A Heuristic Facility Location Method Based on Hybrid Index Acceleration and Multi-Source Data Fusion

Abstract

A heuristic facility location selection method based on hybrid index acceleration and multi-source data fusion, comprising: a) performing grid preprocessing on the geographical space, dividing the geographical space into grids with a fixed side length; b) preprocessing the trajectory data and point of interest (POI) data, mapping the location information therein to grid coordinates, and establishing an index structure between the trajectory data and POI data and the grids; c) constructing a hybrid index structure based on a balanced tree + hash set, and selecting the location with the highest scoring function from it using a heuristic search strategy; d) selecting and updating the bidirectional index of the hybrid index and grid trajectory according to the number of currently selected locations; if the number of selected locations does not meet the requirements, repeat step d), otherwise stop the process and output the finally selected location set.

Description

Technical Field

[0001] The present invention is a heuristic site selection method based on hybrid index acceleration and multi-source spatial data fusion, which involves multiple fields such as spatial data mining, facility site selection, and combinatorial optimization. Background Art

[0002] With the development of GPS and mobile positioning facilities, a large amount of spatial data such as trajectory data and Point of Interest (POI) data has been generated, thus giving rise to a large number of facility site selection applications based on spatial characteristics. Currently, in applications of site selection based on spatial characteristics, usually only the distance between the facility address and the service object is considered, or only the number of trajectories covered by the facility address is considered as the influence magnitude for site selection. However, simply considering trajectory data for regional facility site selection has certain limitations. It often cannot handle some unreasonable spatial positions well and needs to combine more information for multi-dimensional consideration so that the mined positions can have more fine-grained analyzable indicators and rationality. POI data refers to point-like data in Internet electronic maps, generally including four attributes: name, address, coordinates, and category; it originates from the vector data set of point-like map elements in the basic mapping result DLG (Digital Line Graphic) product.

[0003] Chinese Patent Application No. 201810010827 discloses "Urban Rail Transit Emergency Rescue Station Site Selection Method and Device". It establishes a network model between urban rail transit stations and calculates network-free characteristic indicators. A rescue station site selection model is constructed according to the comprehensive vulnerability index and coverage function. However, it does not utilize spatial data such as trajectory data and POI data, and the site selection modeling is carried out for a static geographical spatial distribution characteristic.

[0004] Chinese Patent Application No. 202111346223 discloses "A Resource Allocation Method, Device and Server for Emergency Supplies". It mainly considers some distance, material attributes, and some probability characteristics of positions in the resource allocation of specific emergency supplies to establish a linear programming equation and then solve it, which is different from the method we proposed in terms of the usage scenario, data source, the mathematical model for solving problems, and the computer technology applied.

[0005] Chinese Patent Application No. 202111272208 discloses "A Method and Device for Site Selection of Electric Vehicle Charging Piles", which mainly solves the problem of low utilization efficiency of charging piles that may be caused by inappropriate site selection of electric charging piles in the prior art. By setting the restriction conditions of the area, one or more addresses are reserved, taking into account the use of charging users and the benefits of charging pile operators. In the process of solving the site selection problem, it mainly considers some factors of specific electric vehicles, does not use trajectory data and POI data to mine influence, and does not use the relevant maximum coverage model for site selection, which is different from the method proposed by us in both application scenarios and problem-solving technologies.

[0006] The present invention provides a heuristic site selection method based on hybrid index acceleration and multi-source data fusion. By combining trajectory data and POI data, using the maximum k-coverage model to heuristically select a set of locations, and at the same time using the hybrid index technology to accelerate the entire mining process, so as to achieve the purpose of spatially characteristic and efficient service facility site selection. Summary of the Invention

[0007] The problem to be solved by the present invention is the site selection problem of mining a set of locations with the greatest influence by combining trajectory data and POI data under a multi-source spatial data set. Further, the purpose of the present invention is also to optimize the efficiency of the above-mentioned site selection problem by combining the hybrid index technology.

[0008] The technical solutions adopted to achieve the technical problems to be solved by the present invention are generally as follows: A heuristic facility location site selection method based on hybrid index acceleration and multi-source data fusion, and the specific method steps include:

[0009] a) For the facility location site selection method, the geographical space is gridified, the geographical space is divided into grids with a fixed side length, and the area where the facility site selection is to be carried out is gridified. The longitude and latitude of the points on the boundary of a region can be obtained from the acquired trajectory data and POI data, and then a maximum covering rectangle is determined using these longitude and latitude, and its area is set as A, and the longitude and latitude of one of the points is selected as the origin; then, according to a preset grid side length L, the entire region is divided into at most A / (L*L) grids, each grid has the distance coordinates of the center point and the side length of the grid, and also has a unique identification number (ID); the divided grids are stored in a set for subsequent location selection.

[0010] b) For the facility location selection method described above, preprocess the trajectory data and POI data, and map the location information to grid coordinates. Convert the longitude and latitude information in the trajectory data and POI data into the distance coordinates corresponding to those described in a). Then segment the trajectory data according to time, and save the segmented trajectory data into a set. Additionally, save the POI data with the converted coordinates into the set. Each POI also contains the category to which the POI belongs. Each processed POI and each processed trajectory are assigned a unique ID.

[0011] c) For the facility location selection method described above, establish an index structure between the trajectory data, POI data, and the grid. Let the set of trajectory data be T. Traverse the set of trajectory data. For each trajectory t in the set T, traverse all the coordinate points p in the trajectory i , find the grid g where p i is located, and establish a two-way index between the trajectory t and the grid g. Let Tg[t] represent the set of grid IDs passed by the trajectory t, and Gt[g] represent the set of trajectory IDs covered by the grid g.

[0012] Similarly, let the set of POI data be P. Traverse the set of POI data. For each point p of the POI, find the grid g where p is located, and establish a one-way index between the grid g and the point of interest p. Let Gp[g] represent the set of categories of POIs contained in the grid g.

[0013] d) For the facility location selection method described above, construct a hybrid index structure based on a balanced tree + hash set, and use a heuristic search strategy to select the location with the highest scoring function value from it;

[0014] First, define a scoring function for a grid:

[0015] IF(g) = λ · (α|TC(g)| · |Is| · k + (1 - α)(dis(g) · |Tr|)).

[0016] Where λ is an adjustment coefficient to ensure that the final value of IF(g) is an integer, α is a weight coefficient, with a value range of [0, 1], set by the user, representing the preference degree for the trajectory data and POI data, k represents the number of locations to be selected, TC(g) represents the number of trajectories passed by the grid g, dis(g) represents the number of types of POIs in the grid g, |Tr| represents the total number of trajectories in the entire region, and |Is| represents the total number of types of POIs in the entire region.

[0017] Then, a hybrid index structure based on balanced tree hashing is designed. The outer layer of the hybrid index is a balanced tree, and each node in the tree represents a hash index structure. Its key value is the IF value, and the corresponding value of the node is the set of IDs of the grids whose IF values are this key value and are inserted into the balanced tree. In the initial state, it is necessary to calculate the IF value of each grid and insert each grid into the hybrid index. After that, each time a grid g is randomly selected from the hash set corresponding to the maximum key value of the hybrid index and added to the corresponding result set. The structure of the hybrid index is as shown in the attached Figure 2 Schematic diagram of the hybrid index structure.

[0018] If the number of selected positions reaches the requirement, stop the process and output the finally selected set of positions.

[0019] e) According to the facility location selection method described above, update the bidirectional index of the hybrid index and the grid trajectory based on the number of currently selected positions. If the number of selected positions reaches the requirement, stop the process and output the finally selected set of positions; if the number of selected positions does not reach the requirement, repeat step d, otherwise stop the process and output the finally selected set of positions.

[0020] For each selected grid, all grids containing any trajectory within this grid need to update the Gt[g] index so that the grid index Gt[g] does not contain the already selected trajectories; at the same time, use the tempSet set to collect those grids that need to be updated in this operation, and use a hash map M to record the current IF value of all grids. The specific method is to first remove the grid from the hybrid index according to the old value of each grid g in the tempSet, then insert it into the hybrid index according to the updated new IF value of the grid g, and update the corresponding hash map M; finally, return to execute step d again.

[0021] The present invention preprocesses multi-source data such as trajectory data and point of interest (POI) data, and then establishes index structures of the trajectory data and the POI data with grid positions respectively; uses a heuristic greedy strategy to select a certain number of positions from the set of grid positions to maximize the scoring function of the set composed of these positions. And uses a hybrid index structure to accelerate the entire selection process. By adopting this method, the problem of facility location selection for location-based services can be efficiently solved from the dimension of multiple spatial data sources.

[0022] Beneficial effects: The method provided by the present invention can effectively perform heuristic location selection from multi-source data, involving facility location selection and data fusion. The technology provided by the present invention can define the influence magnitude of a location by combining multiple spatial data, and then efficiently select a set of locations with approximately optimal influence from them, which is applicable to the problem scenario of facility location selection by combining spatial data and the maximum coverage model. The present invention uses two types of spatial data, namely trajectories and POIs, for mining regional facility locations, and can combine the two dimensions of dynamic trajectories and static POIs, so that the final location selection result has more reference value. At the same time, a hybrid index method is designed to handle the time efficiency problem in the location selection process, which can accelerate the entire facility location selection process. By adopting the technical solution of the present invention, engineers can relatively easily implement the relevant software. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of the heuristic location selection method based on hybrid index acceleration and multi-source data fusion of the present invention;

[0024] Figure 2 Schematic diagram of the hybrid index structure;

[0025] Figure 3 Schematic diagram of the distribution of trajectories and POIs based on grid division. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The following will describe the embodiments of the present invention in combination with specific examples:

[0027] Step 1, first divide the area where facility location selection is to be performed into grids. The longitude and latitude of the boundary points of a region can be obtained from the acquired trajectory data and POI data, and then a maximum covering rectangle is determined using these longitudes and latitudes. Let its area be A, and the longitude and latitude of one of the points be selected as the origin. Then, according to a preset grid side length L, the entire region is divided into at most A / (L*L) grids, and each grid has a unique ID number. The divided grids are stored in a set for subsequent location selection. Assume that the result after division in the given example is a rectangular grid (as shown in the appendix Figure 3 shown).

[0028] Step 2, convert the longitude and latitude coordinates of the trajectories and POIs into distance coordinates, and assign corresponding identifiers to the trajectories and POIs. The curves in the appendix represent trajectories, and the circles represent the categories of POIs.

[0029] Step 3, traverse each coordinate point p in each trajectory T, convert the longitude and latitude coordinates (x, y) into grid coordinates and establish a two-way index with the corresponding grid; the same conversion is also done for each POI. According to the appendixFigure 3 The data in can obtain the grid - related index list as shown in Table 1:

[0030] Table 1

[0031] Grid position ID Set of track IDs contained in the grid Set of POI identifiers contained in the grid 1 {T1} {a, b} 2 {T1} {b} 3 {T4} {a} 4 {} {d} 5 {T1} {c, d} 6 {T3} {a} 7 {T4} {f} 8 {T3} {a, c} 9 {T1} {e} 10 {T2, T3} {b, d} 11 {T3, T4} {a, b, c, d} 12 {T3} {f} 13 {T2} {g} 14 {T2} {b} 15 {T3, T4} {e} 16 {T3} {b}

[0032] Meanwhile, the grid index list passed by the trajectory can be obtained, as shown in Table 2:

[0033] Table 2

[0034]

[0035]

[0036] Step 4:

[0037] First, define a scoring function for a grid:

[0038] IF(g)=λ·(α|TC(g)|·|Is|·k+(1 - α)(dis(g)·|Tr|)).

[0039] Where λ is an adjustment coefficient to ensure that the final value of IF(g) is an integer. α is a weight coefficient, with a value range of [0, 1], set by the user, representing the preference degree for trajectory data and POI data. k represents the number of positions to be selected. TC(g) represents the number of trajectories passing through grid g. dis(g) represents the number of POI types in grid g. |Tr| represents the total number of trajectories in the entire region. |Is| represents the total number of POI types in the entire region.

[0040] In this example, set λ to 100, α to 0.5, and k to 3, indicating that a set with a position number of 3 will be selected. Then, calculate the IF value of each grid position according to the scoring function and insert the grid ID into the hybrid index according to the IF value. For example, the number of trajectories passing through the grid with grid ID 1 (abbreviated as grid 1) is 1, so the TC value of grid 1 is 1; it only contains two POI types, a and b, so its dis value is 2; there are a total of 4 trajectories in the entire region, so |Tr| is 4; the total number of POI types in the entire region is 7, so |Is| is 7. Finally, the IF value of grid 1 is:

[0041] IF(g1)=λ·(α|TC(g1)|·|Is|·k+(1 - α)(dis(g1)·|Tr|))

[0042] =100*(0.5*1*7*3+(1 - 0.5)*2*4)=1450

[0043] Similarly, the initial IF value of each grid can be obtained (as shown in Table 3):

[0044] Table 3

[0045]

[0046]

[0047] Then, the positions of each ID in the hybrid index are shown according to the magnitude of the IF value, and a hash map is used to record the IF value corresponding to each grid, as shown in Table 4:

[0048] Table 4

[0049] IF value Grid set 2900 {11} 2500 {10} 2300 {15} 1450 {1,5,8} 1250 {2,3,6,7,9,12,13,14,16} 200 {4}

[0050] Next, successively select the grid with the largest IF value and add it to the result set:

[0051] For the first time, select grid 11 with the largest IF value, and delete grid 11 from the hybrid index and the grid track index Gt[g]; grid 11 contains two tracks T3 and T4, and the index Tg from track to grid can be used to find which grids have covered any one of T3 and T4; use a temporary set tempSet to store the grids {3, 6, 7, 8, 10, 11, 12, 15, 16} involved in T3 and T4; then traverse each grid in tempSet, and delete all information of T3 and T4 from the grid track index Gt[g], so as to obtain the updated grid track index list, as shown in Table 5:

[0052] Table 5

[0053] Grid position ID Set of track IDs contained in the grid 1 {T1} 2 {T1} 3 {} 4 {} 5 {T1} 6 {} 7 {} 8 {} 9 {T1} 10 {T2} 12 {} 13 {T2} 14 {T2} 15 {} 16 {}

[0054] Then traverse the tempSet set again. For each grid g in the set, first use the hash map M to remove the old value of g from the hybrid index, then calculate the new IF value of the current grid g, and insert g into the hybrid index according to the new IF value. At the same time, record the latest value of grid g in the hash map M. After that, obtain the updated hybrid index, as shown in Table 6:

[0055] Table 6

[0056] IF value Grid set 1450 {1,5,10} 1250 {2,9,13,14} 400 {8} 200 {3,4,6,7,12,15,16}

[0057] Second, select grid 1 with the maximum IF value of 1450, and delete grid 1 from the mixed index and the grid trajectory index Gt[g]; grid 1 contains trajectory T1. Use the trajectory grid index Tg to find the grids {1, 2, 5, 9} passed by T1, and store them using the tempSet set; then traverse each grid in tempSet, and delete the information of T1 from the grid trajectory index, so as to obtain the updated grid trajectory index list, as shown in Table 7:

[0058] Table 7

[0059] Grid position ID Set of track IDs contained in the grid 2 {} 3 {} 4 {} 5 {} 6 {} 7 {} 8 {} 9 {} 10 {T2} 12 {} 13 {T2} 14 {T2} 15 {} 16 {}

[0060] Then traverse the tempSet set again. For each grid g in the set, first use the hash map M to remove the old value of g from the mixed index, then calculate the new IF value of the current grid g, and insert g into the mixed index according to the new IF value. At the same time, record the latest value of grid g in the hash map M; then obtain the updated mixed index, as shown in Table 8:

[0061] Table 8

[0062] IF value Grid set 1450 {10} 1250 {13,14} 400 {8} 200 {2,3,4,5,6,7,9,12,15}

[0063] Third, select grid 10 with the maximum value of 1450 from the mixed index. Since k = 3 grid positions have been selected currently, there is no need to perform subsequent update operations, and directly return the result set, that is, {11, 1, 10}.

[0064] The above has specifically described the disclosed heuristic site selection method based on hybrid index acceleration and multi-source data fusion, which can be understood and implemented by those skilled in the art. Without departing from the scope of the present invention, various modifications can be made to the form and details of the site selection method. Therefore, the above suggested but not limited modifications are all within the scope of the present invention.

Claims

1. A heuristic facility location siting method based on hybrid index acceleration and multi-source data fusion, characterized in that the steps Including: a) Perform grid preprocessing on the geospatial space, dividing the geospatial space into grids with a fixed side length; b) Preprocess the trajectory data and Point of Interest (POI) data, mapping the location information therein to grid coordinates; c) Establish an index structure between the trajectory data and the POI data and the grid; let the trajectory data set be T, traverse the trajectory data set, and for each trajectory t in the set T, traverse all the coordinate points p in the trajectory i , find the grid g where p i is located, and establish a two-way index between the trajectory t and the grid g. Let Tg[t] represent the set of grid IDs passed by the trajectory t, and Gt[g] represent the set of trajectory IDs covered by the grid g; Similarly, let the set of POI data be P. Traverse the set of POI data. For each point p of a POI, find the grid g where p is located, establish a one-way index between grid g and POI p, and let Gp[g] represent the set of categories of POIs contained in grid g; d) Construct a hybrid index structure based on a balanced tree + hash set, and use a heuristic search strategy therein to select the location with the highest scoring function value; First, define a scoring function for a grid: ， where λ is an adjustment coefficient to ensure that the final value of IF(g), i.e., the IF value, is an integer, α is a weight coefficient with a value range of [0, 1], set by the user, representing the preference degree for trajectory data and POI data, k represents the number of locations to be selected, TC(g) represents the number of trajectories passing through grid g, dis(g) represents the number of types of POIs in grid g, |Tr| represents the total number of trajectories in the entire region, and |Is| represents the total number of types of POIs in the entire region; Then design a hybrid index structure based on a balanced tree hash. The outer layer of the hybrid index is a balanced tree, and each node in the tree represents a hash index structure. Its key value is the IF value, and the corresponding value of the node is the set of IDs of the grids whose IF values are this key value and are inserted into the balanced tree; In the initial state, it is necessary to calculate the IF value of each grid and insert each grid into the hybrid index; After that, each time randomly select a grid g from the hash set corresponding to the maximum key value of the hybrid index and add it to the corresponding result set; e) Select and update the two-way index of the hybrid index and grid trajectories according to the currently selected number of locations; If the selected number of locations does not meet the requirements, repeat step d), otherwise stop the process and output the finally selected set of locations.

2. The facility location selection method according to claim 1, characterized in that Among them, step a): Perform grid division on the area for facility location; Obtain the longitude and latitude of the boundary points of an area from the acquired trajectory data and POI data, then use these longitude and latitudes to determine a maximum covering rectangle, set its area as A, and select the longitude and latitude of one of the points as the origin; Then, according to a pre-set grid side length L, divide the entire area into at most A / (L*L) grids. Each grid has the distance coordinates of the center point and the side length size of the grid, and also has a unique identification number ID; Store the divided grids in a set for subsequent location selection.

3. The facility location selection method according to claim 2, characterized in that, Among them, step b): Convert the longitude and latitude information in the trajectory data and POI data into the distance coordinates corresponding to those described in step a, then segment the trajectory data according to time, and save the segmented trajectory data in a set; In addition, save the POI data with the converted coordinates in the set, and each POI also contains the category to which the POI belongs; Each processed POI and each processed trajectory are assigned a unique ID.

4. The facility location selection method according to claim 1, characterized in that, Among them, step e): If the number of selected positions reaches the requirement, stop the process and output the set of finally selected positions; if the number of selected positions does not reach the requirement, repeat step d, otherwise stop the process and output the set of finally selected positions; for each selected grid, all grids containing any trajectory within this grid need to update the Gt[g] index so that the grid index Gt[g] does not contain the already selected trajectories; at the same time, use the tempSet set to collect those grids that need to be updated during this operation, and use a hash map M to record the current IF value of all grids; first, remove each grid g from the hybrid index according to its old value in tempSet, then according to the updated new IF value of this grid g, insert it into the hybrid index, and update the corresponding hash map M; Finally, return to execute step d again.

Images