Forest resource asset homogeneity division method based on two-step distance and dynamic constraint

Abstract

The application discloses a forest resource asset homogeneous region division method based on two-step distance and dynamic constraint, which comprises the following steps: (1) collecting attribute data related to forest resource assets, including social economy, hydrology, topography and natural resource asset conditions; and calculating attribute space data sets; (2) proposing a two-step distance algorithm to perform spatial restriction clustering connection to obtain a clustering connection graph; (3) using the sum of square differences SSD to define partition information loss, and deleting connection edges from top to bottom to obtain the final partition result with the minimum information loss as the target; (4) selecting different restriction sets to obtain partition results, and selecting the best restriction set and the corresponding partition result after comparative analysis. The homogeneous region division result obtained by the method has lower information loss, and is suitable for homogeneous region division of different resource or administrative unit levels, and has certain universality and generalization.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of spatial region division, and particularly relates to a forest resource asset homogeneous region division method based on two-step distance and dynamic constraint. BACKGROUND

[0002] In order to scientifically and comprehensively reflect the value of natural resource assets, the Ministry of Natural Resources proposes a natural resource asset value accounting path of "paying equal attention to physical quantity and value quantity, and then value quantity after physical quantity", and also carries out natural resource asset inventory pilot work of the whole people in many places. According to the characteristics of various natural resource assets owned by the whole people, the natural resource asset homogeneous region with consistent natural, social and economic conditions is an important content of the inventory work, and directly affects the advantages and disadvantages of the inventory price system. Therefore, the natural resource asset region division work needs to be carried out under the principles of meeting the integrity of administrative boundaries, unified function and spatial connection and merger, so as to estimate the average price of natural resource assets in homogeneous regions and establish the inventory price system.

[0003] At present, there are two methods for dividing natural resource asset homogeneous regions. One is to modify and improve the existing division results, such as referring to the existing relevant division results (China Forestry Development Division 2011) in the Technical Guide for Comprehensive Inventory of Natural Resource Assets (Trial), and combining the latest administrative boundary data to integrate, connect and merge the administrative boundaries to obtain the corresponding homogeneous region division results. However, the existing natural resource division results are less time-effective, and the economic properties of natural resources are not considered, which leads to the inconsistency between the final division results and the actual situation, and makes it difficult to establish the asset inventory price system. The other is to try to use spatial data analysis tools, such as spatial clustering algorithm, which can divide spatial data into a series of spatial clusters according to the similarity between spatial data, so that the similarity of spatial entities in the same cluster is as high as possible. Traditional spatial clustering algorithms can be mainly divided into three types: partition clustering algorithm, hierarchical clustering algorithm and density clustering algorithm. However, not all spatial clustering algorithms can be applied to regionalization. In order to perform hierarchical clustering, some algorithms are usually constructed to be dissimilarity matrices. Moreover, the existing methods cannot dynamically update the adjacency matrix during the clustering process, which leads to the failure of connecting two clusters that should be adjacent.

[0004] Therefore, in view of the deficiencies of the existing division methods, such as incomplete indicators and clustering algorithms without dynamic updating, the application proposes a regionalization method combining clustering algorithm and dynamic constraint to divide natural resource asset homogeneous regions. SUMMARY

[0005] In order to construct a natural resource asset homogeneous region meeting the requirements of natural resource asset inventory work, the application provides a homogeneous region division method based on multiple types of data, which adopts a two-step distance clustering algorithm to define the cluster distance, and can ensure that the regionalization result is continuous in space, and takes into account the dynamic updating of the cluster distance and the dynamic adjacency condition.

[0006] In order to achieve the above-mentioned purpose, the forest resource asset homogeneous region division method based on two-step distance and dynamic constraint provided by the application comprises the following steps:

[0007] S1: Obtain the county-level vector map of the study area (R), and collect attribute data related to forest resource assets (socio-economic, hydrological, topographical and natural resource asset conditions). The attribute data is processed in the basic unit of the county-level administrative region, and finally a spatial attribute data set (R') connected with the county-level administrative region is formed;

[0008] S2: Obtain a spatial adjacency matrix (C) based on the spatial relationship of the county-level administrative regions in the study area (R), process a first-order edge data set (E) based on the spatial attribute data set (R'), and input the spatial adjacency matrix (C) of the study area, the spatial attribute data set (R') and the first-order edge data set (E) to propose a two-step distance method for spatial restriction clustering connection to obtain a clustering connection graph;

[0009] S3: Define the partition information loss by using the sum of squared differences (SSD), input the clustering connection graph and attribute spatial data, limit the expected number of regions (L) and the minimum number of units in the region (M), and aim to minimize the information loss to delete the connection edge (N-1) from top to bottom to obtain the final partition result;

[0010] S4: Select different limit sets (L, M) to obtain the partition results respectively, and select the best limit set and the corresponding partition result after comparative analysis.

[0011] As a preferred scheme, in the step S1, the attribute data is processed in the basic unit of the county-level administrative region, and finally the attribute space connected with the county-level administrative region is formed, and the specific formula is as follows:

[0012] The expression of the spatial data set is:

[0013] R={R1,R2,…,Rn} (1) n

[0014] In the formula, the study area is regarded as a set R, there are n county-level administrative regions in the study area, and R1 is the first county-level administrative region.

[0015] ​If the selected attribute data after the selection has m classes, the attribute space data set expression is:

[0016]

[0017] In the formula, R' is a space data set, R' nm is R n The data value of the mth attribute of the county.

[0018] Further, in the step S2, the spatial adjacency matrix (C) is obtained based on the spatial relationship of the counties in the research area (R), and the specific process is as follows:

[0019] The spatial adjacency matrix C (n, n) is set, and C (u, v) ∈ C (n, n) is set. If R u , R v Two counties are adjacent in space (i.e., share a certain distance boundary), then C (u, v) = 1.

[0020] In the step S2, the first-order edge data set (E) is obtained based on the spatial attribute data set (R'), and the specific process is as follows:

[0021] When C (u, v) = 1, R u , R v The distance d uv of two counties in spatial attribute is called the first-order edge, and all the first-order edges in the research area are collectively referred to as the first-order edge data set.

[0022]

[0023] In the formula, Nu and Nv are the number of counties included in Ru and Rv, respectively. In this formula, they are both 1. Therefore, the formula can be rewritten as:

[0024]

[0025] For any two counties Ru and Rv, if C (u, v) = 1, then add the edge e uv = <R u , R v > to E.

[0026] In the step S2, the two-step distance algorithm is used to perform spatial restriction clustering connection based on the research area spatial adjacency matrix (C), the spatial attribute data set (R'), and the first-order edge data set (E), and the specific process is as follows:

[0027] S2.1: Input and parameter setting: research area R, spatial adjacency matrix: C, spatial attribute data: R', first-order edge data set: E, set the clustering connection graph to empty set T;

[0028] S2.2: Find the shortest edge e in E lm , and C(l, m) must be 1;

[0029] S2.3: Draw the cluster join graph, and add e lm to T;

[0030] S2.4: Update the study area, let R k = R u U R v , and remove R u , R v from R;

[0031] S2.5: Update the spatial join matrix, for any R x , if C(x, u) = 1 or C(x, v) = 1, then C(x, k) = 1;

[0032] S2.6: Update the first-order edge dataset, for any R x , if C(x, k) = 1, then add e xk to E;

[0033]

[0034] where e xk is the attribute space distance between R x and R k , where R k = R u U R v ;

[0035] S2.7: Repeat steps S2.2-2.6 until |R| = 1, and obtain the complete spatial join graph T.

[0036] Further, in the step S3, the sum of squared differences (SSD) is used to define the loss of partition information, and the specific formula is as follows:

[0037]

[0038] SSD = SSD(R) - SSD(RA) - SSD(RB) (7)

[0039] In formula (6), R is a region, SSD(R) represents the SSD value of R, m is the number of attributes, n is the number of objects in R, R′ ij is the jth attribute value of the ith object, is the average value of the jth attribute of all objects in R;

[0040] In formula (7), R is a region, and R=RA∪RB. In step S3, each time a best connecting edge is removed, a region R is divided into two regions RA and RB, and the information loss degree of this step is calculated using the formula.

[0041] In step S3, the clustering connection graph (T) and the attribute space data (R') are taken as inputs, the expected number of regions (L) and the minimum number of units in a region (M) are taken as limits, and the last partition result is obtained by removing the connection edge (L-1) from top to bottom with the minimum information loss as the target. The specific steps are as follows:

[0042] S3.1: Set the ideal partition set to be empty set Ω;

[0043] S3.2: Calculate the SSD(R') value;

[0044] S3.3: Take R as the segmentation region, remove the connection edge e in the clustering connection graph, divide the region into two parts RA and RB, and calculate the corresponding SSD value; if |RA|<M or |RB|<M, the SSD value of this segmentation method does not need to be calculated;

[0045] S3.4: Find the smallest connection edge e in the SSD values of multiple segmentation methods best ;

[0046] S3.5: Remove e in the clustering connection graph T best , RA and RB are added to the ideal partition Ω;

[0047] S3.6: Let a=SSD(RA')-SSD(RB'), if a>0, then the segmentation region R=RA, remove RA in Ω, otherwise the segmentation region R=RB and remove RB in Ω;

[0048] S3.7: Repeat steps S3.3-S3.6 until |Ω|=L.

[0049] The advantages and beneficial effects of the present application are as follows:

[0050] The natural resource asset homogeneous region division strategy proposed by the present application has high application value, and the specific advantages are as follows:

[0051] Compared with the prior art, in view of the problems that the spatial connectivity is not considered in the method, and the dynamic adjustment is not carried out in the clustering process, the application firstly proposes a dynamic first-order edge data set construction method, obtains a spatial adjacency matrix according to the spatial relationship of the research area, provides spatial constraints for clustering connection, and dynamically adjusts the spatial connection matrix or the first-order edge data set during the spatial connection, solves the problem of inconsistent spatial continuity of the region before and after the connection, then uses the sum of square differences (SSD) to define the partition information loss, and carries out the regional division with the loss as a constraint, so that the information loss of the regionalization result is minimized, and finally the best regionalization scheme is obtained.

[0052] After the method of the application is adopted, firstly, each homogeneous region is continuous in space, meeting the requirement of complete and continuous administrative boundary; secondly, the expected number of regions and the minimum number of units in the region can be modified in the division process to meet different requirements and facilitate comparison to obtain the best number of regions and the corresponding regionalization scheme; finally, the homogeneous region division result based on the method of the application can well reduce the information loss and ensure the consistency of the attributes in each homogeneous region. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 It is a county-level administrative region distribution map of the research area of the application;

[0054] Figure 2 It is a clustering connection map formed in the division process of the application;

[0055] Figure 3 It is a county-level homogeneous region distribution map after division of the application. DETAILED DESCRIPTION

[0056] The application will be further described in detail below in combination with the drawings and specific embodiments.

[0057] Embodiment 1

Taking homogeneous region division of forest resource assets in Qinghai Province as an example

[0058] The data shows that, as of 2020, Qinghai Province has 2 prefecture-level cities, 6 autonomous prefectures, and a total of 45 county-level administrative regions. However, among them, Golmud City manages Tanggula Mountain Town, and there is still a Zhiduo County between the two places. Therefore, when setting the research area, Tanggula Mountain Town of Golmud City is separately taken as a region, that is, there are 46 sub-research areas in the research area (the specific position distribution of the research area R is shown in Figure 1 ).

[0059] After analysis and demonstration, it is considered that the factors affecting the division of homogeneous regions of forest resource assets should be considered from four aspects of hydrological conditions, social and economic conditions, geographical location and forest resource conditions. Based on this, the invention collects and collates the forest land quality grades (attribute 1) of each county in Qinghai Province in 2020, the primary net productivity grades (attribute 2) of each county, the forest stock volume (attribute 3) of each county, the altitude (attribute 4) of each county, the per capita GDP (attribute 5) of each county, and the rainfall (attribute 6) of each county, and obtains the spatial attribute data set R' (Table 1).

[0060] Table 1 Spatial attribute data set R'

[0061]

[0062]

[0063] According to the spatial relationship of the research area, the spatial connection matrix C (Table 2) is obtained, and the first-order edge data set E (Table 3) is obtained by calculating and processing based on the spatial attribute data set. In view of the large amount of data samples, the spatial connection matrix and the first-order edge data set (i.e. Table 2 and Table 3) shown in this embodiment are only part of the data, but they do not affect the qualitative analysis in this embodiment, and they also have sufficient support to prove the effectiveness of the beneficial effects of this embodiment.

[0064] Table 2 Spatial connection matrix C

[0065]

[0066]

[0067] Table 3 First-order edge data set E

[0068]

[0069]

[0070] Next, taking the spatial adjacency matrix of the research area, the spatial attribute data set and the first-order edge data set as input, based on the two-step distance algorithm, the spatial limited clustering connection is carried out to obtain the clustering connection graph T( Figure 2 ).

[0071] Table 4 Clustering connection sequence

[0072] Serial number Coupling content Serial number Coupling content Serial number Coupling content 1 R33-R37 16 R22-R26 31 R6-R17 2 R1-R4 17 R32-R35 32 R22-R27 3 R1-R2 18 R18-R24 33 R6-R18 4 R42-R43 19 R27-R29 34 R6-R20 5 R6-R12 20 R30-R31 35 R28-R30 6 R42-R44 21 R22-R25 36 R6-R8 7 R20-R23 22 R42-R45 37 R5-R6 8 R8-R10 23 R40-R42 38 R36-R38 9 R13-R19 24 R39-R41 39 R34-R36 10 R8-R13 25 R36-R46 40 R34-R40 11 R1-R3 26 R32-R33 41 R22-R34 12 R6-R9 27 R30-R32 42 R22-R39 13 R14-R15 28 R20-R21 43 R5-R22 14 R7-R16 29 R6-R11 44 R1-R5 15 R5-R7 30 R6-R14 45 R1-R28

[0073] Finally, taking the clustering connection graph T( Figure 2), attribute space dataset as input, select different expected number of regions and the minimum number of units in the region as the limit, to minimize the loss of information from top to bottom to delete the connection edge to get a number of partition results. After analysis and comparison and combined with the actual situation and the requirements of the relevant departments, the final forest resource asset homogenization regionalization results are determined (see Figure 3 ).

[0074] The sum of square differences (SSD) is used to define the partition information loss. The regional division results obtained based on the method of the present application are compared with the information loss of the study area without regional division and the original study area (Table 5).

[0075] Table 5 Comparison of information loss of different study area states

[0076] Research area state Original research area Forestry development area division result Invention area division result Whether regional division No Yes Yes Division method — Divided according to experience Based on two-step distance and dynamic constraint Information loss 270 172.4 95.4

[0077] Firstly, according to the partition result map ( Figure 3 ), it can be seen that each homogenization region is spatially aggregated and continuous, which meets the requirements of complete and continuous administrative boundary zoning and meets the actual situation and the requirements of the relevant departments. Secondly, based on Table 5, the information loss of the original study area state without homogenization regionalization is 270, and the information loss of the forestry development zoning result obtained by simple division according to experience is 132.4. It shows that homogenization zoning can well reduce the information loss of homogenization region, but the reduction of information loss by simple division according to experience is limited. The regionalization result obtained based on the method of the present application is only 95.4, which reduces nearly two-thirds of the information loss, and to a certain extent, it can better ensure the consistency of the information within each homogenization region. Therefore, the forest resource asset homogenization zoning strategy proposed in the present application is feasible and superior, and the method can also be used for homogenization zoning of different resources or administrative unit levels according to demand, which has certain universality and generalization.

[0078] The division of forest resource asset homogenization region needs to divide the units with similar attributes and adjacent positions into the same homogenization region, and ensure that no information loss is caused in the division process. Therefore, intuitively, the homogenization region after division is continuous in space, and the information loss caused by division is the least. According to this standard, the results of the example have verified that the present application effectively achieves the purpose of homogenization regionalization with the least information loss.

Claims

1. A method for dividing forest resource asset homogenous regions based on two-step distance and dynamic constraints, characterized in that: The method comprises the following steps: S1: Obtain the county-level vector map of the study area R, and collect attribute data related to forest resource assets, including social economy, hydrology, topography, and natural resource asset conditions; process the attribute data in units of counties, and finally form a spatial attribute data set R' connected with the county-level administrative region space; S2: Obtain a spatial adjacency matrix C based on the spatial relationship of the county-level administrative regions within the study area R, and process the spatial attribute data set R' to obtain a first-order edge data set E; input the spatial adjacency matrix C, the spatial attribute data set R', and the first-order edge data set E, and perform spatial restriction clustering connection based on a two-step distance algorithm to obtain a clustering connection graph; In the step S2, the spatial adjacency matrix C is obtained based on the spatial relationship of the county-level administrative regions within the study area R, and the specific process is as follows: Set the space adjacency matrix C(n, n), let C(u, v) ∈ C(n, n), if R u , R v Two counties are adjacent in space, that is, share a certain distance of the border, then C(u, v) = 1; In the step S2, the first-order edge data set E is obtained by processing the spatial attribute data set R', and the specific process is as follows: C(u, v) = 1 when R u , R v The distance d uv between two counties in spatial attribute is called a first-order edge, and all the first-order edges in the study area are collectively referred to as a first-order edge dataset; In the formula, N u , N v are R u , R v respectively, the number of counties in the administrative region, in this formula, are all 1. Therefore, the formula can be rewritten as: For any two counties R„ Rv, if C(u, v) = 1, then add edge e uv = <R u , R v >, In the step S2, the spatial restriction clustering connection is performed based on the two-step distance algorithm with the spatial adjacency matrix C, the spatial attribute data set R', and the first-order edge data set E as inputs, and the specific process is as follows: S2.1: Input and parameter setting: study area R, spatial adjacency matrix C, spatial attribute data R', first-order edge data set E, and set the clustering connection graph to an empty set T; S2.2: Find the shortest edge e in E lm and C(l,m) must be 1; S2.3: Draw the cluster join graph, drawing e lm is added to T; S2.4: Update the study region, let R k = R u UR v , and remove R u from R v ; S2.5: update the spatial join matrix, C(x, k) = 1 for any R x C(x, k) = 1 if C(x, u) = 1 or C(x, v) = 1. S2.6: Update the first-order side dataset for any R x , if C(x, k) = 1, add e to E xk ; where e xk is R x and R k is the distance between the attribute spaces of the two regions, where R k = R u UR v ; S2.7: Repeat steps S2.2-2.6 until |R| = 1 to obtain a complete spatial connection graph T; S3: Define the partition information loss using the sum of squares SSD, input the clustering connection graph and attribute spatial data, limit the expected number of regions L and the minimum number of units M in the region, and aim to minimize the information loss to delete the connection edge N-1 from top to bottom to obtain the final partition result; S4: Select different restriction sets (L, M) to obtain partition results, compare and analyze them, and select the best restriction set and the corresponding partition result.

2. The forest resource asset homogeneous region division method based on two-step distance and dynamic constraint according to claim 1, wherein: In the step S1, the attribute data is processed in units of county-level administrative regions, and finally an attribute space connected with the county-level administrative region space is formed, and the specific formula is as follows: The spatial data set expression is: R = {R1, R2,..., R n} (4) In the formula, the study area is regarded as a set R, there are n county-level administrative regions in the study area, and R1 is the first county-level administrative region; If the selected attribute data after selection has m types, then the attribute spatial data set expression is: where R' is a spatial data set, R' nm is the data value of the mth attribute of the county. n is the data value of the mth attribute of the county.

3. The forest resource asset homogeneous region division method based on two-step distance and dynamic constraint according to claim 1 or 2, wherein: In the step S3, the partition information loss is defined using the sum of squares SSD, and the specific formula is as follows: SSD = SSD(R) - SSD(RA) - SSD(RB) (7) In Equation (6), R is a region, SSD(R) represents the SSD value of region R, m is the number of attributes, n is the number of objects in R, R' is the complement of R, and ij is the jth attribute value of the ith object, is the average quality of the jth attribute of all objects in R. In the formula (7), R is a region, and R = RA∪RB; in step S3, each time a best connection edge is removed, a region R is divided into two regions RA and RB, and at this time, the formula is used to calculate the information loss degree of this step.

4. The two-step distance and dynamic constraint based forest resource asset homogeneity zoning method according to claim 3, characterized in that: In step S3, the clustering join graph T and the attribute space data R' are taken as inputs, the expected number of regions L and the minimum number of units M in the region are taken as limits, and the last zoning result is obtained by removing the join edges L-1 from top to bottom with the minimum information loss as the goal. The specific steps are as follows: S3.1: Set the ideal partition set to be empty set Ω; S3.2: Calculate the SSD(R') value; S3.3: Take R as the segmentation region, remove the join edge e in the clustering join graph in turn, divide the region into two parts RA and RB, and calculate the corresponding SSD value. If |RA|<M or |RB|<M, there is no need to calculate the SSD value of this segmentation method; S3.4: Find the smallest connected edge e among the SSD values of the various segmentation methods best ; S3.5: Remove e from the clustered join graph T best Add RA and RB to the ideal partition Ω. S3.6: Let a=SSD(RA')-SSD(RB'), if a>0, then the segmentation region R=RA, remove RA from Ω, otherwise the segmentation region R=RB and remove RB from Ω; S3.7: Repeat steps S3.3-S3.6 until |Ω|=L.

Images