Observation point optimization device, observation point optimization method, and program
The observation point optimization device and method address the challenge of determining optimal monitoring points in networks by using advanced computational methods to maximize information gain on network reliability, improving disaster response efficiency.
Patent Information
- Application Number
- JP2021072440
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-04-22
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2041-04-22
AI Technical Summary
Existing methods for evaluating network connectivity and reliability in disaster scenarios fail to efficiently determine optimal observation points for monitoring network conditions, especially under resource constraints, leading to incomplete information about network connectivity.
An observation point optimization device and method that utilizes a combination of Cross Entropy method, Crude Monte Carlo method, and Binary Decision Diagrams to identify optimal observation branches for maximizing the expected utility of information gain about network reliability, considering cost constraints.
Efficiently determines observation points that provide high information gain on network reliability, enabling accurate assessment of connectivity even under resource limitations, thus enhancing disaster response capabilities.
Smart Images

Figure 0007584092000013 
Figure 0007584092000014 
Figure 0007584092000015
Abstract
Description
[Technical field]
[0001] The present invention relates to an observation point optimization device, an observation point optimization method, and a program. [Background technology]
[0002] The concept of connectivity is important in networks such as road networks and communication networks. To transport goods between distant points, those points must be connected by roads, and to communicate between distant devices, those devices must be connected by communication nodes or communication cables. In this way, connectivity is the most basic and universal concept of networks, and when there is concern that the network may be damaged due to a failure or disaster, efforts to evaluate and ensure that connectivity are extremely important.
[0003] Network reliability (hereinafter, also referred to as reliability) is a quantitative representation of the connectivity of a network that may vary probabilistically. Reliability is an evaluation of the probability of connection between certain vertices when the network is regarded as a mathematical object, namely, "a graph in which edges are probabilistically cut." Many techniques for efficiently calculating this reliability have been proposed (e.g., Non-Patent Document 1, Non-Patent Document 2, etc.). In addition, techniques for calculating a method of setting edges to realize a highly reliable network (e.g., Non-Patent Document 3, etc.) and techniques for identifying edges that contribute most to increasing reliability (e.g., Non-Patent Document 4, etc.) have also been proposed. For example, it is believed that by applying these conventional techniques to a road network, it is possible to realize a road network that is highly robust against disasters.
[0004] Incidentally, for example, in the case of the transportation of goods after a disaster or the early recovery of communication infrastructure, it is extremely important to collect information on objects, particularly those with network configurations. For example, consider the following problem.
[0005] Problem: Facility A and Facility B need to be accessible by emergency vehicles in the event of a disaster. If roads are damaged and it becomes impossible to reach Facility A and Facility B by land, alternative means such as air routes must be secured, so the connectivity between Facility A and Facility B must be known quickly and with high accuracy. For this reason, the following pre-event and post-event measures are considered.
[0006] - Advance measures In preparation for disasters, we can consider installing surveillance cameras at several locations on the road network in advance to accurately grasp road conditions. However, the number of surveillance cameras that can be installed is limited due to installation and operating costs. So, where is the best place to install surveillance cameras to maximize the effect in terms of grasping connectivity?
[0007] · Logistics After a disaster occurs, information gathering teams are sent to the site to check the road conditions. However, due to human resources and time constraints, it is not possible to survey all roads that may be affected by the disaster. So, which roads should be surveyed to maximize the effectiveness in terms of understanding connectivity?
[0008] Similar problems can also be considered with respect to connectivity in communication networks.
[0009] To quantitatively measure the connectivity of a damaged network, it is sufficient to calculate its reliability using conventional technology. However, in the above problem, it is possible to monitor and investigate parts of the network, and we are considering a situation in which this makes it possible to determine its reliability deterministically. Thus, when assessing the damage to social infrastructure such as transportation and communication networks, it is extremely important to determine which parts of the network should be observed (monitored, investigated, photographed, measured, etc.) in order to quickly and accurately grasp the global characteristics specific to the network, such as its reliability. [Prior art documents] [Patent documents]
[0010] [Patent Document 1] Patent No. 6389808 [Non-patent literature]
[0011] [Non-Patent Document 1] Saito, Hiroshi. "Geometric evaluation of survivability of disaster-affected network with probabilistic failure." IEEE INFOCOM 2014-IEEE Conference on Computer Communications. IEEE, 2014. [Non-Patent Document 2] Hardy, Gary, Corinne Lucet, and Nikolaos Limnios. "K-terminal network reliability measures with binary decision diagrams." IEEE Transactions on Reliability 56.3 (2007): 506-515. [Non-Patent Document 3] Nishino, Masaaki, et al. "Optimizing network reliability via best-first search over decision diagrams." IEEE INFOCOM 2018-IEEE Conference on Computer Communications. IEEE, 2018. [Non-Patent Document 4] Gertsbakh, Ilya, and Yoseph Shpungin. "Network reliability importance measures: combinatorics and monte carlo based computations." WSEAS Transactions on Computers 7.4 (2008): 216-227. [Non-Patent Document 5] De Boer, Pieter-Tjerk, et al. "A tutorial on the cross-entropy method." Annals of operations research 134.1 (2005): 19-67. [Non-Patent Document 6] Kroese, Dirk P., Kin-Ping Hui, and Sho Nariai. "Network reliability optimization via the cross-entropy method." IEEE Transactions on Reliability 56.2 (2007): 275-287. [Non-Patent Document 7] Ikeda, Yasuhiro, Ryoichi Kawahara, and Hiroshi Saito. "Generating a network reliability formula by using binary decision diagrams." IEICE Communications Express 4.9 (2015): 299-303. Summary of the Invention [Problem to be solved by the invention]
[0012] However, Non-Patent Document 1 and Non-Patent Document 2 aim to evaluate connectivity and reliability, and do not consider a situation in which edges are observed and their states are determined. Non-Patent Document 3 aims to design a network that increases reliability, and does not consider a situation in which an existing network is probabilistically damaged and can be observed. Non-Patent Document 4 can identify edges that contribute greatly to reliability, but does not evaluate the amount of information obtained when edges are observed.
[0013] An embodiment of the present invention has been made in consideration of the above points, and has an object to obtain an observation point from which information regarding the reliability of a target network can be efficiently acquired. [Means for solving the problem]
[0014] In order to achieve the above-mentioned object, an observation point optimization device of one embodiment has an input unit that inputs at least information regarding a target network and a target vertex set indicating a set of vertices to be confirmed for connectivity in the target network, an observation point acquisition unit that acquires an observation branch indicating a branch that will be an observation point from among the branches of the target network, a reliability calculation unit that calculates a reliability regarding the connectivity between vertices included in the vertex set after the observation branch has been acquired, and an observation expected utility calculation unit that uses the reliability to calculate an expected value of an observation utility that represents the gain when the observation branch is observed, and the observation point acquisition unit acquires the observation branch so as to maximize the expected value. Effect of the Invention
[0015] It is possible to obtain an observation point from which information regarding the reliability of the target network can be efficiently obtained. [Brief description of the drawings]
[0016] [Figure 1] FIG. 1 is a diagram showing an example of a graph (part 1). [Diagram 2] FIG. 2 is a diagram showing an example of a graph (part 2). [Diagram 3] FIG. 3 is a diagram showing an example of a graph (part 3). [Figure 4] FIG. 13 is a diagram illustrating an example of an algorithm for an observation point optimization process. [Diagram 5] FIG. 2 is a diagram illustrating an example of a functional configuration of an observation point optimization device according to the present embodiment. [Figure 6] FIG. 2 is a diagram illustrating an example of a hardware configuration of an observation point optimization device according to the present embodiment. [Figure 7] FIG. 1 is a diagram (part 1) for explaining an observation point optimization result in the embodiment. [Figure 8] FIG. 2 is a diagram (part 2) for explaining the observation point optimization result in the embodiment. [Figure 9] FIG. 11 is a diagram (part 3) for explaining the observation point optimization result in the embodiment. [Figure 10] FIG. 4 is a diagram (part 4) for explaining the observation point optimization result in the embodiment. [Figure 11] FIG. 5 is a diagram (part 5) for explaining the observation point optimization results in the embodiment. [Figure 12] FIG. 6 is a diagram for explaining the observation point optimization result in the embodiment. [Figure 13] FIG. 7 is a diagram for explaining the observation point optimization result in the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0017] An embodiment of the present invention will be described below. In this embodiment, an observation point optimization device 10 capable of obtaining an observation point (i.e., an optimal observation point) from which information on the reliability of a target network can be efficiently acquired will be described. Here, the target network refers to a network for which an optimal observation point is to be obtained, and can be any network requiring an evaluation of reliability. For example, a road network, a communication network, a transportation network, a circuit network, etc. can be the target network.
[0018] The observation point optimization device 10 according to this embodiment executes an observation point optimization process described later, and is able to determine a set of observation branches that maximizes the amount of information that can be obtained regarding the reliability of a target network under the cost constraints required for observation. Here, observation refers to monitoring, investigating, photographing, measuring, etc., of branches on a network to confirm whether the branch is functioning (i.e., whether vertices are connected). Moreover, an observation branch refers to a branch on which observation has been performed.
[0019] <Theoretical structure> The theoretical configuration of this embodiment will now be described.
[0020] ≪Problem setting≫ The target network is represented as G(V,E), and the vertex set V and edge set E are
[0021]
number
[0022] In the following, unless there is any risk of misunderstanding, we will use the branch e j ∈E can also be expressed as j∈E. Similarly, when there is no risk of misunderstanding, we write vertex v i ∈V can also be expressed as i∈V.
[0023] Each branch e j ∈E is the utilization rate p j ∈[0,1], each with independent probability p j (i.e., connect vertices) with probability 1-p j The state of the target network (hereafter also referred to as the NW state) is a binary vector y∈{0,1} |E| and its jth component y j When is 1, branch e j indicates that the node is in operation, and when it is 0, the branch e j indicates that a malfunction has occurred.
[0024] In the NW state y, two vertices s, t ∈ V of the target network are said to be connected if there exists a j This means that there exists a path consisting of only =1 edges.
[0025] For a set of vertices K ⊂ V, the K-connectivity indicator function φ K (y):{0,1} |E| →{0,1} is φ when any two vertices in K are connected in the NW state y. K (y=1, otherwise φ K It is defined as a function such that (y)=0.
[0026] Next, we introduce the concept of "observation". For a set of subbranch sets M ⊂ E, observation M is made by observing each branch ej ∈M's operating state y j Such an M is called an observation branch set. Also, the branch e after observation M is performed j ∈M is the utilization rate m j ∈{0,1}. This m j is the operating state j is essentially the same as above, but uses different symbols to emphasize that the availability rate has converged to 0 or 1 after observation.
[0027] The result of observation M, m j The set of branches for which m = 0 (called the faulty branch set) is M0 ⊂ M, m j Let M1 ⊂ M be the set of edges where = 1 (called the working edge set). In this case, the observation result (M0, M1) is a partition of M, and M0 ∪ M1 = M and M0 ∩ M1 = φ hold. Note that simply writing φ represents the empty set.
[0028] Observation incurs a cost (hereinafter referred to as observation cost). In general, an arbitrary observation cost function C:2 E →[0,∞) can be used to define the observation cost. In the following, the sum of the costs of each branch is used as an example of the observation cost function C(·). That is, for each branch e j ∈E is cost c j ≧0, the observation cost when making M observations is C(M)=Σ j∈M c j That is, the observation cost C(M) is the cost of each branch e j Cost of c j It is assumed that the sum of
[0029] Based on the above definitions, we formulate the probability of each event.
[0030] The probability that the NW state is y when not observed is
[0031]
number
[0032] The probability that the set of working branches is M1 and the set of faulty branches is M0 when M observations are made (where (M0, M1) is a division of M) is:
[0033]
number
[0034] Also, the probability that the NW state is y under the observation result (M0, M1) is
[0035]
number
[0036] Now, let us define the K-level reliability under the observation result (M0, M1) of observation M as
[0037]
number
[0038] The above K-level reliability r K Using (M0, M1), we define the observed utility U(M0, M1) as a function that represents the gain when the observed result (M0, M1) is obtained. In this embodiment, the observed utility U(M0, M1) is K Any function with (M0,M1) as arguments can be used. In the following, we use negative binary entropy as an example.
[0039]
number
[0040] The observation utility U(M0,M1) shown in the above formula 6 is r K When (M0,M1)=0.5, it takes the minimum value of -1, and as it approaches 0 or 1, the value becomes larger. K (M0, M1) = 0 or 1, and the maximum value is 0. K (M0,M1)=0.5 means that there is a 50 / 50 chance that K is connected or not, meaning there is a high degree of uncertainty. K (M0,M1)=0 or 1 means that it is determined whether K is disconnected or connected. In other words, the observation result is r K It is preferable that (M0,M1) is close to 0 or 1. Therefore, it can be considered that the larger the value of U(M0,M1), the greater the gain from the observation result (M0,M1). Note that, since the observation utility also depends on K by its definition, when K is explicitly stated, the observation utility U(M0,M1) is expressed as U K It can also be expressed as (M0,M1). In the following, we will consider K as fixed, so the observation utility will be expressed as U(M0,M1).
[0041] Next, we define the observation expected utility EU(M) as a function that represents the effectiveness of making an observation M. This is expressed by the expected value of the observation utility U(M0,M1) regarding the observation result. That is,
[0042]
number
[0043] Finally, the optimization problem to be solved in this embodiment (hereinafter, also referred to as branch-selective observation expected utility maximization problem) is formulated. This problem is written as follows.
[0044]
number
[0045] As described above, the observation utility U(M0,M1) indicates the degree to which the connectivity between the vertices in the vertex set K is determined by the observation M (i.e., the degree to which information on connectivity can be obtained), and the expected observation utility EU(M) is the expected value of the observation utility U(M0,M1) when the observation M is performed. Therefore, for example, if a surveillance camera or the like is installed in advance on the branch included in the observation branch set M that maximizes the expected observation utility EU(M), even if a disaster occurs and the road network is partially blocked, it becomes possible to more definitively determine whether or not specific vertices (between vertices included in K) are reachable by the road network. In other words, by observing the branch included in the observation branch set M that maximizes the expected observation utility EU(M), it becomes possible to efficiently obtain information on the reachability between the vertices included in K (i.e., information on the reliability of the target network (or a part of it)), and it becomes possible to more definitively determine the reachability between the vertices included in K in the event of a disaster, etc. Note that K ⊂ V is a set of vertices whose connectivity (reachability) is to be known, and K may be V.
[0046] <Simple example> In the following, we show that for a simple network, the expected observation utility EU(M) is an appropriate index for obtaining observation points from which information on the reliability of the target network can be efficiently obtained.
[0047] First, consider the network represented by the graph shown in Figure 1. This network has a vertex set V = {v1, v2, v3, v4} and an edge set E = {e1, e2, e3, e4}, with the utilization rate of each edge being given as (p1, p2, p3, p4) = (1.0, 0.5, 0.8, 1.0). The vertex set for which we want to know the connectivity is K = {v1, v4}. Furthermore, the cost of each edge is 1, and the budget is L = 1. In other words, we can only observe a maximum of one edge.
[0048] The graph shown in Figure 1 is a non-simple graph having multiple edges, but if vertices v5 and v6 and edges e5 and e6 are added as shown in Figure 2, and the cost of the added edges e5 and e6 is set to L+1 (i.e., they are not included as candidates for the observation edge), and the operating rate is set to 1.0, the graph shown in Figure 1 can be reduced to a graph without multiple edges as shown in Figure 2. Therefore, by reducing the graph shown in Figure 2 to the graph shown in Figure 2, it is possible to apply the formulations explained so far as they are.
[0049] In this case, the expected utility when observing each branch is calculated as follows:
[0050] If either e1 or e4 is observed r with probability 1.0 K = (1.0-0.5×0.2)=0.9, so EU({e1})=EU({e4})=-1.0×H(0.9)≒-0.47 When e2 is observed r with probability 0.5 K = 1, with probability 0.5, r K = 0.8, so EU({e2}) = -0.5 × H(1) - 0.5 × H(0.8) ≒ -0.5 × 0.72 = -0.36 If e3 is observed r with probability 0.8 K = 1, with probability 0.2, r K = 0.5, so EU({e3}) = -0.8 × H(1) - 0.2 × H(0.5) = -0.2 × 1.0 = -0.2 Therefore, in this case, observing e3 is the most effective. In this example, e2 and e3 are in symmetrical positions in the network topology, but the difference in their operating rates is reflected as a difference in the expected utility of observation.
[0051] Next, consider the network represented by the graph shown in Figure 3. This network has a vertex set V = {v1, v2, v3, v4, v5} and an edge set E = {e1, e2, e3, e4}, with the utilization rate of each edge being given as (p1, p2, p3, p4) = (1.0, 0.5, 1.0, 0.5). The vertex set for which we want to know the connectivity is K = {v1, v4}. Furthermore, the cost of each edge is 1, and the budget is L = 1. In other words, we can only observe a maximum of one edge.
[0052] In this case, the expected utility when observing each branch is calculated as follows:
[0053] If either e1 or e3 is observed r with probability 1.0 K = 0.5, so EU({e1}) = EU({e3}) = -1.0 × H(0.5) = -1.0 When e2 is observed r with probability 0.5 K = 1, with probability 0.5, r K = 0, so EU({e2}) = -0.5 × H(1) - 0.5 × H(0) = 0 When e4 is observed r with probability 0.5 K = 0.5, with probability 0.5 r K = 0.5, so EU({e4}) = -0.5 × H(0.5) - 0.5 × H(0.5) = -1.0 Therefore, in this case, observing e2 would be the most effective. In this example, e2 and e4 have the same operating rate, but the difference in their connections on the network is reflected in the difference in the expected utility of observing.
[0054] As described above, the expected observation utility defined in this embodiment is an index that expresses the effectiveness of observation taking into account both the topology and availability of the network, and by maximizing this, it is possible to obtain an observation point from which information regarding the reliability of the target network can be efficiently obtained.
[0055] <Branch-selective observation expected utility maximization problem> The branch-selective observation expected utility maximization problem shown in the above formula 8 has the following three difficulties (1) to (3).
[0056] (1) Combinatorial optimization It is necessary to optimize the set of observation branches under cost constraints, but there are only two patterns of the set of observation branches. |E| Since there are many ways to do this, the search space grows exponentially, and if executed naively, the number of times EU(·) would be calculated would be enormous.
[0057] (2) Calculating the expected value of EU(·) When calculating EU(M), it is necessary to perform an expectation calculation. The number of patterns of observation results obtained under observation M is 2. |M| Therefore, naively, we would have to calculate U(·,·) an exponential number of times and then calculate the sum of these.
[0058] (3) K-way reliability calculation To calculate U(M0,M1), K It is necessary to perform the calculation of (M0,M1), but r K The computation of (M0,M1) is in general #P-complete, and this computation needs to be repeated as many times as required by (2) above.
[0059] In this embodiment, the above-mentioned difficulties are solved by combining three methods, Method 1 to Method 3, which correspond to the above-mentioned (1) to (3). Examples of these three methods are as follows.
[0060] (Method 1) A method for combinatorial optimization: Importance sampling using the Cross Entropy method (hereinafter also referred to as the CE method) This method uses the CE method (see, for example, Non-Patent Document 5 and Non-Patent Document 6), which is an importance sampling method used for approximate calculations regarding rare events. The CE method is a method that successively brings the proposal distribution closer to the optimal importance sampling distribution limited to rare events based on cross entropy. In this method, by setting "EU(M) being large" as a rare event, a proposal distribution that can efficiently sample M such that EU(M) is large is obtained, and combinatorial optimization is approximately solved using this. The use of the CE method makes it possible to efficiently search within a search space that is exponentially large.
[0061] (Method 2) A method that corresponds to the calculation of the expected value of EU(·): Approximation calculation using the Crude Monte Carlo method (hereinafter also referred to as the CMC method) The operating rate of each branch p j Since is known, the observation result (M0, M1) can be generated by sampling. Therefore, in this method, (M0, M1) is sampled multiple times according to the direct product measure of the Bernoulli measure defined by the utilization rate, and r is generated by (Method 3) described later. K EU(M) is approximated by calculating (M0,M1) and U(M0,M1) and taking the average. This is a simple Monte Carlo method (CMC method). By using the CMC method, it is possible to approximate the expected value expressed as the sum of exponential times with a smaller number of samples.
[0062] (Method 3) A method corresponding to the K-way reliability calculation: Calculation of K-way reliability using a Binary Decision Diagram (hereinafter referred to as BDD) BDD is a data structure for expressing Boolean algebra. In this method, we use a K-connectivity indicator function φ K We construct a BDD to represent (y). Then, we assign a utilization rate p j ya m j As a result, the exact solution is r K(M0, M1) can be calculated efficiently (see, for example, Non-Patent Document 2). Here, the construction of the BDD is generally the limiting factor of the computational complexity, and r K The calculation of (M0, M1) is lightweight. In this embodiment, we calculate r K (M0,M1) needs to be calculated, but the K-connectivity indicator function φ K Once we have constructed a BDD that represents (y), K Even if the arguments of (M0, M1) change, it is efficient because it is only necessary to change the weights on the BDD and perform dynamic programming.
[0063] In the following, the case where each of the methods described in (Method 1) to (Method 3) above is used will be described, but the present embodiment is not limited to these, and other methods can be used instead. For example, instead of the method described in (Method 1) above, another heuristic method capable of performing combinatorial optimization such as a genetic algorithm may be used. Also, for example, instead of the method described in (Method 2) above, importance sampling may be used in which sampling is performed preferentially from the point where the value of U(M0, M1)Pr(M0, M1|M) is large. Also, for example, instead of the method described in (Method 3) above, another network reliability calculation method according to the conventional technology such as the Monte Carlo method may be used, or a method of explicitly calculating a formula for calculating reliability in advance and then substituting arguments into it to reduce the amount of calculation (for example, see Non-Patent Document 7, Patent Document 1, etc.) may be used.
[0064] <<Observation point optimization processing algorithm>> The algorithm for solving the branch-selective observation expected utility maximization problem shown in the above formula 8 to find the optimal observation point (observation point optimization process) will be described with reference to FIG. 4. FIG. 4 is a diagram showing an example of the algorithm for observation point optimization process. The inputs to this algorithm are the target network G(V,E), a vertex set K whose connectivity is to be known, and the utilization rate {p j} j , the edge cost {c j} j, budget L, predefined parameters N1, N2, N3, ρ, α, β, and the output is a set of observation branches M * ⊂ E. Note that N1, N2, and N3 are integers equal to or greater than 1, ρ>0, 0<α<1, and β>0.
[0065] First, in the first line, for a given target network G(V,E) and vertex set K, we calculate the K-connectivity indicator function φ K We construct a BDD to represent (y). Next, in the second line, we initialize the reference vector used in the CE method as a0 = (0.5, , 0.5). Note that the number of elements in the reference vector is |E|, and the j-th element is the j-th branch e j Corresponds to.
[0066] Line 3 represents the iteration of the CE method, repeating lines 4 to 13 for t = 1, , N1. Line 4 represents the iteration of lines 5 to 10 for k = 1, , N2. Line 5 represents the iteration of the reference vector a t-1 According to the observation branch set M k The method for sampling the observation branch set will be described later.
[0067] Line 6 indicates that lines 7 to 9 are repeated for l = 1, , N3 (note that l is a lowercase L). Line 7 indicates that the availability rate {p j} j According to the observation branch set M k From the fault set M0 k,l and the working edge set M1 k,l In line 8, we use the BDD constructed in line 1 to sample r K (M0 k,l ,M1 k,l In line 9, by the definition of U(·,·), we calculate r K (M0 k,l ,M1 k,l ) to U(M0 k,l ,M1 k,l ) in line 10. Then, for l=1, ,N3, we calculate U(M0 k,l ,M1 k,l), EU(M k ) is obtained approximately.
[0068] And on line 11,
[0069]
number
[0070] Next, in line 12, the above γ^ t Using the following update formula, the reference vector a t Update.
[0071]
number
[0072] In line 12 above, the reference vector a t By updating, it becomes easier to sample M with a larger EM(M) in the next iteration of the CE method.
[0073] Next, in line 13, it is determined whether or not the following termination condition is met, and if the termination condition is met, the loop in line 3 is terminated.
[0074]
number
[0075] Then, in line 14, the final reference vector a t We sample an appropriate number of observation branch sets M according to the above, and select the set M that maximizes EU(M). * In addition, in the 14th line, for example, the EU(M k ) that gives the largest value M k M * It is also possible to use the following.
[0076] In addition, this algorithm does not include details of the BDD construction method and the CE method (i.e., the reference vector a t It can be applied regardless of the update formula, termination condition, sampling method of the observation branch set M, etc.
[0077] · Sampling method for the observation branch set M Next, as an example of the sampling method of the observation branch set M in the fifth line of the above algorithm, we will explain the case where the observation cost function C(·) is the sum of the costs of each branch and the knapsack constraint is considered as the constraint of the maximization problem.
[0078] First, let us consider the variable C, which represents the cumulative cost. Temp Initialize to 0. Next, the edge set E = {e1, ,e |E|} uniform random permutation {e σ(1) ,···,e σ(|E|)} is obtained. In the order of this permutation, it is decided whether to adopt the branch (i.e., add the branch to the observation branch set M) or not to adopt the branch (i.e., do not add the branch to the observation branch set M).
[0079] Branch e σ(j) For probability a t,σ(j) provisional adoption, probability 1-a t,σ(j) Next, the tentatively adopted branch e σ(j) For the cumulative cost and edge e σ(j) The cost of C Temp +c σ(j)If the budget exceeds L, the plan is rejected. If it is less than L, the plan is adopted and the cumulative cost is C Temp ←C Temp +c σ(j) and update.
[0080] The above method allows probabilistic sampling of the observation branches set M. Note that if a definition or constraint of C(·) other than the above is used, the sampling method for the observation branches set M will be different, and the update formula in line 12 of the above algorithm will also be different.
[0081] <Functional configuration of the observation point optimization device 10> Next, the functional configuration of the observation point optimization device 10 according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a diagram showing an example of the functional configuration of the observation point optimization device 10 according to this embodiment.
[0082] 5, the observation point optimization device 10 according to this embodiment has an input unit 101, an observation point optimization processing unit 102, and an output unit 103. Each of these units is realized, for example, by processing executed by a processor 205 (described later) or the like of one or more programs installed in the observation point optimization device 10.
[0083] The input unit 101 receives input data (a target network G(V,E), a vertex set K for which connectivity is to be known, and an operating rate {p j} j , the edge cost {c j} j , budget L, etc.) and parameters (N1, N2, N3, ρ, α, β, etc.) are input from any input source (e.g., auxiliary storage device, keyboard, mouse, other device or equipment connected via a communication network, etc.). j} j may be input as information indicating the probability of network damage, such as hazard information, for example.
[0084] The observation point optimization processing unit 102 executes the observation point optimization process represented by the algorithm shown in FIG. 4 using the input data and parameters input by the input unit 101, and obtains an observation branch set M * Calculate ⊂ E.
[0085] The output unit 103 outputs the observation branch set M calculated by the observation point optimization processing unit 102. * is output to a predetermined output destination (e.g., an auxiliary storage device, a display, or another device or equipment connected via a communication network).
[0086] Here, observation point optimization processing unit 102 includes a BDD construction unit 111 , an observation point acquisition unit 112 , an observation expected utility calculation unit 113 , and a reliability calculation unit 114 .
[0087] The BDD construction unit 111 executes construction of the BDD in the first line of the algorithm shown in Fig. 4. The BDD constructed by the BDD construction unit 111 is passed to the reliability calculation unit 114.
[0088] The observation point acquisition unit 112 initializes the reference vector a0 in the second line of the algorithm shown in FIG. 4, controls the repetition in the third line, controls the repetition in the fourth line, and determines the observation branch set M k Sampling of,reference vector a in lines 11 to 12 t , the determination of the termination condition in line 13 and the control to exit the loop in line 3, and the observation branch set M in line 14. * The observation branch set M sampled by the observation point acquisition unit 112 is k is passed to the observation expected utility calculation unit 113. Also, the observation branch set M sampled by the observation point acquisition unit 112 is * is passed to the output unit 103.
[0089] In addition, the set of observation branches M sampled in line 14 of the algorithm shown in Figure 4 is called M *In this case, the observation branch set M is passed to the observation expected utility calculation unit 113, and the returned EU(M)s are compared. Also, as described above, in the 14th line of the algorithm shown in FIG. 4, the EU(M k ) that gives the largest value M k M * It is also possible to use the following.
[0090] The observation expected utility calculation unit 113 performs the repetition control in the sixth line of the algorithm shown in FIG. k,l ,M1 k,l ), the observed utility U(M0 k,l ,M1 k,l ) and the observed expected utility EU(M k The observation expected utility calculation unit 113 receives the observation branch set M from the observation point acquisition unit 112. k When it is passed, lines 6 and 7 are executed, and (M0 k,l ,M1 k,l ) updated by sampling {m j} to the reliability calculation unit 114. The observation expected utility calculation unit 113 also receives the K-way reliability r K If passed, line 9 will be executed.
[0091] The reliability calculation unit 114 receives the availability rate {m j} is passed, the K-way trust r K (M0 k,l ,M1 k,l ) calculation.
[0092] 5 shows an example in which each unit is included in one observation point optimization device 10, but the present invention is not limited to this, and each unit may be distributed among multiple devices. For example, the BDD construction unit 111, the observation point acquisition unit 112, the observation expected utility calculation unit 113, and the reliability calculation unit 114 may be distributed among multiple devices.
[0093] <Hardware configuration of the observation point optimization device 10> Next, the hardware configuration of the observation point optimization device 10 according to this embodiment will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of the hardware configuration of the observation point optimization device 10 according to this embodiment.
[0094] 6, the observation point optimization device 10 according to this embodiment is realized by a general computer or computer system, and includes an input device 201, a display device 202, an external I / F 203, a communication I / F 204, a processor 205, and a memory device 206. Each of these pieces of hardware is connected to each other via a bus 207 so as to be able to communicate with each other.
[0095] The input device 201 is, for example, a keyboard, a mouse, a touch panel, etc. The display device 202 is, for example, a display, etc. Note that the observation point optimization device 10 does not necessarily have to have at least one of the input device 201 and the display device 202.
[0096] The external I / F 203 is an interface with an external device such as a recording medium 203a. The observation point optimization device 10 can read and write data from and to the recording medium 203a via the external I / F 203. Examples of the recording medium 203a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), and a USB (Universal Serial Bus) memory card.
[0097] The communication I / F 204 is an interface for connecting the observation point optimization device 10 to a communication network. The processor 205 is, for example, various types of arithmetic devices such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit). The memory device 206 is, for example, various types of storage devices such as a HDD (Hard Disk Drive), an SSD (Solid State Drive), a RAM (Random Access Memory), a ROM (Read Only Memory), and a flash memory.
[0098] The observation point optimization device 10 according to this embodiment has the hardware configuration shown in Fig. 6, and is therefore able to realize the observation point optimization process described above. Note that the hardware configuration shown in Fig. 6 is merely an example, and the observation point optimization device 10 may have other hardware configurations. For example, the observation point optimization device 10 may have multiple processors 205, and may have multiple memory devices 206.
[0099] <Example> In the following, an embodiment will be described in which a simulation is performed using an artificial network and a communication network.
[0100] In this embodiment, the three types of networks ((A) random graph, (B) random graph, and (C) communication network) shown in Table 1 below were used as target networks.
[0101] [Table 1] (A) and (B) are random graphs generated by the Erdős-Rényi model (ER model) with an edge generation rate of 0.2, and (C) is a communication network described in Non-Patent Document 6. Note that |V|, |E|, L, N2, and N3 are as described in Table 1 above.
[0102] Operating rate {p j} jIn all cases (A) to (C), is given according to a uniform distribution in [0.6, 1.0]. In addition, ρ = 0.1, α = 0.7, N1 = 20, and there is no termination condition (i.e., the CE method is repeated N1 times). Furthermore, the cost of each branch is set to 1. In other words, a maximum of L observation branches can be selected for each graph (A) to (C).
[0103] First, the optimization results of the observation points for (A) are shown in Fig. 7, and the optimization results of the observation points for (B) are shown in Fig. 8. These results are consistent with the exact optimal solution.
[0104] Furthermore, regarding the EU(M) value in (B), the strict maximum value is −0.20625750730195735, whereas the value calculated by the observation point optimization device 10 according to this embodiment is −0.21339758724528224, which roughly matches the strict solution.
[0105] Regarding (B), Fig. 9 shows the change in the value of each element of the reference vector with respect to the iterations of the CE method. In Fig. 9, the vertical axis represents the value of each element, and the horizontal axis represents the number of iterations t. Also, series 1 to series 20 correspond to the 1st element to the 20th element, respectively.
[0106] As shown in Figure 9, we can see that L=5 elements converge to 1 after about 10 iterations, and the other elements converge to 0 (note that the branch corresponding to the element that converged to 1 is the optimal observation branch).
[0107] Also, for (B), the -γ^ t The change in is shown in Figure 10. As shown in Figure 10, it can be seen that this also nearly converges after about 10 iterations.
[0108] From the above, it has been found that the observation point optimization device 10 according to this embodiment can efficiently obtain an approximate solution even if the number of iterations of the CE method is small.
[0109] Next, the results of the optimization of observation points for (C) are shown in Figure 11. An exact solution for (C) cannot be obtained due to its large scale. However, the validity of this result can be understood intuitively. That is, in Figure 11, the result is that the edges connecting to vertices a, m, and x belonging to vertex set K are observed, but if we consider that there are many detours once we leave the periphery of vertex set K, it is natural to understand that the connectivity of several vertices around vertex set K determines the connectivity between vertices K.
[0110] As with (B), the change in the value of each element of the reference vector for (C) with respect to the iterations of the CE method is shown in Fig. 12. t The change in the value of the reference vector is shown in FIG. 13. However, in FIG. 12, in order to avoid complexity, only the changes in the values of the 1st element to the 36th element of the reference vector are shown, and the changes in the values of the 37th element to the 51st element are omitted.
[0111] As shown in FIG. 12 and FIG. 13, similar to (B), (C) also shows that the iteration count is approximately 10 and converges.
[0112] Here, in order to compare with the observation point optimization device 10 according to this embodiment, a method was implemented in which, instead of using the CE method as a sampling method for the observation branch set M, EU(M) was calculated for each observation branch set M using a method of simply randomly sampling 10N2 = 10,000 times (i.e., the value of each element of the reference vector was set to 0.5, and then sampling was performed in the same manner as the CE method without updating the reference vector at all), and the M that gave the maximum value among them was adopted as the optimal observation point set. With this method, the maximum value of EU was achieved as -0.26841349682012183. On the other hand, in the observation point optimization device 10 according to this embodiment, in the 10th iteration of the CE method, which corresponds to the same number of samplings, γ^ was calculated as shown in FIG. 10 ≒0.15 (Note that the EU value is γ^ 10(The expected observation utility EU is larger than the expected observation utility EU of the observation point optimization device 10 according to the present embodiment.) In other words, the method executed by the observation point optimization device 10 according to the present embodiment (the method using the CE method) obtains a solution with a larger expected observation utility EU than the method using only simple sampling with the same number of samples for the observation branches. This also shows the effectiveness of the method proposed in this embodiment.
[0113] The present invention is not limited to the above-described embodiments specifically disclosed, and various modifications, changes, combinations with known technologies, etc. are possible without departing from the scope of the claims. [Explanation of symbols]
[0114] 10. Observation point optimization device 101 Input section 102 Observation point optimization processing unit 103 Output section 111 BDD Construction Department 112 Observation Point Acquisition Department 113 Observation Expected Utility Calculation Unit 114 Reliability Calculation Unit
Claims
1. An input unit for inputting at least information about a target network and a target vertex set indicating a vertex set to be a target for confirming connectivity in the target network; an observation point acquisition unit that acquires an observation branch set, which is a set of observation branches, from among the branches of the target network according to a sampling distribution of observation branches indicating branches that are observation points; a reliability calculation unit that calculates a reliability regarding connectivity between vertices included in the vertex set under the observation branch set being acquired; an observation expected utility calculation unit that calculates an expected value of an observation utility that represents a gain when the observation branch is observed, using the reliability; an update unit that updates the sampling distribution by using a plurality of expected values obtained by sequentially repeating the acquisition of the observation branch set by the observation point acquisition unit, the calculation of the reliability by the reliability calculation unit, and the calculation of the expected value by the observation expected utility calculation unit, so as to obtain an observation branch set with a higher expected value of observation utility; having an observation point optimization device that repeats sequentially the acquisition of the observation branch set by the observation point acquisition unit, the calculation of the reliability by the reliability calculation unit, and the calculation of the expected value by the observation expected utility calculation unit, and the updating of the sampling distribution by the update unit, until a predetermined termination condition is satisfied.
2. a binary decision diagram constructing unit for constructing a binary decision diagram expressing a function that takes the value 1 when any two vertices included in the target vertex set are connected and takes the value 0 when they are not connected, for the target network and the target vertex set; The reliability calculation unit is The observation point optimization device according to claim 1 , wherein the reliability is calculated by dynamic programming using the binary decision diagram.
3. The information about the target network includes an availability rate of each branch of the target network; The observation expected utility calculation unit is Sampling working observation branches indicating working observation branches and non-working observation branches indicating non-working observation branches according to the operation rate; Calculating the observed utility from the working observation branch and the non-working observation branch; 3. The observation point optimization device according to claim 1, wherein the expected value is calculated by calculating a sampling average of the observation utilities.
4. The input unit includes: inputting a cost required for observing an edge of the target network and constraint information related to the cost; The observation point acquisition unit, The observation point optimization device according to claim 1 , wherein the observation branch is acquired so that the cost of the observation branch satisfies the constraint information.
5. The observation point acquisition unit, 5. The observation point optimization device according to claim 1, wherein the observation branches are obtained by sampling the observation branches using a cross-entropy method in accordance with the sampling distribution.
6. 6. The observation point optimization device according to claim 1, wherein the observation utility is a binary entropy of a probability that two vertices included in the target vertex set are connected to each other.
7. An input step of inputting at least information about a target network and a target vertex set indicating a vertex set to be a target for confirming connectivity in the target network; an observation point acquisition step of acquiring an observation branch set, which is a set of observation branches, from among the branches of the target network according to a sampling distribution of observation branches indicating branches that are observation points; a reliability calculation step of calculating a reliability of connectivity between vertices included in the vertex set under the observation branch set being obtained; an expected observation utility calculation step of calculating an expected value of an observation utility representing a gain when the observation branch is observed, using the reliability; an updating procedure for updating the sampling distribution using a plurality of expected values obtained by sequentially repeating the steps of obtaining the observation branch set by the observation point obtaining procedure, calculating the reliability by the reliability calculation procedure, and calculating the expected value by the expected observation utility calculation procedure, so as to obtain an observation branch set with a higher expected value of observation utility; The computer executes an observation point optimization method, which repeats sequentially obtaining the observation branch set by the observation point obtaining procedure, calculating the reliability by the reliability calculation procedure, and calculating the expected value by the observation expected utility calculation procedure, and updating the sampling distribution by the update procedure, until a predetermined termination condition is satisfied.
8. A program for causing a computer to function as the observation point optimization device according to any one of claims 1 to 6.
Citation Information
Patent Citations
Light modulating and deflecting element
JP1988089808A
Computational expression generation device, computational expression generation method, and program
JP2017021552A
Sensor arrangement method and sensor arrangement program
JP2017180598A
