Determination device for laying environment identification

The determination device and method address the issue of erroneous recognition in identifying optical communication cable environments by extracting feature vectors, calculating discrimination functions, and detecting misrecognition points, resulting in improved accuracy through reclassification or rejection of incorrect results.

WO2024246996A1PCT designated stage expired Publication Date: 2024-12-05NT T INC

Patent Information

Application Number
PCT/JP2023/019738
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-26
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Existing methods for identifying the installation environment of optical communication cables using neural network technology suffer from erroneous recognition, leading to difficulties in improving recognition accuracy.

Method used

A determination device and method that extracts feature vectors from longitudinal vibration distributions of optical cables, calculates discrimination functions, compares them with teacher signals, and detects misrecognition points to improve identification accuracy by reclassifying or rejecting erroneous results.

Benefits of technology

The solution effectively reduces misrecognition and enhances the accuracy of identifying the installation environment by correcting or rejecting erroneous identification points, thereby improving the overall recognition accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023019738_05122024_PF_FP_ABST
    Figure JP2023019738_05122024_PF_FP_ABST
Patent Text Reader

Abstract

The purpose of the present invention is to provide a determination device, a determination method, and a program that can reduce misidentification of the environments in which communication equipment has been laid and improve accuracy. The present invention is a determination device that classifies the laying environments of an optical cable 21 and comprises a feature extraction unit 212 that extracts a feature vector xi for each position on the optical cable 21 from a vibration distribution in the longitudinal direction of the optical cable 21, a computation unit 217 that computes an identification function for each position on the optical cable 21 from the feature vector xi and a weighting vector that corresponds to laying environments of the optical cable that are to be classified, a determination unit 220 that, for each position on the optical cable 21, compares the identification function and teaching signals that represent the laying environments and determines the laying environment represented by the teaching signal to which the identification function is closest to be the laying environment of the optical cable, and a mis-determination detection unit 221a that detects mis-determined positions on the optical cable 21 from the relationship between the laying environment of the optical cable determined for each position on the optical cable 21 and the nearby laying environments.
Need to check novelty before this filing date? Find Prior Art

Description

Judgment device for identifying installation environment

[0001] The present disclosure relates to a determination device, a determination method, and a program for identifying an environment in which an optical communication cable is installed.

[0002] Optical fibers have become established as a signal transmission medium for high-speed optical communications due to their low loss and light weight, and in recent years, new ways of utilizing optical fibers for communications have been explored. In optical fiber sensing, which utilizes the phenomenon in which the optical properties of an optical fiber change when an external acoustic vibration is applied to the optical fiber, for example, Non-Patent Document 1 proposes a method for detecting events in a room (such as the movement of people or chairs, or the opening and closing of doors) from optical changes in the input and output signals of an optical fiber installed in a loop in a room.

[0003] We have already verified the practicality of using optical fibers in optical communication cables laid for optical communication purposes as sensors, and for example, by using this in combination with the optical tester reported in Non-Patent Document 2, it is possible to detect distributed vibration in the longitudinal direction of the optical fiber. Our goal is to realize the identification and utilization of the equipment environment around the optical fiber and optical communication cable from this distributed vibration (optical fiber environment monitoring), and the elemental technologies are broadly divided into "optical measurement" for distributed vibration measurement and "data interpretation" technology for identifying the equipment environment.

[0004] This disclosure relates to data interpretation for identifying equipment environments with high accuracy from distributed vibration signals. A waterfall diagram is a method for visually expressing distributed vibration information in three dimensions (measurement distance, vibration magnitude, and time). While a waterfall diagram is a powerful tool for showing the time variation of all vibrations along the optical fiber, it is difficult to directly understand the causes of vibration.

[0005] Therefore, we have disclosed an identification algorithm that uses neural network technology to enable identification of multiple mixed equipment environments from vibration distribution waveforms (Non-Patent Document 3). Non-Patent Document 3 makes it possible to identify equipment environments from data where the equipment environment is unknown by preparing and learning learning data that clearly describes the equipment environment for each class to be identified.

[0006] A. Tomasov, P. Dejdar, A. Cem, P. Munster, T. Horvath, “Event classification from distributed interferometric sensor using neural network”, in 27th International Conference on Optical Fiber Sensors (OFS27), (2022), Th4.58. Z. Pan, K. Liang, Q. Ye, H. Cai, R. Qu, and Z. Fang, “Phase-sensitive OTDR system based on digital coherent detection”, in ACP (2011), 2011, pp. 1-6. M. Inoue, C. Kito, K. Mine, and Y. Koshikiya, “DVS-Signal Classification Analysis Using Widrow-Hoff Learning Algorithm for Cabling Environment”, in 27th International Conference on Optical Fiber Sensors, Technical Digest Series (Optica Publishing Group, 2022), paper W4.2.

[0007] However, the method described in Non-Patent Document 3 has a problem in that the identification results include erroneous recognition, making it difficult to improve the recognition accuracy. Therefore, in order to solve the above problem, an object of the present invention is to provide a determination device, a determination method, and a program that can reduce erroneous recognition of the environment in which communication facilities are installed and improve accuracy.

[0008] In order to achieve the above object, the determination device of the present invention takes advantage of the fact that the installation environment of communication facilities such as optical cables has the property that the distance is relatively long and the distance-wise transition of the identification results is small, and detects erroneous recognition points from the context of the identification results.

[0009] Specifically, the determination device of the present invention is a determination device that classifies the installation environment of an optical cable, and is characterized by comprising: a feature extraction unit that extracts a feature vector for each position of the optical cable from the vibration distribution in the longitudinal direction of the optical cable; a calculation unit that calculates a discrimination function for each position of the optical cable from the feature vector and a weight vector corresponding to the installation environment of the optical cable to be classified; a determination unit that compares the discrimination function with a teacher signal representing the installation environment for each position of the optical cable, and determines that the installation environment represented by the teacher signal to which the discrimination function is closest is the installation environment of the optical cable; and an incorrect determination detection unit that detects a position of the optical cable that has been incorrectly determined based on its relationship with nearby installation environments, regarding the installation environment of the optical cable determined for each position of the optical cable.

[0010] Furthermore, a determination method according to the present invention is a determination method for classifying the installation environment of an optical cable, characterized by comprising the steps of: extracting a feature vector for each position of the optical cable from the vibration distribution in the longitudinal direction of the optical cable; calculating a discrimination function for each position of the optical cable from the feature vector and a weight vector according to the installation environment of the optical cable to be classified; comparing, for each position of the optical cable, the discrimination function with a teacher signal representing the installation environment, and determining that the installation environment represented by the teacher signal to which the discrimination function is closest is the installation environment of the optical cable; and detecting a position of the optical cable that has been misjudged based on its relationship with nearby installation environments for the optical cable installation environment determined for each position of the optical cable.

[0011] By utilizing the above-mentioned properties of the identification results of the installation environment obtained using neural network technology, erroneous recognition points can be detected from the installation environment determined in the distance direction, and the erroneous recognition points can be appropriately processed to improve the recognition accuracy. Therefore, the present invention can provide a determination device and a determination method that can reduce erroneous recognition of the environment in which communication equipment is installed and improve the accuracy.

[0012] Regarding the processing of the erroneously recognized point, for example, the installation environment of the position of the optical cable that has been determined to be erroneously determined is corrected to the installation environment in the vicinity.

[0013] The accuracy of classification can be improved by reclassifying incorrectly recognized results into correct results.

[0014] Furthermore, in processing of erroneous recognition points, for example, the installation environment of the optical cable position that has been determined to be erroneous may be deleted from the data. By discarding suspicious results that are erroneous recognition from the recognition results, it is possible to improve the recognition accuracy.

[0015] The present invention also provides a program for causing a computer to execute the above-described determination method. The determination device of the present invention can also be realized by a computer and a program, and the program can be recorded on a recording medium or provided via a network.

[0016] The above inventions can be combined as much as possible.

[0017] The present invention can provide a determination device, a determination method, and a program that can reduce erroneous recognition of the environment in which communication facilities are installed and improve accuracy.

[0018] Fig. 1 is an image diagram explaining the reclassification of equipment identification and data rejection performed by a determination device according to the present invention. Fig. 2 is a configuration diagram explaining a determination device according to the present invention. Fig. 3 is a diagram explaining the reclassification of identification results performed by a determination device according to the present invention. Fig. 4 is a diagram explaining data rejection performed by a determination device according to the present invention. Fig. 5 is a diagram explaining a determination device according to the present invention.

[0019] The following description of the preferred embodiments of the present invention will be given with reference to the accompanying drawings. The preferred embodiments described below are examples of the present invention, and the present invention is not limited to the preferred embodiments. In this specification and the drawings, components having the same reference numerals are intended to represent the same components.

[0020] (Embodiment 1) An example related to the present disclosure will be described below. Fig. 1 is an image diagram illustrating the reclassification and data rejection of identification results of facility environments obtained using neural network technology, which is the gist of the present disclosure. Here, 11: identification result, 12: correct environmental information, 13: misidentification result, 14: misidentification result identified as a class 1 conduit, 15: identification result identified as a class 0 tunnel, 16: result reclassified to identification result 0, 17, 18, 19: misidentification results excluding misidentification result 14, and 110, 111, 112: reclassified results.

[0021] FIG. 1(A) is a diagram plotting the classification results (identification results 11) for each distance of an optical cable determined using the method of Non-Patent Document 3 in the distance direction. Non-Patent Document 3 describes three-class classification using neural network technology. In this embodiment, the implementation method will be described based on the identification results. The plots of identification results 0, 1, and 2 indicate that the environmental information around the optical cable is classified into tunnel (class 0), conduit (class 1), and aerial (class 2), respectively. The correct environmental information 12 represents the intervals of each environmental information. The ideal identification result is one in which identification result 0 is plotted in the tunnel distance interval, identification result 1 is plotted in the conduit distance interval, and identification result 2 is plotted in the aerial distance interval.

[0022] However, the identification result 11 includes erroneous identification results 13 (many actual identification results include erroneous identification results). For example, a result in which the identification result is 1 despite being a distance section of a telegraph tunnel and classified as a conduit, a result in which a telegraph tunnel with an identification result of 0 is classified as a fictitious distance section of a conduit, or a plot in which a fictitious distance section is classified as a telegraph tunnel with an identification result of 0 corresponds to the erroneous identification result 13. Reclassifying the erroneous identification result 13 as a correct identification result or discarding the result improves the identification accuracy.

[0023] 1B and 1C are conceptual diagrams illustrating how an incorrectly identified result 13 is reclassified into a correct identification result. An example of the reclassification method is described below. First, the number of identification results in the distance direction is counted, and a moving average is calculated and displayed as a histogram. The distances of the target objects, such as cable tunnels, conduits, and aerial tunnels, are relatively long, and the identification results exhibit little change in the distance direction. Based on this characteristic, the moving average value of the number of results in the distance direction is calculated for each identification result, and the moving average value of the incorrectly identified result is smaller than that of neighboring results. Therefore, the moving average values ​​of the number of results are compared with those of other identification results, and the identification result is reclassified into a class with a larger value. For example, as shown in FIG. 1B, the moving average value of the number of incorrectly identified results 14, which were identified as a Class 1 conduit, is smaller than the moving average value of the number of incorrectly identified results 15, which were identified as a Class 0 cable tunnel of the same distance. Therefore, as shown in FIG. 1C, the incorrectly identified result 14 is reclassified into an identification result 16 with a Class 0. Similarly, the incorrectly classified results 17, 18, and 19 in Figure 1(B) are reclassified into classification results 110, 111, and 112 in Figure 1(C). The reclassified classification result 113 in Figure 1(C) is more correctly classified than the original classification result 11 in Figure 1(A).

[0024] Figure 1(D) is an illustration of the rejection of an erroneous classification result 13. Data rejection is performed by quantifying the suspiciousness of the classification result in question from the standard deviation of the group of classification results in nearby locations, assuming the same characteristics as above (few transitions of the classification target), and making a rejection decision. The classification result 114 in which the erroneous classification result in Figure 1(D) has been rejected has a reduced number of plots overall, but the absence of erroneous classification results improves classification accuracy compared to the case of Figure 1(A).

[0025] 2 is a diagram illustrating a determination device 228 of this embodiment. The determination device 228 is a determination device that classifies the installation environment of the optical cable 21, and determines a feature vector x for each position of the optical cable 21 from the vibration distribution in the longitudinal direction of the optical cable 21. i a feature extraction unit 212 that extracts a feature vector x iand a weight vector according to the installation environment of the optical cable to be classified, a determination unit 220 that compares, for each position of the optical cable 21, the discrimination function with a teacher signal representing the installation environment, and determines that the installation environment represented by the teacher signal to which the discrimination function is closest is the installation environment of the optical cable, and an erroneous determination detection unit 221 a that detects a position of the optical cable 21 that has been erroneously determined based on the relationship with nearby installation environments for the optical cable determined for each position of the optical cable 21. The determination device 228 may further include an discrimination dictionary 218 having a dictionary calculation unit 230 and an update unit 227.

[0026] Here, the components are 21: optical cable, 22: ground, 23: tunnel, 24: conduit, 25: aerial, 26: vibration distribution measuring instrument, 27: first vibration distribution data, 28: second vibration distribution data, 29: third vibration distribution data, 210: fourth vibration distribution data, 211: memory unit, 212: feature extraction unit, 213: data reading unit, 214: Fourier transform unit, 215: frequency feature extraction unit, 216: interval maximum / average value extraction unit, 217: discrimination calculation unit, 218: discrimination dictionary, 219: discrimination function calculation unit, 220: classification judgment unit, 221a: erroneous judgment detection unit, 221b: reliability improvement unit, 222: result display unit, 223: discrimination function calculation unit, 224: error calculation unit, 225: teacher signal, 226: squared error calculation unit, 227: weight vector update unit, and 228: judgment device.

[0027] Optical cable 21 is laid below ground level 22 in a conduit (underground tunnel with floorboards) 23 and a conduit (duct pipe buried underground) 24, and then continues to an overhead cable 25 that runs between utility poles above ground. A vibration distribution measuring instrument 26 is installed at one end of the optical cable to detect the distribution of vibration in the longitudinal direction of the optical cable. By continuously detecting the vibration distribution, three-dimensional data of optical cable distance - time - vibration magnitude can be obtained. Hereinafter, this three-dimensional data will be referred to as vibration distribution data.

[0028] In this embodiment, an example of learning and evaluation of a neural network that classifies and determines the installation environment of an optical cable into three classes (k=3): a cable tunnel, a conduit, and a fictitious environment will be described. Note that k is not limited to 3. In preparation for learning, four vibration distribution data are acquired using a vibration distribution measuring instrument 26. The first vibration distribution data 27 is data for an unknown installation environment, and is vibration distribution data that includes a mixture of cable tunnel, conduit, and a fictitious installation environment. The second vibration distribution data 28 is vibration distribution data measured in advance only in a cable tunnel installation environment. The third vibration distribution data 29, like the second vibration distribution data 28, is vibration distribution data measured in advance only in a conduit installation environment. The fourth vibration distribution data 210, like the third vibration distribution data 29, is vibration distribution data measured in advance only in a fictitious installation environment. Since the second vibration distribution data 28, the third vibration distribution data 29, and the fourth vibration distribution data 210 are used as learning data, it is preferable that the first installation environment is different from the installation environment of the unknown vibration distribution data 27. The four vibration distribution data are stored in a memory unit 211.

[0029] The vibration distribution data 27, 28, 29, and 210 are sent to a feature extraction unit 212. A data reading unit 213 reads each of the vibration distribution data 27, 28, 29, and 210 independently, and passes the data to subsequent processes. The vibration distribution data 27, 28, 29, and 210 passed from the data reading unit 213 are converted by a Fourier transform unit 214, which converts the amplitude of vibration continuously observed at each point in the longitudinal direction of the optical cable from the time domain to the frequency domain. Therefore, the vibration distribution data 37, 38, 39, and 310, which are three-dimensional data of optical cable distance-time-vibration amplitude, are converted by the Fourier transform unit 214 into three-dimensional data of optical cable distance-frequency-amplitude (Fourier spectrum data in the longitudinal direction of the optical cable).

[0030] The Fourier spectrum data is passed to the frequency feature extraction unit 215. The frequency feature extraction unit 215 extracts the peak frequency and the magnitude of its amplitude from the Fourier spectrum at each point in the longitudinal direction of the optical cable.

[0031] Furthermore, the vibration distribution data 27, 28, 29, and 210 passed from the data reading unit 213 are used by the section maximum / average value extracting unit 216 to extract the time variation of the magnitude of the vibrations described above at all points on the optical cable.

[0032] The feature vectors thereafter are all or any combination of the peak frequency and amplitude from the Fourier spectrum, the maximum value and average value of the vibration signal for a frame length obtained by dividing the time change of the vibration magnitude into any time interval, or the peak frequency and amplitude from the Fourier spectrum, the maximum value and average value of the vibration signal for a frame length obtained by dividing the time change of the vibration magnitude into any time interval, and the feature vector x. The feature vectors of the first vibration distribution data 27 are passed to the discrimination calculation unit 217. The feature vectors of the second vibration distribution data 28, the third vibration distribution data 29, and the fourth vibration distribution data 210 are passed to the discrimination dictionary 218. Each element of the feature vector x is expressed as x i Let's say.

[0033] The feature vector x of the first vibration distribution data 28 passed to the identification calculation unit 217 i is passed as an input signal to the discriminant function calculation unit 219. The discriminant function calculation unit 219 is composed of a neural network consisting of an input layer, an intermediate layer, and an output layer, and generates three outputs y k is calculated. where j is the jth neuron in the hidden layer of the neural network, k is the kth neuron in the output layer of the neural network, and ω ij is the weight variable given to the jth neuron, and is the element x of the feature vector x. i For ω jk is the weight variable given to the kth neuron, and the output g j where n is the number of first weight variables, m is the number of second weight variables, and α is a constant indicating the degree of smoothness of the sigmoid curve.

[0034] If the longitudinal distance of the optical cable is composed of L points, the calculation unit calculates 3L discriminant functions (each output y (C=0 or 1 or 2) of classes 0, 1, and 2) from the feature vectors of all the first vibration distribution data 27 passed thereto. The 3L discriminant functions are passed to the classification determination unit 220.

[0035] In the classification determination unit 220, The classification result of class c is output based on the feature vector x that is not classified into any class. i If the input is y, the classification result will be incorrect. k indicates the probability of being classified into three installation environments: tunnel, conduit, and fictitious. Therefore, the output y closest to 1 k The classification result is passed to the erroneous judgment detection unit 221a. The classification result that has been reclassified or data discarded by the reliability improvement unit 221b is passed to the result display unit 222.

[0036] The classification accuracy is determined by the weight variable ω' received from the classification dictionary 218. jk and ω' ij The weight variable varies depending on where ρ is the learning coefficient, b k is the teacher signal.

[0037] The method for updating the weight variables will be described below. The feature quantities of the second vibration distribution data 28, the third vibration distribution data 29, and the fourth vibration distribution data 210 passed to the discrimination dictionary 218 are passed to the discrimination function calculation unit 223 as input signals. At this time, similarly to the discrimination calculation unit 217, The discriminant function is calculated.

[0038] The calculated discriminant function is passed to the error calculation unit 224, and the error y k -b k The error is passed to a square error calculation unit 226, and the sum of squares of the error is calculated. is calculated.

[0039] The error is passed to the weight vector update unit 227, and the weight variables are updated using equation (3). The discrimination dictionary 218 repeatedly calculates equations (3) and (4) until the sum of squares of the error reaches a minimum and stabilizes.

[0040] (Example 1 for Improving Recognition Accuracy) In this example, reclassification of classification results will be described. Fig. 3 is a diagram for explaining reclassification of classification results based on a moving average histogram. Here, 31 is a table of classification results 11, 32 is a table of classification results divided by class, 33 is a group of classification results of any length, 34 is calculation results, 35 is the calculation results of distances 9.6 and 11.2, and 36 is the calculation results of distances 2992.0 and 2993.6.

[0041] Table 31 is a table with the distances in the first column and the classification results in the second column for the classification results 11 in Figure 1. In order to focus on the number of classification results for each class, the classification results are divided by class. In table 32 of classification results divided by class, one cell is filled with the classification result for each distance, and the rest are blank. Here, in the figure, the classification results to be filled in use the values ​​assigned to each class, but any value may be used.

[0042] The erroneous determination detection unit 221a detects erroneous determination positions and results from the classification results determined by the classification determination unit 220. An example of a method for detecting erroneous determination is described below. First, a classification result group 33 of any length (5 in the figure, for example) is extracted for each distance classification result. In the figure, as an example, classification results before and after each distance classification result are extracted, but classification results before or after each distance classification result may also be extracted. The number of items in the classification result group 33 is counted and the average value is calculated.

[0043] For example, the classification result group for class 0 at a distance of 4.8 contains all values, so the average value is 1.0. The classification result groups for class 1 and class 2 do not contain any classification results, so the calculated value is 0.0. The next classification result group for class 0 at a distance of 6.4 contains one blank, so the calculated value is 0.8. In this way, the extraction of classification result groups of any length and the calculation of the average number of classification results continue while moving in the distance direction. When classification targets exist in the same class at a certain distance, classification results for locally different classes are suspected to be erroneous classification results. In other words, the misclassification detection unit 221a can detect areas where misclassification has occurred based on the relationship with nearby classification results.

[0044] The reliability improvement unit 221b corrects the installation environment of the optical cable position that was determined to be incorrect to the installation environment in the vicinity. The specific operation of the reliability improvement unit 221b is described below. Calculation results 34 are a table summarizing the results of the above calculations for all distances and all classes. Based on the calculation results 34, the calculated values ​​of each class are compared for each distance, and the class of the identification result is reclassified to the class with the largest value. For example, calculation results 35 for distances of 9.6 and 11.2 were originally classified as class 1, but when the calculated values ​​were compared, class 0 was found to be the highest, so the result is reclassified from class 1 to class 0. Similarly, calculation results 36 for distances of 2992.0 and 2993.6 are reclassified from class 0 to class 2.

[0045] As a result, the classification result 11 is reclassified to the classification result 113, which is a more accurate classification result. In addition, although the calculation value in this embodiment is explained as an average value obtained by counting the number of items, reclassification may be performed using a value obtained by counting the number of items in the classification result group instead of the average value.

[0046] (Example 2 for Improving Recognition Accuracy) In this example, rejection of the classification results will be described. Fig. 4 is a diagram for explaining the rejection of the classification results based on the Chauvenet criterion. Here, 41: table of classification results 11, 42: classification result group of arbitrary length n, 43: average value, 44: standard deviation σ c , 45: Standard deviation magnification A, 46: Normal error integral, 47: Rejection decision, 48: Rejection of classification result, 49: Retention of classification result. is shown.

[0047] Table 41 is a table with the distance in the first column and the classification result in the second column for the classification result 11 in Figure 1 (same as Table 31 in Figure 3). As with Table 31 in Figure 3, a classification result group of an arbitrary length n (for example, 5) for the classification result of each distance is extracted in Table 41.

[0048] The misjudgment detection unit 221a detects the positions and results of misjudgments from the classification results determined by the classification determination unit 220. An example of a method for detecting misjudgments is described below. First, an average value 43 and a standard deviation 44 are calculated based on the extracted classification result group 42 of an arbitrary length n. The respective formulas are as follows: Then, from the two values ​​of the average value 45 and the standard deviation 44, the standard deviation magnification 45 for the classification result is calculated using the following formula.

[0049] The standard deviation multiplication factor is calculated by multiplying the classification result by A times the standard deviation (Aσ) from the average value 43 using the function P() of the normal error integral 46. c Here, the probability that the deviation falls within the normal error integral 46 is calculated as a function P(Aσ c The probability that the classification result falls within a range of A times the standard deviation can be determined from the following equation based on normal distribution (Gaussian distribution). Here, z is the value of the integral interval -1≦z≦1.

[0050] Here, the classification result is average value 43 to Aσ c The probability of deviation of more than (1-P(Aσ c For example, a determination 47 is made as to whether or not the recognition is erroneous based on (1-P(Aσ c If the difference (within the range of distance) is less than 0.5 (yes), the classification result is deemed to be incorrect, and if it is not less than 0.5 (no), the classification result is deemed to be correct. In other words, the error determination detection unit 221a can detect an incorrectly determined part based on the relationship with nearby classification results. This determination is called Chauvenet's criterion. The error determination detection unit 221a applies Chauvenet's criterion to the classification results for all distances and determines whether the determination is correct or incorrect.

[0051] Then, the reliability improvement unit 221b discards 48 the identification result determined to be an incorrect determination, and retains 49 the identification result determined to be a correct determination. By discarding the identification result determined to be an incorrect determination, the identification result 11 becomes an identification result 114, and the correct identification result can be retained.

[0052] By using the above-described algorithm, it is possible to reclassify and reject results to improve the accuracy of the identification results of the facility environment.

[0053] (Embodiment 2) The determination device 228 described in embodiment 1 can also be realized by a computer and a program, and the program can be recorded on a recording medium or provided via a network. Figure 5 shows a block diagram of a system 100. The system 100 includes a computer 105 connected to a network 135.

[0054] Network 135 is a data communications network. Network 135 may be a private or public network and may include any or all of the following: (a) a personal area network, e.g., covering a room; (b) a local area network, e.g., covering a building; (c) a campus area network, e.g., covering a campus; (d) a metropolitan area network, e.g., covering a city; (e) a wide area network, e.g., covering an area spanning city, region, or country boundaries; or (f) the Internet. Communications are conducted over network 135 by electronic and optical signals.

[0055] Computer 105 includes a processor 110 and a memory 115 connected to processor 110. Although computer 105 is depicted herein as a stand-alone device, it is not limited to such, but rather may be connected to other devices not shown in a distributed processing system.

[0056] Processor 110 is an electronic device made up of logic circuits that responds to and carries out instructions.

[0057] The memory 115 is a tangible computer-readable storage medium on which a computer program is encoded. In this regard, the memory 115 stores data and instructions, i.e., program code, that can be read and executed by the processor 110 to control its operation. The memory 115 can be implemented as a random access memory (RAM), a hard drive, a read-only memory (ROM), or a combination thereof. One component of the memory 115 is a program module 120.

[0058] The program modules 120 contain instructions for controlling the processor 110 to perform the processes described herein. Although operations are described herein as being performed by the computer 105 or a method or process or sub-process thereof, those operations are actually performed by the processor 110.

[0059] The term "module" is used herein to refer to a functional operation that may be embodied as either a stand-alone component or an integrated configuration of multiple subcomponents. Thus, program module 120 may be implemented as a single module or as multiple modules operating in cooperation with each other. Furthermore, although program module 120 is described herein as being installed in memory 115 and therefore implemented in software, it may be implemented in any of hardware (e.g., electronic circuitry), firmware, software, or a combination thereof.

[0060] While the program modules 120 are shown as already loaded into memory 115, they may also be configured to reside on storage device 140 for later loading into memory 115. Storage device 140 is a tangible, computer-readable storage medium that stores the program modules 120. Examples of storage device 140 include compact discs, magnetic tape, read-only memory, optical storage media, a memory unit consisting of a hard drive or multiple parallel hard drives, and a universal serial bus (USB) flash drive. Alternatively, storage device 140 may be random access memory or another type of electronic storage device located in a remote storage system (not shown) and connected to computer 105 via network 135.

[0061] System 100 further includes data source 150A and data source 150B, collectively referred to herein as data sources 150, that are communicatively connected to network 135. In practice, data sources 150 may include any number of data sources, i.e., one or more data sources. Data sources 150 may include unstructured data and may include social media.

[0062] The system 100 further includes a user device 130 operated by the user 101 and connected to the computer 105 via a network 135. The user device 130 includes an input device, such as a keyboard or a voice recognition subsystem, that allows the user 101 to communicate information and command selections to the processor 110. The user device 130 also includes an output device, such as a display device or a printer or a voice synthesizer. A cursor control, such as a mouse, trackball, or touch-sensitive screen, allows the user 101 to manipulate a cursor on the display device to communicate further information and command selections to the processor 110.

[0063] The processor 110 outputs the results 122 of the execution of the program modules 120 to the user device 130. Alternatively, the processor 110 can provide the output to a storage device 125, such as a database or memory, or via a network 135 to a remote device not shown.

[0064] For example, a program that performs the operations of the feature extraction unit 212, the identification dictionary 218, the identification calculation unit 217, the erroneous judgment detection unit 221a, and the reliability improvement unit 221b described with reference to Figures 2 to 4 may be the program module 120. The system 100 can be operated as a determination device 228.

[0065] The terms "comprising" or "comprising" should be interpreted as specifying the presence of the stated features, integers, steps or components, but not excluding the presence of one or more other features, integers, steps or components or groups thereof. The terms "a" and "an" are indefinite articles and therefore do not exclude embodiments having a plurality thereof.

[0066] (Other Embodiments) The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention. In short, the present invention is not limited to the above-described embodiment, and the components can be modified and embodied in the implementation stage without departing from the spirit of the present invention.

[0067] Furthermore, various inventions can be formed by appropriately combining the multiple components disclosed in the above embodiments. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components from different embodiments may be appropriately combined.

[0068] (Effects of the Invention)

[0069] The neural network device, method, and program for identifying the environment around an optical communication cable according to the present disclosure are believed to have the following advantages.

[0070] The accuracy of identification can be improved by reclassifying the identification results of the facility environment obtained using neural network technology based on the number of similar identification results in the vicinity or by rejecting the results based on the standard deviation of the identification results.

[0071] [Note] The system of the embodiment comprises: a vibration distribution measuring instrument connected to one end of an optical cable and acquiring a vibration distribution waveform in the longitudinal direction; a memory unit 211 for storing the vibration distribution data acquired from the vibration distribution measuring instrument; a feature extraction unit 212 for acquiring from the memory first vibration distribution data 27 showing the vibration distribution of four different equipment environments in which the equipment environment is unknown, second vibration distribution data 210 in which the equipment environment is a tunnel, third vibration distribution data 29 in which the equipment environment is a conduit, and fourth vibration distribution data 28 in which the equipment environment is fictitious, and for assigning different teacher signals to each of the data, and extracting features of the vibration waveform at each distribution point in the longitudinal direction of the optical cable; a discriminant function calculation unit 217 for outputting the result of an activation function in which an output calculated from a linear sum of weighting coefficients multiplied by the feature amount is input; a classification determination unit 220 for identifying classes by comparing the discriminant functions; an erroneous determination detection unit 221a / reliability improvement unit 221b for reclassifying classes or rejecting the identification result based on the identification result. An identification dictionary 218 is provided for correcting and storing the weighting factors.

[0072] As an example, the erroneous determination detection unit 221a / reliability improvement unit 221b has a function of dividing the identification results by class, calculating the average value of the number of identical identification results in the vicinity of all the identification results, thereby detecting erroneous identification results, and reclassifying them into a class with a larger average value. In this example, for each class into which each distribution point is classified by identification, a moving average value along the cable length of the number of distribution points classified into that class in a predetermined distance section of the cable is calculated, and the identification result of each distribution point is corrected (reclassified) into a class with a larger (largest) moving average value in the vicinity of that distribution point.

[0073] As another example, the error detection unit 221a / reliability improvement unit 221b has a function of calculating the average value and standard deviation of the identification results in the vicinity of the identification result, calculating a magnification of the standard deviation of the identification result, selecting from a normal error integral function the probability that the identification result will deviate from the average value by more than the value obtained by multiplying the magnification and the standard deviation, and rejecting the identification result as an error if the value is less than 0.5. In this example, for each class into which each distribution point is classified by identification, the change in the standard deviation of the number of distribution points classified into that class in the cable longitudinal direction is calculated, and the identification result is rejected based on the Chauvenet criterion using the moving average value and the standard deviation.

[0074] 100: System 101: User 105: Computer 110: Processor 115: Memory 120: Program Module 122: Results 125: Storage Device 130: User Device 135: Network 140: Storage Device 150: Data Source

Claims

1. A determination device for classifying the installation environment of an optical cable, comprising: a feature extraction unit that extracts a feature vector for each position of the optical cable from the vibration distribution in the longitudinal direction of the optical cable; a calculation unit that calculates a discrimination function for each position of the optical cable from the feature vector and a weight vector corresponding to the installation environment of the optical cable to be classified; a determination unit that compares the discrimination function with a teacher signal representing the installation environment for each position of the optical cable, and determines that the installation environment represented by the teacher signal to which the discrimination function is closest is the installation environment of the optical cable; and an error determination detection unit that detects a position of the optical cable that has been incorrectly determined based on its relationship with nearby installation environments, regarding the installation environment of the optical cable determined for each position of the optical cable.

2. The determination device according to claim 1, further comprising a reliability improvement unit that corrects the installation environment of the position of the optical cable that has been erroneously determined to be the installation environment of the nearby position.

3. The determination device according to claim 1, further comprising a reliability improvement unit that deletes from the data the installation environment of the position of the optical cable that has been determined to be erroneously determined.

4. A method for classifying the installation environment of an optical cable, comprising the steps of: extracting a feature vector for each position of the optical cable from the vibration distribution in the longitudinal direction of the optical cable; calculating a discrimination function for each position of the optical cable from the feature vector and a weight vector corresponding to the installation environment of the optical cable to be classified; comparing the discrimination function with a teacher signal representing the installation environment for each position of the optical cable, and determining that the installation environment represented by the teacher signal to which the discrimination function is closest is the installation environment of the optical cable; and detecting a position of the optical cable that has been misjudged based on its relationship with nearby installation environments for the optical cable determined for each position of the optical cable.

5. The method according to claim 4, further comprising correcting the installation environment of the optical cable position determined to be erroneously determined to be the adjacent installation environment.

6. The method according to claim 4, further comprising the step of deleting the installation environment of the optical cable position determined to be erroneously determined from the data.

7. A program for causing a computer to execute the determination method according to any one of claims 4 to 6.

Citation Information

Patent Citations

  • Optical facility identification system, light reflection device, and optical facility identification method

    JP2011149807A

  • Optical cable laying environment measuring method, optical cable laying environment measuring device and optical cable laying environment measuring system

    JP2011164074A

  • Distributed sensing over switched fiber optic networks.

    JP2023036668A

  • Event detection device, event detection system, and event detection method

    WO2023021680A1

Cited By

  • Vibration analysis processing device

    WO2026163343A1

  • Machine learning algorithm for vibration determination

    WO2026190976A1