Railway subgrade deformation risk identification method and device

Abstract

The present application provides a kind of railway subgrade deformation risk identification method and device, based on grey wolf algorithm, the historical deformation data of each monitoring point along the target railway is divided into blocks according to distribution similarity, and independent principal component analysis model is established respectively, to obtain prior knowledge, obtain the prior solution mixing matrix corresponding to each prior sub-block, and the control limit corresponding to prior statistical quantity and prior statistical quantity;When processing the newly generated data of each monitoring point, grouping is carried out according to the prior block obtained based on historical deformation data, and the posterior statistical quantity and posterior statistical quantity are quickly calculated using the corresponding prior solution mixing matrix, combined with the control line obtained in prior knowledge to carry out local risk monitoring, calculate bayesian integrated statistics for posterior statistical quantity and posterior statistical quantity, monitor global deformation risk using preset significance level value, realize the systematic risk identification of local and global for hidden danger area, improve the discrimination efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of railway risk supervision, and in particular to a railway subgrade deformation risk identification method and device. BACKGROUND

[0002] In the prior art, the safety state of railway infrastructure can be effectively monitored by deploying Beidou high-precision positioning terminals in the risk hidden danger area along the railway. The risk hidden danger area along the railway mainly includes geological disaster danger points along the railway, key infrastructure, etc., and its span varies from a few hundred meters to thousands of meters. Therefore, multiple Beidou terminals are often deployed on site to jointly monitor external risk hidden dangers along the railway. Existing research mainly gives a Beidou deformation monitoring data early warning method for a certain monitoring point through a data-driven method. Although the safety state of railway subgrade deformation can be effectively identified, the analysis result only represents the risk hidden danger condition of a local area and cannot describe the overall safety state of the risk hidden danger area. Therefore, it is urgent to jointly analyze the monitoring data of each group in the hidden danger area along the railway and realize the demand for systematic risk identification of the hidden danger area. Moreover, the analysis and processing speed is slow and cannot meet the demand for rapid response in a large range. SUMMARY

[0003] In view of this, the embodiments of the present application provide a railway subgrade deformation risk identification method and device to eliminate or improve one or more defects in the prior art and solve the problem that the prior art cannot use prior knowledge of historical data to quickly and efficiently monitor deformation risks of the railway locally and as a whole.

[0004] One aspect of the present application provides a railway subgrade deformation risk identification method, which comprises the following steps:

[0005] Obtaining historical deformation data of a target railway collected by each monitoring point in cooperation with a reference station; the reference station and the monitoring points are deployed at preset positions along the target railway;

[0006] Grouping the historical deformation data of each monitoring point into multiple candidate subblocks two by two, constructing multiple grey wolves based on a grey wolf algorithm as solutions of the corresponding groups of the candidate subblocks, minimizing the sum of the KL divergences of the historical deformation data of each subblock in the solutions, and solving to obtain multiple prior subblocks of an optimal group;

[0007] Constructing an independent principal component analysis model for each prior subblock and outputting the prior statistic quantity and its first-type control limit, prior statistic quantity and its second-type control limit, and recording the prior solution mixing matrix corresponding to each prior subblock;

[0008] acquire deformation data to be analyzed collected by the monitoring points in a target period, and construct sub-blocks to be analyzed according to the grouping of the prior sub-blocks;

[0009] In combination with the prior disentangling matrix, calculate posteriori statistics and posteriori statistics for each sub-block to be analyzed;

[0010] fuse the posteriori statistics and the posteriori statistics of each sub-block to be analyzed through Bayesian comprehensive inference, to obtain Bayesian comprehensive statistics;

[0011] In combination with the first type of control limit and the second type of control limit, compare the posteriori statistics and the posteriori statistics of each sub-block to be analyzed, to monitor local deformation risk and its occurrence time; in combination with a preset significance level, compare the Bayesian comprehensive statistics to monitor global deformation risk and its occurrence time.

[0012] In some embodiments, the historical deformation data and the deformation data to be analyzed are displacement data sequences collected at a set frequency.

[0013] In some embodiments, a plurality of grey wolves are constructed based on a grey wolf algorithm as solutions of the grouping corresponding to the candidate sub-blocks, the sum of the KL divergences of the historical deformation data in each sub-block in the solutions is minimized, and a plurality of priori sub-blocks of an optimal grouping are solved, including:

[0014] Initialize a grey wolf group, and each grey wolf represents a potential solution of grouping the historical deformation data of each monitoring point in pairs to obtain a plurality of sub-blocks;

[0015] Calculate the sum of the KL divergences of the historical deformation data in each sub-block corresponding to each grey wolf as an adaptive function; find the grey wolf with the highest adaptive function, the grey wolf with the second highest adaptive function, the grey wolf with the third highest adaptive function, and the remaining grey wolves as the grey wolves according to the adaptive functions corresponding to each grey wolf;

[0016] In the hunting stage, each grey wolf updates its own position according to the position of the wolf in this round, and the expression is as follows:

[0017] ;

[0018] ;

[0019] ; ​​​​​

[0020] ;

[0021] in, Let X(t) represent the position of the prey, X(t) represent the position of the gray wolf in round t, D represent the distance the gray wolf needs to move, and A and C are both coordination coefficients. This represents the convergence factor, which decreases linearly from 2 to 0 based on the number of iterations. and

[0022] Represents a random number between 0 and 1;

[0023] During the hunting phase, each gray wolf is determined according to its current round. Wolf, Wolf, The wolf updates its position using the following expression:

[0024] ;

[0025] ;

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] in, express The wolf's current location express The wolf's current location express The wolf's current location; Indicates the location of each individual gray wolf; , , Represents a random number between 0 and 2; express The distance between the wolf and its prey, express The distance between the wolf and its prey express The distance between the wolf and its prey; , , Represents the distance coefficient vector; express The updated location of the wolf. represents the updated position of the wolf, represents the updated position of the wolf; is the first iteration of the position of the grey wolf individual;

[0032] updating the social rank of the grey wolf group;

[0033] The algorithm ends when the maximum number of iterations is reached or the KL divergence and convergence to a set threshold are reached.

[0034] In some embodiments, an independent principal component analysis model is constructed for each of the prior sub-blocks, and a prior statistic and its first type control limit, prior statistic and its second type control limit corresponding to each of the prior sub-blocks are output, and a prior unmixing matrix corresponding to each of the prior sub-blocks is recorded, including:

[0035] For the historical deformation data of the target railway containing m variables, n samples, the data set is divided into p prior sub-blocks, each of which is represented as ; the variables within the i-th prior sub-block are represented as , represents the j-th measurement variable in the i-th prior sub-block;

[0036] The i-th prior sub-block is decomposed using an independent principal component analysis model, and is represented as a linear combination of unknown independent components , represented as:

[0037] ;

[0038] In the formula, represents a mixing matrix, represents an independent component matrix, and E represents a residual matrix;

[0039] Neglecting the residual matrix, the independent principal component analysis model corresponding to the i-th prior sub-block is represented as:

[0040] ;

[0041] Then the independent component matrix estimated by the historical deformation data is represented as:

[0042] ;

[0043] In the formula, ​represents the estimated independent component matrix of the original deformation monitoring data based on the FastICA algorithm, and the prior demixing matrix is solved .

[0044] In some embodiments, the posterior statistic of each to-be-analyzed sub-block is calculated in combination with the prior demixing matrix statistic, including:

[0045] The posterior statistic of each to-be-analyzed sub-block is calculated, and the expression is:

[0046] ;

[0047] wherein, represents the posterior statistic of the i-th to-be-analyzed sub-block, represents the newly input sample data of the i-th to-be-analyzed sub-block, represents the prior demixing matrix, T represents the matrix transpose, represents the calculated independent component matrix; The posterior

[0048] statistic of each to-be-analyzed sub-block is calculated, and the expression is:

[0049] ;

[0050] wherein, represents the posterior statistic of the i-th to-be-analyzed sub-block, is obtained by projecting the new sample

[0051] to the independent component space learned by the model and reconstructing according to the independent components, and the expression is:

[0052] .

[0053] In some embodiments, the first type of control limit of each of the prior sub-blocks is obtained by performing a univariate kernel density estimation on the corresponding prior statistic;

[0054] The second type of control limit of each of the prior sub-blocks is obtained by performing a univariate kernel density estimation on the corresponding prior statistic.

[0055] In some embodiments, the posterior statistic of each to-be-analyzed sub-block and the posterior​ statistic, obtaining a Bayesian combined statistic, comprising:

[0056] the posterior statistic and the posterior statistic of each of the sub-blocks to be analyzed is converted into a conditional probability, expressed as:

[0057] ;

[0058] ;

[0059] ;

[0060] ;

[0061] wherein, represents the deformation data to be analyzed of the i-th sub-block to be analyzed, F represents a fault state, and N represents a normal state, represents the SPE confidence limit of the i-th sub-block to be analyzed corresponding to the prior sub-block, represents the SPE confidence limit of the i-th sub-block to be analyzed corresponding to the prior sub-block, ; represents the posterior statistic of the i-th sub-block to be analyzed, represents the posterior statistic of the i-th sub-block to be analyzed; represents the probability of being normal based on the statistic index ; represents the prior probability of being abnormal based on the statistic index ; represents the probability of being normal based on the SPE statistic index ; represents the prior probability of being abnormal based on the SPE statistic index ;

[0062] then the posterior probability of being abnormal can be expressed as:

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] wherein, represents the probability of being normal based on the statistical quantity index posterior probability of abnormality, representing probability of abnormality, based on statistical quantity index probability of abnormality; representing the SPE statistical quantity index posterior probability of abnormality, representing the SPE statistical quantity index probability of abnormality;

[0068] based on the preset significance level , define as , define as , the Bayesian comprehensive statistical quantity expression is:

[0069] ;

[0070] wherein B represents the number of the sub-blocks to be analyzed.

[0071] In some embodiments, the posterior probability of abnormality of each sub-block to be analyzed is compared with the first control limit and the second control limit. statistical quantity and the posterior statistical quantity monitors the local deformation risk and the occurrence time thereof, including:

[0072] marking the position higher than the first control limit in the posterior statistical quantity of each sub-block to be analyzed as having a local deformation risk and marking the corresponding risk period;

[0073] and marking the position higher than the second control limit in the posterior statistical quantity of each sub-block to be analyzed as having a local deformation risk and marking the corresponding risk period;

[0074] The Bayesian comprehensive statistical quantity is compared with the preset significance level value to monitor the global deformation risk and the occurrence time thereof, including marking the position higher than the significance level value in the Bayesian comprehensive statistical quantity as having a global deformation risk and marking the corresponding risk period.

[0075] In another aspect, the present application also provides a computer readable storage medium having computer programs / instructions stored thereon, which, when executed by a processor, implement the steps of the above method.

[0076] In another aspect, the present application also provides a computer program product comprising computer programs / instructions, characterized in that the computer programs / instructions, when executed by a processor, implement the steps of the above method.

[0077] The beneficial effects of the present application are at least:

[0078] The railway subgrade deformation risk identification method and device provided by the present application divide the historical deformation data of each monitoring point along the target railway into blocks according to the distribution similarity based on the grey wolf algorithm, and respectively establish independent principal component analysis models to obtain prior knowledge, obtain the prior solution mixing matrix corresponding to each prior sub-block, and obtain the control limit corresponding to the prior statistic and the prior statistic; when processing the newly generated data of each monitoring point, the prior sub-block obtained based on the historical deformation data is used for grouping, and the corresponding prior solution mixing matrix is used for quickly calculating the posterior statistic and the posterior statistic, and the control line obtained in the prior knowledge is used for local risk monitoring, and the posterior statistic and the posterior statistic, and the control line obtained in the prior knowledge is used for local risk monitoring, and the posterior statistic and the posterior statistic, and the control line obtained in the prior knowledge is used for local risk monitoring, and the posterior

[0079] The additional advantages, objects, and features of the present application will be in part apparent and in part pointed out hereinafter in the description, and will be learned from a reading of the following specification and by practicing the present application according to the following best mode thereof. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereinafter.

[0080] It will be understood by those skilled in the art that the objects and advantages of the present application can be realized and attained by the structure particularly pointed out in the written description and claims hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0081] The drawings described herein are intended to provide a further understanding of the present application, and constitute a part of this application, and do not constitute a limitation of the present application. In the drawings:

[0082] Figure 1 The flowchart of the railway subgrade deformation risk identification method according to an embodiment of the present application is shown.

[0083] Figure 2 The monitoring point distribution diagram in the railway subgrade deformation risk identification method according to an embodiment of the present application is shown.

[0084] Figure 3 The logic diagram of the railway subgrade deformation risk identification method according to an embodiment of the present application.

[0085] Figure 4 Fig. 2 is a diagram of deformation displacement variation of monitoring points 1-10.

[0086] Figure 5 Fig. 3 is a diagram of probability density fitting of deformation displacement of monitoring points 1-10.

[0087] Figure 6 Fig. 4 is ten independent components of deformation displacement of monitoring points 1-10 separated by the ICA method.

[0088] Figure 7 Fig. 5 is a diagram of the statistic of the ICA model of sub-block 1. Figure 6 Fig. 6 is a diagram of the statistic of the ICA model of sub-block 2. Fig. 7 is a diagram of the statistic of the ICA model of sub-block 3. Figure 6 Fig. 8 is a diagram of the statistic of the ICA model of sub-block 4. Fig. 9 is a diagram of the statistic of the ICA model of sub-block 5.

[0089] Figure 8 Fig. 5 is a diagram of the statistic of the ICA model of sub-block 1. Fig. 6 is a diagram of the statistic of the ICA model of sub-block 2. Figure 8 Fig. 7 is a diagram of the statistic of the ICA model of sub-block 3. Fig. 8 is a diagram of the statistic of the ICA model of sub-block 4. Figure 9 Fig. 5 is a diagram of the statistic of the ICA model of sub-block 1. Fig. 6 is a diagram of the statistic of the ICA model of sub-block 2.

[0090] Figure 9 Fig. 7 is a diagram of the statistic of the ICA model of sub-block 3. Fig. 8 is a diagram of the statistic of the ICA model of sub-block 4.

[0091] Figure 10 Fig. 5 is a diagram of the statistic of the ICA model of sub-block 1. Fig. 6 is a diagram of the statistic of the ICA model of sub-block 2. Figure 10 Fig. 7 is a diagram of the statistic of the ICA model of sub-block 3. Fig. 8 is a diagram of the statistic of the ICA model of sub-block 4.

[0092] Figure 11 Fig. 5 is a diagram of the statistic of the ICA model of sub-block 1. Fig. 6 is a diagram of the statistic of the ICA model of sub-block 2. Figure 11 Fig. 7 is a diagram of the statistic of the ICA model of sub-block 3. Fig. 8 is a diagram of the statistic of the ICA model of sub-block 4.

[0093] Figure 12 Fig. 5 is a diagram of the statistic of the ICA model of sub-block 1. Fig. 6 is a diagram of the statistic of the ICA model of sub-block 2. Figure 12 Fig. 7 is a diagram of the statistic of the ICA model of sub-block 3. Fig. 8 is a diagram of the statistic of the ICA model of sub-block 4.

[0094] Figure 13A BIC statistic diagram of the points 1-10 is monitored. DETAILED DESCRIPTION

[0095] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the embodiments and drawings. Herein, the illustrative embodiments of the present application and their descriptions are used to explain the present application, but are not as limitations to the present application.

[0096] Herein, it also needs to be explained that, in order to avoid the present application being obscured by unnecessary details, only the structures and / or processing steps closely related to the solutions according to the present application are shown in the drawings, and other details not closely related to the present application are omitted.

[0097] It should be emphasized that the term "comprises / comprising" as used herein is used to indicate the presence of a feature, element, step or component, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0098] The existing railway subgrade deformation monitoring scheme is mostly independent analysis for a single monitoring point, focuses on the monitoring of local deformation risk, and lacks overall systematic evaluation. In the face of the expanding railway network, the demand for large-scale systematic deformation risk assessment is increasingly prominent, and the demand for rapid and efficient deformation risk assessment detection capability is increasingly high.

[0099] One aspect of the present application provides a railway subgrade deformation risk identification method, as shown in Figure 1 The method comprises the following steps S101-S107:

[0100] Step S101: acquiring historical deformation data of a target railway collected by each monitoring point in cooperation with a reference station; the reference station and the monitoring points are deployed at preset positions along the target railway.

[0101] Step S102: grouping the historical deformation data of each monitoring point into multiple candidate sub-blocks, constructing multiple grey wolves as candidate sub-block corresponding grouping solutions based on the grey wolf algorithm, minimizing the sum of the KL divergence of the historical deformation data of each sub-block in the solution, and solving to obtain multiple prior sub-blocks of the optimal grouping.

[0102] Step S103: constructing an independent principal component analysis model for each prior sub-block, and outputting the prior statistic and its first type control limit, prior statistic and its second type control limit corresponding to each prior sub-block, and recording the prior solution confusion matrix corresponding to each prior sub-block.

[0103] Step S104: acquiring the deformation data to be analyzed collected by the monitoring points in a target period, and constructing the to-be-analyzed sub-block according to the grouping of the prior sub-block.

[0104] Step S105: Calculate the posteriori statistic of the deformation data to be analyzed of each sub-block to be analyzed in combination with the priori demixing matrix statistic and posteriori statistic.

[0105] Step S106: Fuse the posteriori of each sub-block to be analyzed by Bayesian synthesis inference statistic and posteriori statistic, to obtain the Bayesian synthesis statistic.

[0106] Step S107: Compare the posteriori of each sub-block to be analyzed in combination with the first type control limit and the second type control limit statistic and posteriori statistic to monitor the local deformation risk and its occurrence time; combine the preset significance level value to monitor the global deformation risk and its occurrence time.

[0107] In step S101, as shown in the figure, monitoring points are set for the roadbed or the building itself of the target railway, for collecting displacement data as deformation data by Beidou positioning, GPS positioning or other positioning forms. In this process, in order to improve the detection accuracy, a reference station is introduced for auxiliary detection. Figure 2

[0108] In some embodiments, the historical deformation data and the deformation data to be analyzed are displacement data sequences collected at a set frequency.

[0109] In step S102, referring to Figure 3 , the historical deformation data generated by each monitoring point is used to obtain priori knowledge in offline mode. The risk hidden danger area can be regarded as a whole, and there is some kind of association. In order to improve the data utilization rate and the stability and robustness of the deformation risk monitoring, the historical deformation monitoring data generated by multiple monitoring points can be divided into several sub-blocks according to the distribution characteristics, and independent principal component analysis models (ICA early warning models) are established for each sub-block to obtain the priori demixing matrix, the control limit of the statistic, the control limit of the statistic, which are used as priori knowledge for subsequent online early warning process.

[0110] The monitoring points are grouped in pairs to construct sub-blocks. The historical deformation data is used to construct priori sub-blocks in offline mode, and the grouping generated in the subsequent analysis process is the sub-block to be analyzed. The sub-block to be analyzed is directly generated based on the grouping result of the priori sub-block, without additional processing process, and therefore the priori knowledge obtained from the historical deformation data can be directly applied.

[0111] ​The core purpose of the block is to distribute the data of the monitoring points with similar data distribution characteristics into a class and monitor them at the same time to optimize the detection efficiency.

[0112] In some embodiments, a plurality of gray wolves are constructed based on the gray wolf algorithm as solutions of grouping corresponding to the candidate sub-blocks, the sum of the KL divergences of the historical deformation data in each sub-block is minimized, and the plurality of prior sub-blocks of the optimal grouping are obtained by solving.

[0113] Step S201: Initialize the gray wolf group, and each gray wolf represents a potential solution of grouping the historical deformation data of each monitoring point into a plurality of sub-blocks.

[0114] Step S202: Calculate the sum of the KL divergences of the historical deformation data in each sub-block in the potential solution corresponding to each gray wolf as the fitness function; find the gray wolf with the highest fitness function, the gray wolf with the second highest fitness function, the gray wolf with the third highest fitness function, and the remaining gray wolves as the gray wolves.

[0115] Step S203: In the hunting stage, each gray wolf updates its position according to the position of the wolf in this round, and the expression is as follows:

[0116] ;

[0117] ;

[0118] ;

[0119] ;

[0120] wherein, represents the position of the prey, X(t) represents the position of the gray wolf in the tth round, D represents the distance to be moved by the gray wolf, and A and C are both cooperation coefficients; represents a convergence factor, which is linearly decayed from 2 to 0 according to the number of iterations; and represent random numbers between 0 and 1;

[0121] Step S204: In the hunting stage, each gray wolf updates its position according to the position of the wolf in this round, the wolf, the wolf, the wolf, and the expression is as follows:

[0122] ;

[0123] ; ​​​​​

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] ;

[0129] wherein, denotes the current position of the wolf, denotes the current position of the wolf, denotes the current position of the wolf; denotes the position of each grey wolf individual; , , denotes a random number between 0 and 2; denotes the distance of the wolf from the prey, denotes the distance of the wolf from the prey, denotes the distance of the wolf from the prey; , , denotes the distance coefficient vector; denotes the updated position of the wolf, denotes the updated position of the wolf, denotes the updated position of the wolf; is the position of the grey wolf individual after the th iteration.

[0130] Step S205: updating the social rank of the grey wolf swarm.

[0131] Step S206: the algorithm ends when the maximum number of iterations is reached or the KL divergence and converge to a set threshold.

[0132] wherein, the KL divergence is calculated as follows:

[0133] ;

[0134] wherein, denotes the KL divergence, and denotes two historical deformation data random variables in each sub-block group The probability distribution of the two distributions is similar, the closer the KL divergence is to 0, and vice versa.

[0135] In step S103, the distribution probability of the target railway deformation data is obtained by referring to the historical deformation data of the target railway. Figure 3 For the plurality of prior sub-blocks obtained after grouping, an independent principal component analysis model (ICA early warning model) is established for each prior sub-block.

[0136] In some embodiments, an independent principal component analysis model is constructed for each prior sub-block, and the prior statistic and its first type control limit, prior statistic and its second type control limit corresponding to each prior sub-block are output, and the prior demixing matrix corresponding to each prior sub-block is recorded, including steps S301-S303:

[0137] Step S301: For the historical deformation data of the target railway containing m variables and n samples, the data set is divided into p prior sub-blocks, and each prior sub-block is represented as ; the variables in the i-th prior sub-block are represented as , the i-th prior sub-block.

[0138] Step S302: Decompose the i-th prior sub-block using the independent principal component analysis model, and represent as a linear combination of unknown independent components , represented as:

[0139] ;

[0140] In the formula, is a mixing matrix, is an independent component matrix, and E is a residual matrix.

[0141] Step S303: Ignore the residual matrix, and the independent principal component analysis model corresponding to the i-th prior sub-block is represented as:

[0142] ;

[0143] Then the independent component matrix estimated from the historical deformation data is represented as:

[0144] ;

[0145] In the formula, is an independent component matrix estimated from the original deformation monitoring data based on the FastICA algorithm, and the prior demixing matrix is obtained by solving.​

[0146] In some embodiments, the first type of control limit of each prior sub-block is obtained by performing a univariate kernel density estimation on the corresponding prior statistic.

[0147] The second type of control limit of each prior sub-block is obtained by performing a univariate kernel density estimation on the corresponding prior statistic.

[0148] Assume is an independent and identically distributed sample of random variable y, the mathematical description of the univariate kernel density estimation is as follows:

[0149] ;

[0150] In the formula, is the kernel density estimation of the unknown probability density function , m represents the sample size, represents the process data of the sub-block, represents the kernel function, represents the smoothing parameter or bandwidth. For each sub-block, all the statistics and their corresponding control limits are established, and if the statistic is less than the corresponding control limit, it indicates that the overall region represented by the sub-block is stable, otherwise it indicates that there is a certain risk, that is:

[0151] ;

[0152] .

[0153] The prior statistic of each prior sub-block and the prior statistic are calculated by performing a univariate kernel density estimation, which is used as the corresponding control limit, that is, the first type of control limit and the second type of control limit .

[0154] In step S104, referring to Figure 3 , in the online process, the monitoring points collect data to perform the railway subgrade deformation risk discrimination task, and constantly generate new deformation data to be analyzed. Based on prior knowledge, the monitoring points are grouped according to the prior sub-block, and the corresponding deformation data to be analyzed is quickly solved by using the prior disentangling matrix to obtain the posterior statistic and the posterior statistic, and the local deformation risk is monitored based on the first type of control limit and the second type of control limit .

[0155] In some embodiments, in combination with the prior disentangling matrix, the posterior Statistical quantity and posterior Statistical quantity, including steps S401-S402:

[0156] Step S401: Calculate the posterior of each sub-block to be analyzed Statistical quantity, expressed as:

[0157] ;

[0158] Wherein, represents the posterior of the i-th sub-block to be analyzed Statistical quantity, represents the newly input sample data of the i-th sub-block to be analyzed, represents the prior unmixed matrix, and T represents the matrix transpose, represents the calculated independent component matrix. Step S402: Calculate the posterior of each sub-block to be analyzed

[0159] Statistical quantity, expressed as:

[0160] ;

[0161] Wherein, represents the posterior of the i-th sub-block to be analyzed Statistical quantity, is obtained by projecting the new sample

[0162] to the independent component space learned by the model and reconstructing according to the independent components, expressed as:

[0163] .

[0164] In step S106, with reference to Figure 3 , based on the posterior of each sub-block to be analyzed calculated in steps S401-S402 Statistical quantity and posterior Statistical quantity, calculate the Bayesian comprehensive statistical quantity to reflect the overall deformation of the target railway section.

[0165] In some embodiments, the posterior Statistical quantity and posterior Statistical quantity of each sub-block to be analyzed are fused by Bayesian comprehensive reasoning to obtain a Bayesian comprehensive statistical quantity, including steps S501-S502:

[0166] Step S501: Convert the posterior Statistical quantity and the posterior Statistical quantity of each sub-block to be analyzed into conditional probability, expressed as: ​

[0167] ;

[0168] ;

[0169] ;

[0170] ;

[0171] wherein, represents the deformation data to be analyzed of the i-th sub-block to be analyzed, F represents a fault state, and N represents a normal state, represents the SPE confidence limit of the i-th sub-block to be analyzed corresponding to the prior sub-block, represents the SPE confidence limit of the i-th sub-block to be analyzed corresponding to the prior sub-block; represents the posterior statistic of the i-th sub-block to be analyzed, represents the posterior statistic of the i-th sub-block to be analyzed; represents the probability of being normal based on statistic index ; represents the prior probability of being abnormal based on statistic index ; represents the probability of being normal based on SPE statistic index ; represents the prior probability of being abnormal based on SPE statistic index ;

[0172] then the posterior probability of being abnormal can be represented as:

[0173] ;

[0174] ;

[0175] ;

[0176] ;

[0177] wherein, represents the posterior probability of being abnormal based on statistic index ; represents the probability of being abnormal, the probability of being abnormal based on statistic index ; ​​Bayesian comprehensive statistics posterior probability of abnormality, Bayesian comprehensive statistics probability of abnormality;

[0178] Step S502: based on the preset significance level , define as , define as , the Bayesian comprehensive statistics expression is:

[0179] ;

[0180] Wherein, B represents the number of sub-blocks to be analyzed.

[0181] In step S107, the posterior statistics and the posterior statistics of each sub-block to be analyzed are compared with the first type control limit and the second type control limit, the local deformation risk and its occurrence time are monitored, including steps S601-S602:

[0182] Step S601: mark the place higher than the first type control limit in the posterior statistics of each sub-block to be analyzed as existing local deformation risk and mark the corresponding risk period.

[0183] Step S602: mark the place higher than the second type control limit in the posterior statistics of each sub-block to be analyzed as existing local deformation risk and mark the corresponding risk period.

[0184] The Bayesian comprehensive statistics is compared with the preset significance level value, the global deformation risk and its occurrence time are monitored, including step S701: mark the place higher than the significance level value in the Bayesian comprehensive statistics as existing global deformation risk and mark the corresponding risk period.

[0185] In another aspect, the application also provides a computer readable storage medium, which stores computer programs / instructions, and the computer programs / instructions are executed by a processor to realize the steps of the above method.

[0186] In another aspect, the application also provides a computer program product, which comprises computer programs / instructions, and the computer programs / instructions are executed by a processor to realize the steps of the above method.

[0187] The application will be described below in combination with simulation experiments:

[0188] To verify that the proposed method can effectively realize the active early warning of railway infrastructure, the typical section of infrastructure is selected for data analysis. This section of infrastructure is located in the North China Plain, with complex geological conditions. In order to prevent major risk hazards and ensure safe railway transportation, the railway safety management department has set up 1 reference station and 10 monitoring stations in the risk hazard area to monitor the displacement of the infrastructure. The maximum distance between the reference station and the monitoring station is not more than 300 meters, and the monitoring stations are evenly distributed on both sides of the railway, Figure 2 The monitoring data time span is from September 1, 2022 to October 10, 2022, a total of 40 days, the data acquisition frequency is 10 seconds per epoch, a calculation cycle is 10 minutes, and a total of 5760 data points.

[0189] Figure 4 (a)-(j) are the original data time series of each monitoring point in the risk hazard area. From the figure, it can be seen that the monitoring data can reflect the deformation trend of the railway infrastructure in the 40 days, and the more serious burr characteristics caused by external environmental factors can also be observed. From the figure, it can be observed that most of the monitoring points change relatively smoothly in 40 days, but monitoring point 5 and monitoring point 9 still show a relatively obvious subsidence trend. It is also important for railway safety management personnel to determine the overall safety state of the risk area through the state information of each monitoring point. Based on this, the present application proposes to realize the systematic risk identification of the risk area based on Bayesian and ICA theory.

[0190] First, the statistical characteristics of each group of monitoring data in the risk area are analyzed, Figure 5 (a)-(j) are the probability density fitting graphs of each monitoring point, where the blue column represents the probability density distribution of the original data, the red color represents the traditional Gaussian probability density distribution fitting, and the green color represents The fitting parameter values of each monitoring point stable distribution. From Figure 5 (a)-(j) can be seen, the stable distribution fitting can better describe the real probability density distribution of the data, The stable distribution curve and the traditional Gaussian distribution curve do not completely coincide, which shows that each group of monitoring data in this area does not obey the Gaussian distribution. From the fitting results, each group of data does not show serious tailing or peak characteristics, which shows that the data in this risk area has not undergone serious deformation, which is consistent with the result that each group of monitoring data time series changes relatively smoothly. The The stable distribution parameters are all not equal to 2, further verifying that the monitoring data of each group follow a non-Gaussian distribution. For this type of non-Gaussian distributed monitoring data, ICA modeling can be considered to identify the abnormal state of the potential hazard area.

[0191] Table 1 Probability density fitting parameters

[0192]

[0193] Figure 6 The graph shows 10 independent components separated from 10 sets of monitoring data using the ICA method. The overall trend of these 10 independent components is relatively flat, but sets ICA-1, ICA-2, and ICA-3 showed significant oscillations on September 11th, indicating that these three components experienced substantial disturbances within their respective regions. The ICA model was constructed using the previous 30 days' data as the training set to obtain... and The control limits are used to determine the stability of the entire region. Then, the data from the next 10 days is used as a test set to verify and determine the stability of the region. The principle of ICA is actually to calculate the statistic at each time point and compare it with the control limits. If the statistic exceeds the corresponding control limits, it indicates that an anomaly has occurred in the region.

[0194] Figure 7 (a) and (b) show the ICA modeling analysis performed on 10 sets of monitoring data, in which... Figure 7 Figure (a) shows the monitoring data at different time points. Statistics, and Figure 7 Figure (b) shows the monitoring data at various time points. The black line in the graph represents the statistics obtained by the ICA method for the corresponding time, while the red line represents the control limits obtained from the training data. As can be seen from the graph, regardless of... Statistical measure or The statistical values, overall, did not completely exceed the corresponding control limits, indicating that the geology of the area is relatively stable. Furthermore, it can be clearly observed that around October 3rd... and The statistical values ​​all exceeded the control limits, indicating that the geological region was relatively active during this period. The results of the two statistical values ​​corroborate each other, further demonstrating the accuracy and reliability of the method. Furthermore, it can also be seen that... The statistical volume also exceeded the control limit around October 10th, and its Control limits relative The control limits are also relatively small, which indicates that Compared to It has higher sensitivity to detect abnormal signals.

[0195] To further improve the robustness of the ICA model and the utilization rate of monitoring data, the risk state of the hidden danger area is evaluated in multiple dimensions, and a distributed ICA model is established by grouping each group of monitoring data. The grey wolf algorithm model is used for optimization, first set the grey wolf pack, that is, the pairwise arrangement combination of each group of monitoring data, calculate the corresponding KL divergence, and get the grouping corresponding to the minimum KL divergence, that is, the head wolf, then iterate until all grouping results are selected, the purpose is to make the distribution of data in the group similar. The grouping results are shown in Table 2, a total of 10 groups of monitoring data, which are evenly divided into 5 sub-blocks, that is, monitoring points 5 and 6 form a group, monitoring points 1 and 10 form a group, monitoring points 2 and 4 form a group, monitoring points 8 and 9 form a group, and monitoring points 3 and 7 form a group.

[0196] Table 2 Grouping results of each monitoring point in the hidden danger area

[0197]

[0198] Figures 8 to 12 To the five groups of monitoring data corresponding and statistics, it can be observed from the figure that the monitoring data in each group does not appear to exceed the control limit in a large range, which is basically consistent with the ICA model results of the above 10 groups of monitoring data. In addition, it can also be observed from the figure that the statistics of each group and the 10 groups also present local differences, such as the statistics of sub-block 4 appear abnormal around October 8, while the Figure 8 statistics do not appear obvious abnormal in this section, the reason is that Figure 8 statistics is a risk discrimination index based on the entire hidden danger area, while each group is based on two groups of variables, so it can better reflect the specific details. In addition, the and control limits of each sub-block are different from the control limits shown in Figure 7 , the reason is that the statistical limits of each sub-block are based on the training data of each sub-block.

[0199] Figure 13 The control limit of the distributed ICA model established according to the statistics of the five sub-blocks is the corresponding confidence. The red line in the figure is the corresponding confidence control limit, and the black line is the statistics corresponding to each time, it can be observed from the figure that the statistics in each group also does not appear to exceed the control limit in a large range, which is basically consistent with the ICA model results of the above 10 groups of monitoring data. In addition, it can also be observed from the figure that the distributed ICA model also appears abnormal on October 8 in addition to the local abnormalities on October 3 and October 10, which shows that the distributed ICA model can reflect the local abnormal information of each group compared with the traditional ICA model. In addition, from the And Statistically, the area was indeed more active on October 3, but the overall still did not reach the more serious standard. In fact, the control limit is only a threshold for alarm, and the distance between the statistics and the control limit can also be used to determine the stability of the overall risk area. Whether it is a traditional ICA model or a distributed ICA model, there are individual discrete points exceeding the control limit threshold, but for the overall risk area, a continuous number of statistics exceeding the control limit threshold can be set, and then the overall area is determined to enter a more serious risk state.

[0200] Corresponding to the above method, the application also provides a device / system, which comprises a computer device including a processor and a memory, the memory storing computer instructions, and the processor being configured to execute the computer instructions stored in the memory, so that the device / system implements the steps of the above method.

[0201] The embodiment of the application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above edge computing server deployment method. The computer readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0202] In summary, the railway subgrade deformation risk identification method and device provided by the application divide the historical deformation data of each monitoring point along the target railway into blocks according to the distribution similarity based on the grey wolf algorithm, and independently establish independent principal component analysis models to obtain prior knowledge, obtain the prior solution mixing matrix corresponding to each prior sub-block, and obtain the control limit corresponding to the prior Statistics and prior Statistics; when processing newly generated data of each monitoring point, grouping is performed according to the prior block obtained based on the historical deformation data, and the corresponding prior solution mixing matrix is used to quickly calculate the posterior Statistics and posterior Statistics, and the control line obtained in the prior knowledge is combined to perform local risk monitoring on the posterior Statistics and posterior Statistics, and the control line obtained in the prior knowledge is combined to perform local risk monitoring on the posterior Statistics and posterior The statistical quantity is fused based on Bayesian comprehensive reasoning, realizes the systematic risk identification of the local and global of the hidden danger area, and improves the discrimination efficiency and accuracy.

[0203] Those of ordinary skill in the art will appreciate that the various illustrative components, systems and methods described in connection with the embodiments disclosed herein can be implemented as hardware, software, or a combination thereof. The various illustrative components, systems and methods disclosed herein can be implemented in the context of a general computer or in the context of other computing devices including special purpose computers. The particular sequences of operations described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination thereof, as discussed above. The particular sequences of operations described in connection with the embodiments disclosed herein can be implemented on a single computer or distributed across several computers depending upon the circumstances commensurate with a specific design constraint. The described embodiments are not limited to any particular type of computer or network.

[0204] It is to be understood that the present application is not limited to the particular configurations, processing steps, and materials described herein and illustrated in the drawings. Well-known methods have not been described or illustrated in detail in order to not unnecessarily obscure the present application. In the above-described embodiments, a number of specific steps are described and illustrated as examples. However, the methods of the present application are not limited to the specific steps described and illustrated, and one of ordinary skill in the art can make various changes, modifications, and additions, or can change the order of steps, without departing from the spirit of the present application.

[0205] In the present application, features described and / or illustrated in connection with one embodiment can be used in the same or a similar way or in conjunction with or in place of features of another embodiment.

[0206] The above description is only preferred embodiments of the present application, and is not intended to limit the present application. The embodiments of the present application can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the present application.

Claims

1. A railway subgrade deformation risk identification method, characterized in that, The method comprises the following steps: acquiring historical deformation data of a target railway collected by each monitoring point in cooperation with a reference station; the reference station and the monitoring points are arranged at preset positions along the target railway; grouping the historical deformation data of each monitoring point into multiple candidate sub-blocks, constructing multiple wolves based on a grey wolf algorithm as solutions of the corresponding groups of the candidate sub-blocks, minimizing the sum of the KL divergences of the historical deformation data in the solutions, and solving to obtain multiple prior sub-blocks of an optimal group; constructing an independent principal component analysis model for each of the prior sub-blocks, and outputting a prior statistic and a first type control limit thereof, a prior statistic and a second type control limit thereof, and recording a prior disentangling matrix corresponding to each of the prior sub-blocks; wherein the first type control limit of each of the prior sub-blocks is obtained by performing a univariate kernel density estimation on the corresponding prior statistic; the second type control limit of each of the prior sub-blocks is obtained by performing a univariate kernel density estimation on the corresponding prior statistic; the second type control limit of each of the prior sub-blocks is obtained by performing a univariate kernel density estimation on the corresponding prior acquiring deformation data to be analyzed collected by the monitoring points in a target period, and constructing sub-blocks to be analyzed according to the grouping of the prior sub-blocks; combining the a priori unmixing matrix, calculating a posteriori statistics and a posteriori statistics combining the posteriors of the sub-blocks to be analyzed by Bayesian synthesis inference the statistics and the posteriors combining the statistics to obtain a Bayesian synthesis statistics combining the first type of control limit and the second type of control limit to compare the posterior of each sub-block to be analyzed statistic and posterior The statistic monitors the local change risk and its occurrence time; combining the preset significance level value, the Bayesian comprehensive statistic monitors the global change risk and its occurrence time; Wherein the posterior statistics and the posterior statistics, obtaining a Bayesian integrated statistic, comprising: said posterior statistics and said posterior statistics into conditional probabilities, expressed as: ； ； ； ； In the formula, Let F represent the deformation data to be analyzed for the i-th sub-block, F represent the fault state, and N represent the normal state. This represents the SPE confidence limit of the prior sub-block corresponding to the i-th sub-block to be analyzed. This indicates that the prior sub-block corresponding to the i-th sub-block to be analyzed is Confidence limit; The posterior represents the i-th sub-block to be analyzed. Statistic, The posterior represents the i-th sub-block to be analyzed. Statistic; Indicates based on Statistical indicators Normal probability Indicates based on Statistical indicators The prior probability of an anomaly; Indicates based on SPE statistic indicators Normal probability Indicates based on SPE statistic indicators The prior probability of an anomaly; Then The posterior probability of abnormality can be expressed as: ； ； ； ； wherein, denotes the posterior probability of an anomaly based on statistical quantity indicator the posterior probability of an anomaly, denotes the probability of an anomaly, based on statistical quantity indicator the probability of an anomaly; denotes the posterior probability of an anomaly based on the SPE statistical quantity indicator the posterior probability of an anomaly, denotes the probability of an anomaly based on the SPE statistical quantity indicator the probability of an anomaly; based on the preset significance level , define as , define as , then the Bayesian synthesis statistical expression is: ； wherein B represents the number of the sub-blocks to be analyzed.

2. The method of claim 1, wherein the method further comprises: The historical deformation data and the deformation data to be analyzed are displacement data sequences collected at a set frequency.

3. The method of claim 1, wherein the method further comprises: The multiple wolves are constructed based on the grey wolf algorithm as the solutions of the corresponding groups of the candidate sub-blocks, the sum of the KL divergences of the historical deformation data in the solutions is minimized, and multiple prior sub-blocks of an optimal group are solved, comprising: initializing a wolf group, and each wolf represents a potential solution of grouping the historical deformation data of each monitoring point into multiple sub-blocks; The KL divergence of the historical deformation data in each sub-block in the potential solution corresponding to each gray wolf is calculated and summed as a fitness function; the gray wolf with the highest fitness function is found according to the fitness functions corresponding to the gray wolves , the gray wolf with the second highest fitness function , the gray wolf with the third highest fitness function , and the remaining gray wolves as . In the hunting stage, each gray wolf updates its position according to the current round The position of the wolf is updated as follows: ； ； ； ； wherein, represents the position of the prey, X(t) represents the position of the gray wolf at t rounds, D represents the distance to be moved by the gray wolf, and A and C are both coordination coefficients; represents a convergence factor, which is linearly attenuated from 2 to 0 according to the number of iterations; and represents a random number within 0 to 1; In the hunting phase, each gray wolf updates its position according to this round wolf, wolf, wolf updates its position according to the following expression: ； ； ； ； ； ； ； wherein, represents the current position of the wolf, represents the current position of the wolf, represents the current position of the wolf; represents the position of each grey wolf individual; , , represents a random number between 0 and 2; represents the distance of the wolf from the prey, represents the distance of the wolf from the prey, represents the distance of the wolf from the prey; , , represents the distance coefficient vector; represents the updated position of the wolf, represents the updated position of the wolf, represents the updated position of the wolf; is the position of the grey wolf individual after the th iteration; updating the social ranks of the wolf group; when the maximum number of iterations is reached or the sum of the KL divergences converges to a set threshold, the algorithm ends.

4. The method of claim 2, wherein the step of determining the risk of deformation of the railway subgrade is performed by using a risk index. constructing an independent principal component analysis model for each of the prior sub-blocks, and outputting a prior corresponding to each of the prior sub-blocks the statistical quantity, the first type control limit, and the prior the statistical quantity, the second type control limit, and the prior, and recording a prior unmixing matrix corresponding to each of the prior sub-blocks, comprising: For the historical deformation data of the target railway The dataset contains m variables and n samples, and is divided into p prior sub-blocks, each of which is represented as... The variables within the i-th prior sub-block are then represented as follows: , Indicates the i-th prior sub-block. One measurement variable; The i-th prior sub-block is decomposed using an independent component analysis model, yielding a linear combination of unknown independent components denoted by ； wherein denotes the mixing matrix, denotes the independent component matrix, E denotes the residual matrix; Neglecting the residual matrix, the independent principal component analysis model corresponding to the ith prior sub-block is represented as: ； then the independent component matrix estimated by the historical deformation data is represented as: ； In the formula, represents the estimated independent component matrix of the original deformation monitoring data based on the FastICA algorithm, and the prior solution mixing matrix is solved .

5. The railway subgrade deformation risk identification method according to claim 4, characterized in that, combining the a priori unmixing matrix, calculating a posteriori statistic and a posteriori statistic, comprising: computing the posterior for each sub-block to be analyzed a statistic, expressed by the formula: ； wherein represents the posterior of the i-th sub-block to be analyzed statistic, represents the posterior of the i-th sub-block to be analyzed represents the newly input sample data of the i-th sub-block to be analyzed, represents the prior demixing matrix, T represents the matrix transpose, represents the calculated independent component matrix; computing the posterior for each sub-block to be analyzed a statistic, expressed by the formula: ； wherein, denotes the posterior of the i-th sub-block to be analyzed statistic, is obtained by combining the new sample the expression obtained by projecting onto the independent component space learned by the model and reconstructing according to the independent components is: 。 6. The method of claim 1, wherein: combining the first type of control limit and the second type of control limit to compare the posterior of each sub-block to be analyzed statistic and posterior The statistic monitors the risk of local deformation and the time of its occurrence, including: the posterior probability of each sub-block to be analyzed is calculated the statistical quantity above the first control limit is marked as a local deformation risk and the corresponding risk period is marked; and the posteriori probability of each sub-block to be analyzed is calculated the statistical quantity above the second control limit is marked as having a local deformation risk and the corresponding risk period is marked; combining a preset significance level value to compare the Bayesian comprehensive statistic to monitor the global deformation risk and the occurrence time thereof, comprising: marking the positions higher than the significance level value in the Bayesian comprehensive statistic as positions with a global deformation risk and marking the corresponding risk period.

7. A computer readable storage medium having stored thereon computer programs / instructions, characterized in that, The computer program / instructions are executed by the processor to implement the steps of the method of any one of claims 1 to 6.

8. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions are executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images