Routing-optimized graph with dynamically updated Function Road Class values

Abstract

The invention relates to a method for adapting a route planning system, wherein a traffic network is modeled by a graph having edges and nodes, wherein weights and a respective parameter L and a respective parameter S are assigned to the edges, wherein the weights are dynamically adapted to current conditions of the traffic route by being initialized (S1) and replaced by an updated weight based on a quotient L / S (S2), wherein a Dijkstra algorithm is executed for route planning and the parameter S is incremented by one when the edge associated with the respective parameter S is visited by the Dijkstra algorithm, and wherein the parameter L is incremented by one when the edge associated with the respective parameter L is used by the Dijkstra algorithm as part of the route found.

Description

[0001] The invention relates to a method for adapting a route planning system, as well as a corresponding route planning system.

[0002] The most common optimization criterion for route calculations, especially for vehicles, is the fastest route between the start and destination. This is typically calculated using a type of Dijkstra search, such as bidirectional searches, A*, or similar methods, on a graph. This means that reducing the size of the graph directly leads to runtime optimization. Roads are typically categorized into so-called FRC scores (FRC stands for "Function Road Class"). The lower the FRC score of a road, the more important and suitable it is for route planning. These FRC scores are typically represented by weights on the edges of a graph. The weight values ​​can vary significantly depending on the map provider and are primarily dependent on the type of road, e.g., FRC = 1 for a highway or FRC = 5 for a dirt road. Route planning then typically proceeds as follows: - Start a bidirectional Dijkstra algorithm from start and target; - Avoid switching from a better (especially lower) FRC to a higher and therefore worse FRC.

[0003] This significantly reduces the number of edges and, indirectly, the number of nodes for the search, thus considerably shortening the execution time of a Dijkstra algorithm. However, due to the FRC reduction of the graph, optimality of the Dijkstra search can no longer be guaranteed. FRC creation is not necessarily based on the property "is part of a very fast route." Furthermore, the property "is part of a very fast route" is also variable, as, for example, roadworks can make a previously very fast section of highway significantly less suitable, making it more sensible to exit (switch to a less efficient FRC). However, this is often overlooked during route planning due to the FRC. Moreover, the FRC is only one parameter, which, in case of doubt, is only optimized for a Dijkstra search with respect to the shortest route.If it is desired to filter the roads according to criteria that, for example, contribute to a route with little elevation gain, this classification would have to be carried out again, as the previous weighting values ​​may be unsuitable for the simple or combined optimization criterion. Therefore, the weighting values ​​would have to be recalculated for each optimization criterion (speed, elevation gain, scenic views, etc.).

[0004] US 2015 / 0149078 A1 concerns a one-to-many route search method that includes: searching map data to select connections that form a multitude of routes from a departure node according to a cost function; and determining the extent of each route according to an objective function, the search in the map data being performed up to a search limit based on the cost of an unreachable node.

[0005] US Patent 2010 / 0228472 A1 concerns a method for determining a route from a starting point to a destination in a road network using an estimator function for multiple nodes of the road network. The estimator function provides a lower bound for the cost of a route connecting a node of the road network to the destination. The method includes: defining a tiling that covers an area containing at least part of the road network; determining a resistance value for each tile of the tiling; and determining an estimator function value for a node of a tile located on a tile boundary, depending on the resistance values ​​of the tiles of the tiling.The resistance value of a given tile represents the costs associated with routes connecting nodes of the road network at a boundary of the given tile, such that a lower bound for the costs of routes connecting nodes at the boundary of the given tile can be derived from the resistance value of the given tile.

[0006] Furthermore, DE 10 2009 045 039 A1 discloses a navigation device for searching for a route from a departure point to a destination, comprising: an information storage unit that stores map data in a hierarchical structure, excluding the map data of at least one hierarchy level, and a route search unit for reading the map data in the hierarchical structure from the information storage unit, for performing a route search in each predetermined area on a departure point side and a destination side, and for determining the shortest cost route in an area overlapping the search areas. DE 10 2008 042 748 A1 describes a method for route calculation taking into account utilization rates.German patent DE 10 2006 030 271 A1 describes a navigation system with a route calculation unit for providing a navigation route from a starting point to a destination based on an evaluation of a plurality of cost parameters by an evaluation function. German patent DE 10 2010 040 587 A1 relates to a navigation system and a method for calculating the total cost of a route.

[0007] In the known prior art, static weighting values ​​used for route planning typically have to be laboriously created manually with respect to one or more criteria and are not adapted to new circumstances. This leads to poorer route planning results and may require constant maintenance. The object of the invention is to solve this problem and to implement an improved use of weighting values ​​in route planning.

[0008] The invention is defined by the features of the independent claims. Advantageous further developments and embodiments are the subject of the dependent claims.

[0009] A first aspect of the invention relates to a method for adapting a route planning system, wherein a road network is modeled by a graph having edges and nodes, wherein weights and a respective parameter L and a respective parameter S are assigned to the edges, wherein each of the weights describes a respective attribute of a road in a road network and is selected from a predetermined finite set, wherein the weights are dynamically adapted to current conditions of the road by initializing the weights with a respective predetermined value, and the respective initialized value is replaced by an updated weight based on a quotient L / S, wherein a Dijkstra algorithm is executed for route planning and the parameter S is incremented by one when the edge associated with the respective parameter S is visited by the Dijkstra algorithm.and where the parameter L is incremented by one if the edge associated with the respective parameter L is used by Dijkstra's algorithm as part of the route found.

[0010] If the route planning system is adapted as described, a new route planning request can be executed using the adapted route planning system, which in turn leads to an adaptation of the route planning system, and so on, so that the adaptation of the route planning system and its use for route planning occur alternately.

[0011] This provides an algorithm that dynamically generates weight values ​​for a specific type of Dijkstra query and automatically adjusts them to new circumstances. The "type of Dijkstra query" refers to the optimization function with one or more optimization criteria, such as shortest route, lowest gradient, lowest fuel consumption, good network coverage, or others.

[0012] Each weight preferentially specifies an FRC value, and thus a road classification value. First, each edge is preferentially initialized with the least favorable value from the finite set of possible weights. Then, normal routing queries are performed depending on one or more optimization criteria. Each edge holds two additional parameters: S := part of the search space and L := part of the solution. If the Dijkstra algorithm visits an edge, S is incremented by one. If this edge is part of the resulting path, L is incremented by one.

[0013] Furthermore, a specific behavior is implemented whereby Dijkstra's algorithm does not switch from a good FRC to a worse one. This means that edges that are actually part of the solution, in the sense of a found route, are favored in the FRC classification, while edges that were considered but not part of the solution are penalized. The L / S ratio is continuously updated, particularly with each invocation of Dijkstra's algorithm for a new route search, allowing the FRC classes to be dynamically modified. Thus, for example, roadworks on the highway—which reduce speed—are implicitly factored into the FRC class.

[0014] Insbesondere gilt S ≥ L und 0 ≤ L / S ≤ 1 (für S > 0). Je öfter eine Kante tatsächlich an der Lösung der Routenplanung, dh der geplanten Route, beteiligt ist, sprich je größer L ist, desto größer ist das Verhältnis L / S. Somit gilt: je größer L / S desto besser soll der Wert der Gewichtung sein.

[0015] It is advantageous if each subgraph G_i := {e ∈ E: weight(e) ≤ i}, with 1 ≤ i ≤ 5, consists of only one connected component. This ensures that there is a possible route from every node to every other node within an FRC class. However, this is not necessarily the case with the approach described above, and an FRC subgraph may consist of multiple connected components. To generate a single-component graph from G_i, Kruskal's algorithm (Minimal Spanning Tree) is preferably used to generate a minimum spanning tree.

[0016] The method according to the first aspect of the invention presents an algorithm that automatically initializes and dynamically updates edge weights, such as the Functional Road Class (FRC) values ​​of a road network modeled as a graph. The weight values ​​are automatically adjusted to the optimization criterion, such as one of the following objectives: shortest route, fastest route, or most fuel-efficient route. In contrast, conventional methods require manually initializing weight values, which then remain static, resulting in limited flexibility and significant manual effort. The problem of static edge weight values ​​is solved by a dynamic, self-regulating approach. This approach offers the advantages that the weight values ​​do not need to be manually set initially and are dynamically updated during route planning.The proposed algorithm increases flexibility and reduces manual workload by continuously adjusting the weighting values ​​in response to changing optimization criteria, thereby improving the efficiency and adaptability of route planning systems.

[0017] According to an advantageous embodiment, the respective predetermined value for initializing the weights is the least favorable value from the predetermined finite set of possible weights.

[0018] According to a further advantageous embodiment, a respective updated weighting is determined from a respective quotient L / S by an invariant predetermined rule.

[0019] The mapping of a quotient L / S to an updated weighting therefore occurs with a fixed, predefined scale, which can be specified by a table, for example, L / S > 0.5 → weighting = 1, L / S > 0.4 → weighting = 2, L / S > 0.3 → weighting = 3, L / S > 0.2 → weighting = 4, L / S otherwise → weighting = 5. This automatically increases the weighting of roads that are frequently part of the route found by Dijkstra's algorithm. However, a problem arises because, if a graph with weighted edges is present, edges with a less favorable weighting are no longer considered if an adjacent road represented as an edge has a less favorable weighting, since Dijkstra's algorithm does not switch to roads with less favorable weightings. Thus, dynamically updating L and S is difficult, as it is hard to deviate from an existing classification.

[0020] According to a further advantageous embodiment, a respective updated weighting is determined from a respective L / S quotient by a hierarchical classification from the best L / S quotients to the worst L / S quotients.

[0021] This involves mapping an L / S ratio to an updated weighting based on a classification from the best to the worst L / S ratios. For example, this means that the edges with the best 20% L / S values ​​are classified with a weighting of FRC=1 (represented by the FRC value), the next 20% with FRC=2, and so on.

[0022] The fundamental behavior of Dijkstra's algorithm using weights is that a search tree does not switch from an edge with a weight of k to a neighboring edge with a weight greater than k. To put it simply: once a fast road (e.g., a highway) is planned for the route, it should not switch to a slower country road. Once on a highway (for example, with FRC=1), the route planning remains on the highway. Once on a country road (for example, with FRC=3), the route planning does not switch to a dirt road (for example, with FRC=5). This means that other, potentially better, route options are not considered, even if, for example, leaving a highway with FRC=1 for a country road with FRC=3 would lead to better achievement of one or more combined optimization criteria.While this achieves the desired reduction in computation time, and is sensible when using fixed weights, applying this behavior unchanged to the dynamic weight adjustment procedure described above would, as mentioned above, eliminate the possibility in some cases of choosing the now-better road. Therefore, for a certain number of routing queries, it is necessary to use road-representing edges with less favorable weights so that these roads can be included in the search space and thus in the potential solution space of Dijkstra's algorithm. This results in the L / S ratio changing accordingly, allowing the weights to be adjusted dynamically. To avoid this issue of using road-representing edges with less favorable weights, such as the one mentioned above...To still allow for higher FRC values ​​during route search, further parameters are introduced in the following embodiment.

[0023] According to a further advantageous embodiment, a random value p with 0 is generated before each route planning step performed by the Dijkstra algorithm. <p<1 ermittelt, wobei geprüft wird, in welchem Bereich einer vorgegebenen Verteilungsfunktion mit Wahrscheinlichkeiten P_i der Zufallswert liegt, und abhängig vom Bereich mit dem Zufallswert bestimmt, bis zu welchem Wert einer Gewichtung von einem Wechselverbot im Dijkstra-Algorithmus zu einer ungünstiger gewichteten Kante im Graph abgewichen werden kann.

[0024] Before each route request, a random value of 0 is used. <p < 1 ermittelt. Dieser Wert bestimmt, bis zu welchem Wert einer Gewichtung weitere Straßen repräsentierende Kanten mit ungünstigerer Gewichtung, insbesondere mit höherem FRC-Wert, in der Routenplanung berücksichtigt werden. Die Zufallszahl p ist aus dem Bereich [0, 1] zufällig gewählt. In einer vorgegebenen Verteilungsfunktion sind den Gewichtungen zugeordnete Wahrscheinlichkeiten P_i vorgegeben, beispielsweise die Wahrscheinlichkeiten 0,8, 0,9, 0,95, 0,98, 1. Es wird geprüft, in welchem Bereich der Verteilungsfunktion der Zufallswert p liegt. Ist p beispielsweise 0,93, dann liegt p zwischen P_2=0,9 und P_3=0,95. Somit sind folgende Wechsel in der Route auf Straßen mit Gewichtungen, die durch FRC-Werte ausgeprägt sind, während dieser Suchanfrage erlaubt: Kanten mit... [FRC=1 → FRC=2], [FRC 1 → FRC 3], [FRC 2 → FRC 3]. Anders ausgedrückt werden die Kanten mit FRC 1 bis 3 für diese Suchanfrage gleichgesetzt.For example, the path [highway with FRC = 1 → rural road with FRC = 3 → highway with FRC = 1] is considered, and if this is indeed classified as a better route option, the rural road can be upgraded (FRC 3 to FRC 2) and the highway section downgraded (FRC 1 to FRC 2). Analogous to the mapping of the L / S ratio to the weighting, the procedure can be applied here in a modular fashion.

[0025] As an example, consider a method using fixed values: The fixed values ​​can be found in a table. For instance, if p is less than 0.8, the weighting restriction is not relaxed. If, for example, the value is between 0.8 and 0.9, it is always possible to switch to a road with FRC=2, and so on for the values ​​0.95, 0.98, and 1, where a value of 1 corresponds to an FRC value of 5 for a possible switch. Clearly, the more often unfavorable and therefore high weights are included in the search of Dijkstra's algorithm, the more accurate the dynamic updating of the weights becomes; however, this increases the average route calculation time. Furthermore, it is possible to use a correction factor for updating the L / S ratio of the edges.Depending on the respective probabilities for p from a table like the one described above, an edge that is only part of the solution due to the expanded weighting consideration can receive a stronger L / S ratio as a factor. Thus, the new values ​​for L and S would be increased by a rounded-up value of 1 / (1-p) instead of just by one. This means that changes to the map are reflected more quickly in the L / S values ​​and therefore in the weightings. For example: An edge with FRC=2 only became part of the solution because it was also included in the search space due to the value p = 0.8. Therefore, the respective L and S values ​​of the edge are increased by 1 / (1-0.8) = 5.

[0026] According to another advantageous embodiment, the distribution function is statically predetermined and stored in a memory.

[0027] According to a further advantageous embodiment, when executing Dijkstra's algorithm, a node N is classified as a neighbor of a node K if and only if there is an edge in the graph connecting nodes N and K with a weight value that is less than or equal to the value determined to be the maximum permissible deviation from the prohibition of switching.

[0028] Another aspect of the invention relates to a route planning system designed to use a model of a traffic network as a graph, wherein the graph has edges and nodes, and wherein the route planning system is designed to assign weights and a respective parameter L and a respective parameter S to the edges, wherein each of the weights describes a respective attribute of a traffic route and is selected from a predetermined finite set, wherein the route planning system is designed to dynamically adapt the weights to current conditions of the traffic route by initializing the weights with a respective predetermined value, and wherein the respective initialized value is designed to be replaced by a respective updated weight based on a quotient L / S, execute a Dijkstra algorithm for route planning and increment a respective parameter S by one.when the associated edge is visited by Dijkstra's algorithm, and to increment the parameter L by one when the edge is used by Dijkstra's algorithm as part of the planned route.

[0029] Advantages and preferred further developments of the proposed route planning system result from an analogous and substantive transfer of the above statements made in connection with the proposed procedure.

[0030] Further advantages, features and details will become apparent from the following description, in which - possibly with reference to the drawing - at least one embodiment is described in detail.

[0031] They show: Fig. 1: A method for adapting a route planning system according to an embodiment of the invention. Fig. 2: A graph that models a traffic network according to an embodiment of the invention. Fig. 3: The graph of the Fig. 2 with an exemplary starting and ending point, between which an optimal route can be found. Fig. 4: The graph of the Fig. 3 with results of the route search according to an embodiment of the invention. Fig. 5: A procedure in algorithmic form for adapting a route planning system according to an embodiment of the invention.

[0032] The representations in the figures are schematic and not to scale.

[0033] Fig. Figure 1 shows a method for adapting a route planning system, wherein a transport network is modeled by a graph having edges and nodes, wherein the edges are assigned weights and a respective parameter L and a respective parameter S, wherein each of the weights describes a respective attribute of a transport route in a transport network and is chosen from a predefined finite set, wherein the weights are dynamically adapted to current conditions of the transport route by initializing the weights with a respective predefined value S1, and the respective initialized value being replaced by an updated weight based on a quotient L / S S2, wherein a Dijkstra algorithm is executed for route planning and the parameter S is incremented by one when the edge associated with the respective parameter S is visited by the Dijkstra algorithm, and wherein the parameter L is incremented by one.when the edge corresponding to the respective parameter L is used by Dijkstra's algorithm as part of the route found. Furthermore, before each route planning step performed by Dijkstra's algorithm, a random value p is generated, starting with 0. <p<1 ermittelt, wobei geprüft wird, in welchem Bereich einer vorgegebenen Verteilungsfunktion mit Wahrscheinlichkeiten P_i der Zufallswert liegt, und abhängig vom Bereich mit dem Zufallswert bestimmt, bis zu welchem Wert einer Gewichtung von einem Wechselverbot im Dijkstra-Algorithmus zu einer ungünstiger gewichteten Kante im Graph abgewichen werden kann.,

[0034] Fig. Figure 2 shows a graph that models a transportation network. The graph has nodes and edges. Initially, each edge has L = S = 0. Subsequently, the most favorable weight is '1' and the least favorable weight is '5'. Each edge of the graph is initially initialized with a weight of '5'.

[0035] Fig. Figure 3 shows the graph of the Fig. 2 with a given starting node S of the route to a desired destination, the end E of the route to be searched. Since in the first route search from a starting node S to a destination node E all edges have the value of weight '5', a regular Dijkstra algorithm is used to search for a route between S and E. Fig. Figure 4 shows the result of Dijkstra's algorithm: All edges within the search space of Dijkstra's algorithm are marked with dashed lines. The shortest path, and thus the result of Dijkstra's algorithm, is marked with a dash-dot line of increased line thickness. This results in the following parameters for the edges in the scheme "Edge type | L | S | L / S |": remaining | 0 | 0 | - | dashed line | 0 | 1 | 0 | Dash-Dot | 1 | 1 | 1 |

[0036] After many route queries, for example one thousand, in the route planning system on the graph, corresponding parameter values ​​for L and S are obtained, and thus for the quotient (L / S) for the edges of the graph, where the values ​​of the quotient take on values ​​such as 200 / 750, 120 / 700, 50 / 200, 62 / 600, etc. The quotient L / S of the respective edge directly assigns a weight from the set {1, 2, 3, 4, 5} to that edge.

[0037] Fig. Figure 5 shows a complete step-by-step procedure encompassing the processes of Dijkstra's algorithm. While X1 is the initial step of the procedure and thus marks the start, X7 describes the end of the procedure. After starting in X1, all nodes of the graph are initialized. This takes place in X2. The initialization has infinite costs, except for the starting node of the graph, for which costs can be set to zero. Step H1 then follows, in which, as described below, random values ​​for the probability distribution are determined. In the verification step C1, it is checked whether the goal has already been achieved.If this is the case, the process proceeds via path y to step H4, in which, as described below, the value of the parameter L is incremented by one for all edges of the solution path through the graph. Following this, in step H5, the weight value for all considered edges is recalculated as described below, and the procedure concludes in step X7. However, if the result of the check in check step C1 is negative, the process continues with step X3, in which the node of the graph with the lowest cost in the current state is identified. As described in more detail below, step X3 is followed by the procedure steps H2 and H3. In H2, a set of neighbors N of node K is determined according to the P_I probabilities, and in step H3, the parameter S is incremented for all edges to neighbors from K.After step H3, the intermediate step X4 follows, in which all neighbors of the current node are selected. In step X5, the new costs are calculated by adding the current costs and the specified weight of the edge. In check step C2, it is verified whether the new costs are lower than the previous costs. If so, in step X6, the costs of the neighbor and the neighbor's predecessor node are updated. If check step C2 is negative, the process proceeds to check step C3, just as it does after step X6. In C3, it is checked whether all neighbors N of node K have been visited. If the result of check step C3 is negative, the process continues with step X4, as described above. If it is positive, the process returns to check step C1, which, as described above, checks whether the goal has been reached.Is the respective result of a test step in the . Fig. If 5 is positive, the process continues with path y. If it is negative, the process continues with the path marked n. Steps H1 to H5 have the following instructions and properties: - H1: Determine random p; the distribution function of each P_i assigned to the weights can be chosen arbitrarily, e.g., P_1 = 87%, P_2 = 7%, P_3 = 3%, P_4 = 2%, P_5 = 1%; - the exact numbers are not relevant; however, they are preferably chosen to increase logarithmically and so that they cumulatively equal one. - H2: A node N is a neighbor of K if and only if there is an edge with weight ≤ allowed weight connecting N and K, where the allowed weight is determined by the random value p and the underlying distribution function with probabilities P_i. - H3: All edges to neighbors are now part of the search space, therefore the parameter S of the edge is increased by one. - H4: Increase the parameter L by one for all edges of the solution set, i.e., the route found by Dijkstra's algorithm. - H5: Determine the updated weight for all edges in parameter S.

[0038] Although the invention has been further illustrated and explained in detail by means of preferred embodiments, the invention is not limited by the disclosed examples, and other variations can be derived from them by a person skilled in the art without departing from the scope of protection of the invention. It is therefore clear that a multitude of possible variations exist. It is also clear that the embodiments mentioned as examples are truly only examples and are not to be understood in any way as limiting, for example, the scope of protection, the possible applications, or the configuration of the invention.Rather, the preceding description and the description of the figures enable the person skilled in the art to implement the exemplary embodiments in concrete terms, whereby the person skilled in the art, with knowledge of the disclosed inventive concept, can make various changes, for example with regard to the function or the arrangement of individual elements mentioned in an exemplary embodiment, without leaving the scope of protection defined by the claims and their legal equivalents, such as further explanations in the description.

Claims

[1] A computer-implemented method for adapting a route planning system, wherein a road network is modeled by a graph having edges and nodes, wherein the edges are assigned weights and a respective parameter L and a respective parameter S, wherein each of the weights describes a respective attribute of a road in a road network and is chosen from a given finite set, wherein the weights are dynamically adapted to current conditions of the road by initializing the weights with a respective given value (S1), and replacing the respective initialized value with an updated weight based on a quotient L / S (S2), wherein a Dijkstra algorithm is executed for route planning and the parameter S is incremented by one when the edge associated with the respective parameter S is visited by the Dijkstra algorithm.and where the parameter L is incremented by one if the edge associated with the respective parameter L is used by Dijkstra's algorithm as part of the route found. [2] Method according to claim 1, wherein the respective predetermined value for initializing the weights is the least favorable value from the predetermined finite set of possible weights. [3] Method according to one of claims 1 to 2, wherein a respective updated weighting is determined from a respective quotient L / S by an invariant predetermined rule. [4] Method according to one of claims 1 to 2, wherein a respective updated weighting is determined from a respective L / S quotient by a hierarchical classification from the best L / S quotients to the worst L / S quotients. [5] Method according to one of the preceding claims, wherein, prior to each route planning performed by the Dijkstra algorithm, a random value p with 0 <p<1 ermittelt wird, wobei geprüft wird, in welchem Bereich einer vorgegebenen Verteilungsfunktion mit Wahrscheinlichkeiten P_i der Zufallswert liegt, und abhängig vom Bereich mit dem Zufallswert bestimmt wird, bis zu welchem Wert einer Gewichtung von einem Wechselverbot im Dijkstra-Algorithmus zu einer ungünstiger gewichteten Kante im Graph abgewichen werden kann. [6] Method according to claim 5, wherein the distribution function is statically predetermined and stored in a memory. [7] Method according to one of claims 5 to 6, wherein, in the execution of the Dijkstra algorithm, a node N is classified as a neighbor of a node K if and only if there is an edge connecting the nodes N and K in the graph with a weight value that is less than or equal to the value that is determined to be the one up to which the prohibition of switching can be waived. [8] Route planning system designed to use a model of a transport network as a graph, wherein the graph has edges and nodes, and wherein the route planning system is designed to assign weights and a respective parameter L and a respective parameter S to the edges, wherein each of the weights describes a respective attribute of a transport route and is chosen from a given finite set, wherein the route planning system is designed to dynamically adapt the weights to current conditions of the transport route by initializing the weights with a respective given value, and wherein the respective initialized value is designed to be replaced by a respective updated weight based on a quotient L / S, execute a Dijkstra algorithm for route planning and increment a respective parameter S by one,when the associated edge is visited by Dijkstra's algorithm, and to increment the parameter L by one when the edge is used by Dijkstra's algorithm as part of the planned route.

Images