A multi-objective sequential path planning method based on dominance relationship pruning

Abstract

The present invention discloses a multi-objective sequenced path planning method based on dominance relationship pruning. First, a city model centered on a location is constructed, and the location, type and score of the location are extracted from a city point of interest data set. In the city model, the distance between any two locations is calculated and saved in a distance matrix. Secondly, the constraints of the sequenced path are determined, and the user provides the path planning constraints and the multi-objective path evaluation function. Finally, the dominance relationship pruning algorithm is used to find a location sequence in the city model to form an optimal path that meets the user's needs.

Description

Technical Field

[0001] The invention relates to a multi-objective sequencing path planning method based on dominance relationship pruning. Background Art

[0002] With the popularization of mobile Internet and smart phones, navigation services based on path planning have gradually become an indispensable part of people's daily lives. The main goal of traditional path planning is to help users find the best route from the starting point to the end point, which greatly facilitates people's travel. With the rapid development of new network applications such as the Internet of Things technology, path planning based purely on the starting point and the end point is difficult to meet the increasingly complex personalized needs and preferences of users in modern society. For example, users may propose a path planning requirement that passes through multiple points of interest (POI) in sequence while minimizing the total cost (distance and time). At the same time, with the rapid popularization and development of mobile smart terminals and applications, users are exposed to more and more descriptive information such as POI ratings. Therefore, in addition to the type dimension of POI, the quality dimension has also become a key factor for users to consider.

[0003] In recent years, the academic community has conducted in-depth research on the core theme of personalized path planning, and has produced research results in the areas of Optimal Sequenced Route (OSR) planning, trajectory-based path recommendation, keyword-aware path query, and constrained shortest path (CSP). Aiming at different application scenarios in various complex environments, and considering users' requirements in terms of the type and quality of points of interest, path cost, etc., multi-objective optimal sequenced path planning can more flexibly and closely fit user preferences. Optimal sequenced path planning is a key technology in the field of personalized path planning. It can support users to define the type information (points of interest) and the order of visits of the passing places according to their actual needs, and then plan the optimal path for them. If there are nodes containing points of interest in the graph, this type of technology can pass through at least one candidate point of each given type in sequence and find the optimal path.

[0004] Optimal ordered path planning can use methods such as multi-criteria path planning, path skyline search, and constrained shortest path query. However, the above methods all have certain limitations when applied. Multi-criteria path planning aggregates different weights with a specific function, which can transform multiple objectives into a single-objective optimization problem, so the computational overhead is relatively small. However, the specific aggregation function is difficult to adapt to the combination of optimization objectives in different application scenarios, so the query versatility of this method is poor. Constrained shortest path query usually finds the best path based on one optimization objective, while requiring other indicators to meet predefined constraints. This type of query requires the user to provide a constraint threshold, so the user is required to have domain knowledge and know the path measurement indicators in advance, otherwise it may not be possible to find a suitable path, and the calculation is huge. Path skyline search considers multi-objective optimization based on dominance relations and returns Pareto sets that cannot dominate each other on all optimization objectives. However, this process may generate a large number of invalid paths, which need to be screened and cannot be sorted, so the calculation is also very large.

[0005] Therefore, it is necessary to study a method with good query versatility, no need for domain knowledge and a certain degree of optimization of computational complexity to overcome the above limitations and solve the problem of multi-objective optimal sequencing path planning. Summary of the invention

[0006] Purpose of the invention: In view of the problems existing in the above-mentioned prior art, the present invention proposes a multi-objective sequencing path planning method based on dominating relationship pruning, focusing on solving the problem of multi-objective optimal sequencing path planning. By constructing a location-centered city model, determining the constraints of the sequencing path, and using the dominating relationship pruning algorithm, the optimal path that meets user needs is constructed.

[0007] Technical solution: To achieve the purpose of the present invention, the technical solution adopted by the present invention is: a multi-objective sequencing path planning method based on dominance relationship pruning, the method comprising the following steps:

[0008] Step 1. Build a location-centric city model and extract the location, type, and score of the location from the city point of interest dataset. Calculate the distance between any two locations and save it in the distance matrix;

[0009] Step 2. Calculate the path that meets the user's needs, determine the constraints of the sequenced path, the user gives the path planning constraints and the multi-objective path evaluation function;

[0010] Step 3. Use the dominance relationship pruning algorithm to find the location sequence P in the city model. <p 0 ,p 1 ,…,p n >Construct an optimal path to meet user needs.

[0011] Furthermore, in step 1, a location-centric city model is constructed. The specific steps are as follows:

[0012] Step 101. Extract the location, type, and score of a place from the city interest point dataset. A set of places V is obtained from the city interest point dataset. Each place v∈V has a type type(v), a score score(v), and latitude and longitude coordinates los(v).

[0013] Step 102. Calculate the distance between any two locations and save it in the distance matrix. i The coordinates of loc(v i )=(x i ,y i ), then location v i and v j The distance is calculated by the Euclidean distance formula Construct a two-dimensional array as a distance matrix, and the calculation results are saved in the i-th row and j-th column of the distance matrix dis.

[0014] Through the above mathematical description and formula expression, urban locations can be modeled. The model contains the location, type and score information of the location. On this basis, the distance between any two points is calculated to form a distance matrix.

[0015] In step 2, the path that meets the user's needs is calculated. The specific steps are:

[0016] Step 201. The user sets the path planning constraints, starting location s∈V, end location t∈V, and the type sequence of the passing locations C=*c 1 ,c 2 ,…,c n-1 +.

[0017] Step 202. The user gives a multi-objective path evaluation function, and denotes a path as P= <p 0 ,p 1 ,…,p n >, first of all, we need to ensure that p 0 =s,p n = t, and type(p i )=c i , otherwise the cost of path P is positive infinity. Then, let represents the total distance, and Represents the total score, then the evaluation function The two optimization objectives of distance and score are integrated into one optimization objective, and the multi-objective optimization problem is transformed into a single-objective optimization problem.

[0018] Through the above mathematical description and formula expression, the path that meets the user's needs can be calculated by setting path planning constraints and giving a multi-objective path evaluation function.

[0019] In step 3, the dominance relationship pruning algorithm is used. The specific steps are as follows:

[0020] Step 301. First, initialize the result set Ψ, assign the upper bound upper to the maximum value, and then start the iterative search.

[0021] Step 302. Construct a multi-segment graph. First, calculate the sum of the distances from location p to the starting point s and the end point t, i.e., dis(p, s) + dis(p, t). Then, select k locations with the smallest sum of distances as candidate locations for each type that the user needs to visit. Then, take the starting point s, the end point t, and all the candidate locations as the vertices of the multi-segment graph. The starting point s is the first layer of the multi-segment graph, and the type c is the first layer of the multi-segment graph. i The candidate vertex of is the i-th layer of the multi-segment graph, and the end point t is the last layer of the multi-segment graph. There is an edge connecting every two vertices in two adjacent layers, and the weight of the edge is the distance between the corresponding locations.

[0022] Step 303: Update the memorandum matrix DM to store the optimal sub-path length and reactivate the path previously stored in the candidate path set. The memorandum matrix DM can be calculated using dynamic programming. i ] indicates arrival at any type c i A node v i The minimum distance.

[0023] Step 304: Expand and construct feasible paths layer by layer. Construct the current path cr based on the previous path pr, and calculate the distance D, score P, global lower bound GLB, local lower bound LLB, etc. of the new path. i , the path size is i, satisfying type C i-1 The subpaths of type C are all preceded by pr; i , the path size is i+1, which satisfies type C i The sub-paths of are all the current path cr. The global lower bound GLB is divided into two categories: the global lower bound of the node and the global lower bound of the sub-path. The node global lower bound GLB (v i ) means starting from the starting point s and going around to the current node v i Then go directly to the end point t, considering the current node score q(v i ) and the cumulative value Q of the highest score of other types gm : Subpath global lower bound Indicates the end node v of the current subpath iStart and go directly to the end point t, considering the current score of the subpath and the cumulative value Q of the optimal score of all subsequent types bm : The local lower bound LLB means that starting from the end node of the subpath, the path score of reaching the end point by sequentially passing through a series of candidate nodes of subsequent types considering only the distance factor is:

[0024] Step 305: Prune according to the subpath dominance relationship. <c 1 ,c 2 ,...,c m > and two subpaths of the same size and 1≤m≤M. If And satisfy D(SR 1 )≤D(SR 2 )∧Q(SR 1 )>Q(SR 2 ) or D(SR 1 ) <D(SR 2 )∧Q(SR 1 )≥Q(SR 2 ), then it is called subpath SR 1 Dominant subpath SR 2 , denoted as SR 1 >SR 2 . Moreover, the dominance relationship is transitive: Let upper be the optimal path currently found. If GLB(cr) exceeds upper, prune directly and mark the next layer of path as invalid; if LLB(cr) exceeds upper, temporarily abandon it, wait for activation, and mark the next layer of path as abandoned; otherwise, perform dominance relationship detection, if it fails, prune, and mark the next layer of path as invalid. In addition, there are naturally valid paths that pass the above detection, which will be stored in the next layer of path set, or updated to the result set Ψ at the end layer.

[0025] Step 306: Repeat steps 302-305 until the optimal path is found.

[0026] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0027] (1) It supports users’ path planning needs that include multiple POI preferences. It comprehensively considers dimensions such as path cost, POI type, and POI quality, and can plan the optimal path based on multiple objectives to meet users’ increasingly complex personalized needs and preferences.

[0028] (2) In view of the problems of existing multi-objective optimization path planning methods, such as poor versatility, small computational complexity, requirement of domain knowledge reserve, and lack of support for sorting, a path planning method with higher versatility, optimizable computational complexity, no need for domain knowledge reserve, and support for sorting is designed. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flowchart of multi-objective sequential path planning method based on dominance relationship pruning DETAILED DESCRIPTION

[0030] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0031] The goal of the present invention is to design a multi-objective sequencing path planning method that comprehensively considers path costs and user points of interest to meet the user's comprehensive requirements in terms of point of interest type, service quality, distance and time.

[0032] The present invention designs a multi-objective sequencing path planning method based on dominating relationship pruning from the perspectives of city model construction, determination of sequencing path constraint conditions and optimization of dominating relationship pruning algorithm. To this end, the present invention specifically performs the following steps:

[0033] Step 1. Build a location-centric city model and extract the location, type, and score of the location from the city point of interest dataset. Calculate the distance between any two locations and save it in the distance matrix. The specific steps are:

[0034] Step 101. Extract the location, type, and score of a place from the city interest point dataset. A set of places V is obtained from the city interest point dataset. Each place v∈V has a type type(v), a score score(v), and a latitude and longitude coordinate loc(v).

[0035] Step 102. Calculate the distance between any two locations and save it in the distance matrix. i The coordinates of loc(v i )=(x i ,y i ), then location v i and v j The distance is calculated by the Euclidean distance formula Construct a two-dimensional array as a distance matrix, and the calculation results are saved in the i-th row and j-th column of the distance matrix dis.

[0036] Through the above mathematical description and formula expression, urban locations can be modeled. The model contains the location, type and score information of the location. On this basis, the distance between any two points is calculated to form a distance matrix.

[0037] Step 2. Determine the constraints of the sequenced path. The user gives the path planning constraints and the multi-objective path evaluation function. The specific steps are:

[0038] Step 201. The user sets the path planning constraints, starting location s∈V, end location t∈V, and the type sequence of the passing locations C=*c 1 ,c 2 ,...,c n-1 +.

[0039] Step 202. The user gives a multi-objective path evaluation function, and denotes a path as P= <p 0 ,p 1 ,...,p n >, first of all, we need to ensure that p 0 =s,p n = t, and type(p i )=c i , otherwise the cost of path P is positive infinity. Then, let represents the total distance, and Represents the total score, then the evaluation function The two optimization objectives of distance and score are integrated into one optimization objective, and the multi-objective optimization problem is transformed into a single-objective optimization problem.

[0040] Through the above mathematical description and formula expression, the path that meets the user's needs can be calculated by setting path planning constraints and giving a multi-objective path evaluation function.

[0041] Step 3. Use the dominance relationship pruning algorithm to find the location sequence P in the city model. <p 0 ,p 1 ,…,p n >Construct an optimal path that meets user needs. The specific steps are as follows:

[0042] Step 301. First, initialize the result set Ψ, assign the upper bound upper to the maximum value, and then start the iterative search.

[0043] Step 302: Construct a multi-segment graph. First, calculate the sum of the distances from location p to the starting point s and the end point t, i.e., dis(p,s)

[0044] dis(p,t). Then, for each type that the user needs to visit, select k locations with the smallest sum of distances as candidate locations. Then, take the starting point s, the end point t, and all candidate locations as vertices of the multi-segment graph. The starting point s is the first layer of the multi-segment graph, and the type c i The candidate vertex of is the i-th layer of the multi-segment graph, and the end point t is the last layer of the multi-segment graph. There is an edge connecting every two vertices in two adjacent layers, and the weight of the edge is the distance between the corresponding locations.

[0045] Step 303: Update the memorandum matrix DM to store the optimal sub-path length and reactivate the path previously stored in the candidate path set. The memorandum matrix DM can be calculated using dynamic programming. i ] indicates arrival at any type c i A node v i The minimum distance.

[0046] Step 304: Expand and construct feasible paths layer by layer. Construct the current path cr based on the previous path pr, and calculate the distance D, score P, global lower bound GLB, local lower bound LLB, etc. of the new path. i , the path size is i, satisfying type C i-1 The subpaths of type C are all preceded by pr; i , the path size is i+1, which satisfies type C i The sub-paths of are all the current path cr. The global lower bound GLB is divided into two categories: the global lower bound of the node and the global lower bound of the sub-path. The node global lower bound GLB (v i ) means starting from the starting point s and going around to the current node v i Then go directly to the end point t, considering the current node score q(v i ) and the cumulative value Q of the highest score of other types gm : Subpath global lower bound Indicates the end node v of the current subpath i Start and go directly to the end point t, considering the current score of the subpath and the cumulative value Q of the optimal score of all subsequent types bm : The local lower bound LLB means that starting from the end node of the subpath, the path score of reaching the end point by sequentially passing through a series of candidate nodes of subsequent types considering only the distance factor is:

[0047] Step 305: Prune according to the subpath dominance relationship. <c 1 ,c 2 ,...,c m > and two subpaths of the same size and 1≤m≤M. If And satisfy D(SR 1 )≤D(SR 2 )∧Q(SR 1 )>Q(SR 2 ) or D(SR 1 ) <D(SR 2 )∧Q(SR1 )≥Q(SR 2 ), then it is called subpath SR 1 Dominant subpath SR 2 , denoted as SR 1 >SR 2 . Moreover, the dominance relationship is transitive: Let upper be the optimal path currently found. If GLB(cr) exceeds upper, prune directly and mark the next layer of path as invalid; if LLB(cr) exceeds upper, temporarily abandon it, wait for activation, and mark the next layer of path as abandoned; otherwise, perform dominance relationship detection, if it fails, prune, and mark the next layer of path as invalid. In addition, there are naturally valid paths that pass the above detection, which will be stored in the next layer of path set, or updated to the result set Ψ at the end layer.

[0048] Step 306: Repeat steps 302-305 until the optimal path is found.

Claims

1. A multi-objective sequential path planning method based on dominance relationship pruning, characterized in that: The method comprises the following steps: Step 1. Build a location-centric city model, extract the location, type, and score of the location from the city point of interest dataset; calculate the distance between any two locations and save it in the distance matrix; Step 2. Calculate the path that meets the user's needs, determine the constraints of the sequenced path, the user gives the path planning constraints and the multi-objective path evaluation function; Step 3. Use the dominance relationship pruning algorithm to find the location sequence P in the city model. <p0,p1,…,p n >Construct an optimal path that meets user needs; use the dominant relationship pruning algorithm, the specific steps are: Step 301. First, initialize the result set Ψ, assign the upper bound upper to the maximum value, and then start the iterative search; Step 302: construct a multi-segment graph. First, calculate the sum of the distances from location p to the starting point s and the end point t, that is, dis(p,s)+dis(p,t); Then, for each type that the user needs to visit, select k locations with the smallest sum of distances as candidate locations; Then, the starting point s, the end point t, and all candidate locations are taken as vertices of the multi-segment graph. The starting point s is the first layer of the multi-segment graph, and the type c i The candidate vertices of are the i-th layer of the multi-segment graph, and the end point t is the last layer of the multi-segment graph; there are edges connecting the vertices of two adjacent layers, and the weight of the edge is the distance between the corresponding locations; Step 303: Update the memorandum matrix DM to store the optimal sub-path length and reactivate the path previously stored in the candidate path set; the memorandum matrix DM is calculated using dynamic programming, DM[i][v i ] indicates arrival at any type c i A node v i The minimum distance; Step 304: Expand and construct feasible paths layer by layer; construct the current path cr based on the previous path pr, and calculate the distance D, score P, global lower bound GLB, and local lower bound LLB of the new path; For Type C i , the path size is i, satisfying type C i-1 The subpaths of are all the preceding paths pr; For Type C i , the path size is i+1, which satisfies type C i The sub-paths are all the current path cr; The global lower bound GLB is divided into two categories: the global lower bound of the node and the global lower bound of the subpath. The node global lower bound GLB (v i ) means starting from the starting point s and going around to the current node v i Then go directly to the end point t, considering the current node score q(v i ) and the cumulative value Q of the highest score of other types gm : Among them, d(s,v i ) means from location s to v i The distance, d(v i ,t) indicates that from v i The distance to t; Subpath global lower bound Indicates the end node v of the current subpath i Start and go directly to the end point t, considering the current score of the subpath and the cumulative value Q of the optimal score of all subsequent types bm : in, Indicates the path Length, Indicates the path score; The local lower bound LLB means that starting from the end node of the subpath, the path score of reaching the end point by sequentially passing through a series of candidate nodes of subsequent types considering only the distance factor is: Among them, DM[i][v i ] indicates arrival at any type c i A node v i The minimum distance; Step 305. Prune according to the sub-path domination relationship; given a type sequence C = <c1, c2,..., c m > and two sub-paths of the same size and 1 ≤ m ≤ M; if it satisfies and satisfies D(SR1) ≤ D(SR2) ∧ Q(SR1) > Q(SR2) or D(SR1) < D(SR2) ∧ Q(SR1) ≥ Q(SR2), then it is said that the sub-path SR1 dominates the sub-path SR2, denoted as Moreover, the domination relationship has transitivity: Let upper be the currently found optimal path. If it is found that GLB(cr) exceeds upper, directly prune and mark the next-layer path as invalid; if LLB(cr) exceeds upper, temporarily discard it and wait for activation, and mark the next-layer path as discarded; otherwise, perform domination relationship detection, and prune if it fails, and also mark the next-layer path as invalid; in addition, there are naturally valid paths that pass the above detection. These paths will be stored in the next-layer path set or updated to the result set Ψ at the end layer; Step 306: Repeat steps 302-305 until only one path remains in the result set Ψ, and return this path as the optimal path.

2. A multi-objective sequential path planning method based on dominance relationship pruning according to claim 1, characterized in that: In step 1, a location-centric city model is constructed. The specific steps are as follows: Step 101. Extract the location, type, and score of the place from the city interest point dataset; obtain a set V of places from the city interest point dataset; each place v∈V has a type type(v), a score score(v), and a latitude and longitude coordinate loc(v); Step 102. Calculate the distance between any two locations and save it in the distance matrix; record the location V i The coordinates of loc(v i )=(x i ,y i ), then location v i and v j The distance is calculated by the Euclidean distance formula Construct a two-dimensional array as the distance matrix dis, and the calculation results are saved in the i-th row and j-th column of the distance matrix dis.

3. The multi-objective sequential path planning method based on dominance relationship pruning according to claim 1 is characterized in that: In step 2, the path that meets the user's needs is calculated. The specific steps are: Step 201. The user sets the path planning constraints, the starting location s∈V, the end location t∈V, and the type sequence of the passing locations C={c1,c2,…,c n-1 }; Step 202. The user gives a multi-objective path evaluation function, and denotes a path as P= <p0,p1,…,p n >, first, we need to ensure that p0 = s, p n = t, and type(p i )=c i , otherwise the cost of path P is positive infinity; then, let represents the total distance, and Represents the total score, then the evaluation function The two optimization objectives of distance and score are integrated into one optimization objective, and the multi-objective optimization problem is transformed into a single-objective optimization problem.

Images