading and grading results are determined for all 4 types of POIs;

3) Construct feature vectors at street level and train them as follows:

① Count the total number of grids contained in a street and the number of grids at each level of each auxiliary data determined according to step 2);

② Divide the number of grids at each level by the total number of grids to obtain the proportion of the number of grids of each level of the four types of POI, and construct the feature vector of the level proportion;

③ Perform steps ① and ② on all streets in Wuhan to obtain the feature vectors of all streets;

④ Input the feature vector of each street and the number of street population into the regression random forest model for training.

4) Construct feature vectors at grid level and make predictions, the steps are as follows:

① For a grid unit, according to the number of various POIs in the grid, which level the grid belongs to, assign the eigenvalue to 1 at the eigenvector code corresponding to the level, and the rest are 0

②Construct according to step ① to obtain the feature vectors of all grids in Wuhan;

③ Input the feature vector of the grid into the trained random forest model, and output the population weight of each grid;

Wherein 3) the feature vector construction method of 4) step is as shown in Figure 4;

5) In all streets, normalize the population weights of the grids it contains, and then assign the population of the streets to each grid according to the weights.

6) At the community level, the grid population it contains is aggregated and compared with the real community statistical population to measure the accuracy of population spatialization.

The beneficial effects of the invention are as follows: the invention proposes a cross-scale statistical index spatialization method, which classifies the grid cells according to the statistical information, so that the statistical features in the coarse-grained administrative units have the attributes of the fine-grained grid cells, The final spatialization result is then obtained by transferring the rules obtained from the administrative unit modeling to the grid.

This method can effectively select auxiliary data that is similar to the spatial distribution pattern of the statistical indicators to be spatialized through correlation analysis, and the spatialization accuracy is high; as shown in Figure 5, three groups of POIs with different correlations are used as modeling aids. The results show that although there are anomalies under the 1000m grid cell size, the spatialization accuracy generally improves with the improvement of the correlation of the modeling data. At the same time, the optimal classification is determined based on the attribute level distribution characteristics of grid cells, which has lower errors compared with methods such as arbitrary classification that do not consider the attribute level characteristics of grid cells; as shown in Figure 6, the DBI index is used to select the classification of grid cells The results show that there is no absolutely constant optimal grading, and appropriate evaluation indicators can assist in determining a better grading scheme. Finally, compared with the feature statistics and model transfer of the traditional method, this method overcomes the problem of directly transferring the laws learned by administrative unit modeling to the grid due to the lack of grid-scale training data for statistical indicators to a certain extent. downscaling problem; as shown in Fig. 7, the traditional method outperforms the present method at a grid cell size of 1000m, while at 500m and 200m grid sizes, the present method outperforms the traditional method, and as the grid becomes larger As shown in Figure 8, taking the comparison of the spatialization results of the 200m grid cell population in Wuhan as an example, it can be seen that the traditional method generally overestimates the grid population at the edge of Wuhan. In addition, the visualization effect is easy to appear patchy, and the results of the method of the present invention improve both. From the street population error map, the method of the present invention has significantly improved the prediction accuracy in the three rectangular box areas as shown in the figure compared with the traditional method; the results It is shown that the method of the present invention is more suitable for grids with smaller scales, and has greater advantages over the traditional methods in the spatialization of statistical indicators with larger scale spans.

Although the preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiment and all changes and modifications that fall within the scope of the present invention.

Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, provided that these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (9)

1. A cross-scale statistical index spatialization method considering grid cell attribute classification, characterized in that, comprising: Step S1: Analyze the correlation between the statistical indicators to be spatialized and the multi-source data at the coarse-grained administrative unit scale, and select data from the multi-source data that have a predetermined degree of correlation with the statistical indicators to be spatialized as modeling aids data; Step S2: grading the grid statistics of various types of modeling auxiliary data by using a classification method, and determining the optimal classification result of each type of modeling auxiliary data through classification evaluation indicators; Step S3: at the coarse-grained administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, construct a level proportion feature vector, and input the level proportion feature vector into the regression model for training, and obtain the training After the regression model; S4: At the fine-grained grid unit scale, construct feature vectors for each grid unit according to the optimal classification results of various auxiliary data, and input the trained regression model to obtain the statistical index weight of each grid unit; S5: Normalize the statistical index weights of the grid units included in each administrative unit, and distribute the total value of the statistical indicators to be spatialized in the administrative unit to each grid unit according to the weight, to obtain the final grid statistical value. 2. The method of claim 1, wherein step S1 specifically comprises: Step S1.1: Count m types of multi-source data attribute values that can be quantified at the grid unit scale on all n administrative units, where both n and m are positive integers greater than 0; Step S1.2: Calculate the Pearson correlation coefficient between the statistical index to be spatialized and the attribute value of the multi-source data at the administrative unit level; Step S1.3: Select multi-source data with a correlation coefficient greater than a threshold T as the final M-type modeling auxiliary data, where M is a positive integer greater than 0. 3. The method of claim 1, wherein step S2 specifically comprises: Step S2.1: Count the quantified values of M types of modeling auxiliary data on all N grid units, then each grid unit corresponds to M statistical values; Step S2.2: Grid statistics for the t-th type of modeling auxiliary data

The grid is divided into different grades by a grading method, and a preset evaluation index is used to measure the results of each classification to determine the optimal classification result. 4. The method of claim 1, wherein step S3 specifically comprises: Step S3.1: At the coarse-grained administrative unit scale, count the number and proportion of each level grid unit in each type of modeling auxiliary data, and construct a level proportion feature vector β _i ,

Among them, Ni represents the total grid number of the _ith administrative unit,

Indicates the number of grids belonging to the k-th level of the t-th type of modeling auxiliary data, one administrative unit corresponds to one feature vector, and the feature vector contains the proportions of multiple levels of various modeling auxiliary data; Step S3.2: Take an administrative unit as a sample, take the rank proportion feature vector of the administrative unit as the input, and the spatialized statistical index as the output, train the random forest regression model, and obtain the trained random forest regression model. 5. The method of claim 1, wherein step S4 specifically comprises: Step S4.1: For a grid unit, according to the optimal classification result, determine the level of the grid in various types of auxiliary modeling data, and construct a grid according to the level of the grid in various auxiliary modeling data. the eigenvectors of the net elements; Step S4.2: Input the constructed feature vectors of all grid cells into the trained random forest regression model, and output the statistical index weights of the grid. 6. The method of claim 1, wherein in step S5, the total value of the statistical index to be spatialized in the administrative unit is distributed to each grid unit by weight, and the calculation method is as follows:

Among them, i represents the ith administrative unit, j represents the jth grid, SI _ij represents the final statistical index value of the jth grid of the ith administrative unit, and SI _i represents the statistics of the ith administrative unit to be spatialized The total value of the index, W _ij and W _iu are the j-th and u-th grid weights of the _ith administrative unit, respectively, and Ni represents the total number of grids of the ith administrative unit. 7. The method according to claim 1, wherein the indicators to be spatialized include but not limited to population and grain, and the total value of the statistical indicators to be spatialized in the administrative unit is allocated to each grid unit according to the weight, to obtain Final grid statistics, including: The population and grain statistics of the administrative unit are distributed to each grid unit according to the weight, and the final grid population and grain output are obtained. 8. The method of claim 1, further comprising: After the grid statistical indicators are obtained, the predicted grid values obtained are summarized in the next administrative unit of the model training scale, and compared with the actual statistical indicators of the administrative unit to verify the accuracy. 9. The method of claim 8, wherein the method for verifying accuracy is specifically: The indicators to measure the error are the mean absolute error MAE and the root mean square error RMSE, and the calculation formula is as follows:

in,

represents the predicted statistical index value of the i-th next-level administrative unit,

Represents the true statistical index value of the i-th next-level administrative unit, and ′ represents the number of the next-level administrative unit.

Images