Non-insulation track circuit compensation capacitor fault diagnosis method

Abstract

The invention discloses an uninsulated track circuit compensation capacitor fault diagnosis method, which is used for carrying out compensation capacitor fault diagnosis research by jointly applying a signal detection system arranged on a high-speed comprehensive detection train and relying on track circuit information detected by a high-speed comprehensive detection vehicle. According to the research, data which is collected by a dynamic detection system and reflects the working state of a track circuit are fused mainly based on phase-space reconstruction, a pseudo-phase diagram which reflects the state characteristics of the system is obtained through phase-space reconstruction, and fault characteristics of a compensation capacitor are extracted, so that fault diagnosis of the compensation capacitor is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit signal system operation and maintenance, and in particular to a method for diagnosing faults of compensation capacitors in a non-insulated track circuit. Background Art

[0002] Track circuit is a key equipment for safe railway operation. It is responsible for section occupancy detection, train integrity inspection and transmission of train control information. The ZPW-2000A non-insulated track circuit system includes the main track circuit and tuning area, forming a complete signal transmission network.

[0003] During the signal transmission process, the transmitter generates a variety of modulated frequency-shifted low-frequency signals according to the coded instructions, and transmits the signals to tracks of different frequencies through the precise distribution of the cable network and matching reactors. After matching and tuning, the signal passes through the rail to the receiving end of the track circuit, and flows through the tuning unit and matching transformer back to the receiving end. At the same time, part of the signal is received by the receiver in the front section and forwarded to the short track. Based on the information of the main track and the short track, the track relay determines the occupancy status of the section, thereby ensuring the safety and efficiency of train operation. Failure of the compensation capacitor of the uninsulated track circuit is a common cause of the high failure rate of the track circuit. The compensation capacitor is easily damaged by aging and environmental factors in the outdoor environment, and these devices are widely distributed and large in number. Once a failure occurs, its detection and maintenance becomes a challenge.

[0004] At present, the detection work relies on manual inspection by railway site staff, analysis of track circuit receiving voltage, and regular inspection by electrical inspection vehicles. Although these methods provide a basis for track circuit maintenance, they have the following shortcomings:

[0005] (1) Manual investigation: It consumes human resources, is inefficient, and relies on the professional knowledge and experience of the operators.

[0006] (2) Receiving voltage analysis: Affected by the external environment, it lacks the ability to accurately locate faults.

[0007] (3) Regular inspections: The diagnostic timeliness and real-time performance are not ideal, which can easily lead to missed opportunities to discover and handle faults in a timely manner.

[0008] Therefore, the present invention proposes a non-insulated track circuit compensation capacitor fault diagnosis method based on phase space reconstruction multi-source data fusion, which is carried out in conjunction with the signal detection system equipped on the high-speed comprehensive detection train, and relies on the track circuit information detected by the high-speed comprehensive detection vehicle to carry out compensation capacitor fault diagnosis research. This study is mainly based on phase space reconstruction to fuse the data reflecting the working state of the track circuit collected by the dynamic detection system, and through phase space reconstruction, obtain a pseudo-phase diagram reflecting the system state characteristics, extract the fault characteristics of the compensation capacitor, and thus realize the fault diagnosis of the compensation capacitor. Summary of the invention

[0009] The purpose of the present invention is to provide a method for diagnosing the fault of a compensation capacitor of a non-insulated track circuit in order to solve the above-mentioned problem.

[0010] To achieve the above object, the technical solution adopted by the present invention is:

[0011] The following steps are involved:

[0012] Step 1: Acquire data. The selected data that can reflect the working state variables of the track circuit compensation capacitor include: compensation capacitor induction signal, carrier frequency, and frequency shift signals with carrier frequencies of 1700 Hz, 2000 Hz, 2300 Hz, and 2600 Hz;

[0013] Among them, the compensation capacitor sensing signal is expressed as , according to the carrier frequency, the frequency shift signal corresponding to the carrier frequency is ,in is the signal length;

[0014] Step 2: Data Preprocessing

[0015] Step 2.1: Perform maximum-minimum normalization processing on the compensation capacitance sensing signal CC and the frequency shift signal IV respectively to obtain the normalized compensation capacitance sensing signal CC_norm and the frequency shift signal IV_norm, as shown in formula (1):

[0016]

[0017] Step 2.2: Calculate the normalized compensation capacitance sensing signal CC_norm i and the frequency shift signal IV_norm i The isotropic processing is performed to obtain the compensated capacitance sensing signal CC_T and the frequency shift signal IV_T after isotropic processing, which are specifically shown in formula (2):

[0018]

[0019] Step 2.3: wavelet denoising, using wavelet transform to perform denoising on the assimilated compensation capacitor sensing signal CC_T and the frequency shift signal IV_T;

[0020] Specifically, the time series data is input into the wavelet transform denoising algorithm, and the BayesShrink method and soft thresholding are used at the three-layer wavelet decomposition level. The symmetric wavelet (sym7) is selected for decomposition and reconstruction, and the signal variance is recalibrated to ensure the stability and accuracy of the denoising effect, and the compensated capacitive sensing signal after denoising is obtained. With frequency shift signal ;

[0021] Step 3: Phase space reconstruction, using the mutual information method and GP algorithm to calculate the pre-processed compensation capacitor sensing signal one by one With frequency shift signal The delay time and embedding dimension of

[0022] Step 3.1: Mutual information method to calculate delay time

[0023] Define the signal to be analyzed as , where t represents the time index, N is the sequence length, and in order to measure the delay of the sequence under different time delays, the mutual information metric MI(T) is introduced to represent the signal A sequence delayed by T time units from itself The mutual information between them is calculated as follows:

[0024]

[0025] in, and Respectively represent the sequence and sequence The value space, is delaying When the observed and The joint probability of and yes and The marginal probability distribution of

[0026] Mutual Information It reflects the reduction in uncertainty about another part of the sequence when the information of one part is known. When the peak value is reached, it means that Under the delay of , the sharing between the historical information and future information of the sequence is the largest, so by finding The peak value of :

[0027]

[0028] The peak value indicates that Moment, time series The past value of The values ​​after the moment have the highest interdependence;

[0029] Step 3.2: GP algorithm to find the embedding dimension m

[0030] Setting distance threshold , and calculate the distances of all point pairs to obtain the distribution function of point pairs in phase space :

[0031]

[0032] in, is the total number of points in phase space, is the unit step function, Yes and The Euclidean distance between

[0033] Embedding Dimension By correlation dimension It is estimated that is defined as:

[0034]

[0035] In practice, different Value calculation , and plotted on a log-log coordinate right The picture, in and The slope of the linear fit is the correlation dimension , by choosing an appropriate slope, we can estimate the embedding dimension ,make sure To meet the requirements of phase space reconstruction;

[0036] Step 3.3: Determine the final delay time and embedding dimension

[0037] The compensated capacitance sensing signal is calculated based on the mutual information method And the frequency shift signal corresponding to the carrier frequency is selected according to the carrier frequency: The delay time is , the embedding dimension calculated by the GP algorithm is In order to ensure that the phase space of each variable can be fully expanded and the correlation between data is guaranteed without compressing the phase trajectory, the maximum embedding dimension is selected according to formula (7): and minimum delay time as phase space reconstruction parameters;

[0038]

[0039] Step 3.4: According to the final determined delay time and embedding dimension, use the delay embedding method to respectively compensate the capacitance sensing signal and the frequency-shifted signal corresponding to the carrier frequency Reconstruct the phase space and obtain the trajectory matrix of the two variables and As shown in formula (8) and formula (9):

[0040]

[0041]

[0042] Mode and middle, is the embedding dimension, is the delay time, is the number of state vectors obtained by phase space reconstruction, ;

[0043] Step 4: Fuse the trajectory matrices of each variable according to the adaptive weighted fusion algorithm;

[0044] Step 5: Feature extraction. After adaptive weighted fusion, the trajectory matrix Y is used to extract the pseudo-phase diagram that can reflect the state characteristics of the system by using singular value decomposition, thereby extracting the state characteristics of the compensation capacitor. The pseudo-phase diagram is a projection of the state change curve of the original system in the embedded phase space. There is a deterministic mapping relationship between the system state and the pseudo-phase diagram. Since the reconstructed phase space is usually multi-dimensional, it becomes a challenge to intuitively represent the multi-dimensional pseudo-phase diagram, which needs to be processed in order to extract the essential characteristics of the system.

[0045] Let the singular value decomposition of the fused trajectory matrix Y be as follows: As shown:

[0046]

[0047] in, is a diagonal matrix, as shown in the formula, , called matrix The singular values ​​of , the corresponding normalized and orthogonal eigenvectors are ,matrix and Each column vector is orthogonal to each other, so we have , The column vectors of the matrix are the fused matrix In the base The projection on express No. Column vector, based on the relationship between the pseudo phase diagram projection and the compensation capacitor position, extracts the state characteristics of the compensation capacitor;

[0048] Step 6: Fault diagnosis, pseudo phase diagram in The projection on the axis shows high sensitivity to different fault levels and locations of the compensation capacitor; when a fault occurs in the compensation capacitor, the projection amplitude at that location will decrease accordingly, and the severity of the fault is proportional to the proportion of the amplitude reduction. A threshold is set. To determine whether the compensation capacitor is faulty, If , the compensation capacitor is faulty; otherwise, the compensation capacitor is normal.

[0049] Furthermore, the adaptive weighted fusion algorithm includes several main steps: data segmentation, information entropy evaluation, fuzzy logic method adaptive weight allocation and weight fusion:

[0050] Step 4.1: Data chunking

[0051] First, define the block strategy; transform the trajectory matrix Y CC and Y IV Each column of is divided into blocks of size l; the part less than size l is also treated as an independent block;

[0052] Step 4.2: Information entropy evaluation

[0053] Assign weights to each block of each data source, calculate the information entropy of each block, evaluate the entropy difference between different data sources, and determine the "information value" of each block in different sources. The higher the information entropy value, the higher the uncertainty and complexity of the block. The information entropy function is defined as :

[0054]

[0055] Among them, p(x) is the probability of element x appearing in the data block;

[0056] Step 4.3: Fuzzy logic adaptive weight assignment

[0057] A fuzzy logic method is used to adaptively assign weights to each block. First, a fuzzy logic system and an associated set of fuzzy rules are established to map information entropy to the weights of data blocks. A fuzzy membership function is used to assign a subordinate measure to each observed entropy level. The membership function is defined according to each information complexity interval (low, medium, and high) and is expressed by the formula The following three functions are used to reflect this:

[0058]

[0059] Secondly, the weight is calculated based on the information entropy of each data block and its membership under each membership function. In the present invention, a data block with low entropy is assigned a higher weight, while a data block with high entropy is assigned a lower weight, which means that the present invention prefers data blocks with high certainty. By multiplying its membership with a predetermined weight factor and summing them, the weight W of each data block can be obtained as shown in formula (12):

[0060]

[0061] Among them, α, β, and γ are the corresponding weight factors, α=0.5, β=0.3, and γ=0.2.

[0062] Step 4.4: Weighted Fusion

[0063] Use step 4.3 to optimize the weight of each block, and perform weighted fusion to obtain the final fused trajectory matrix Y; first, the weights of each block corresponding to each variable are normalized according to formula (13):

[0064]

[0065] Secondly, the weighted fusion method is used to fuse each block of each variable. The weighted fusion formula is as follows: As shown:

[0066]

[0067] in, and denote the j-th block of the i-th column of variables CC and IV, respectively, and denote the weights of the jth block in the ith column of variables CC and IV, respectively;

[0068] Finally, the fused blocks are spliced ​​to obtain the final fused trajectory matrix Y. Compared with the prior art, the present invention has the following beneficial effects:

[0069] The present invention proposes a method for diagnosing compensating capacitor faults in non-insulated track circuits based on phase space reconstruction and multi-source data fusion, which provides a new fusion method for fusing multi-source data monitored by heterogeneous sensors, and can effectively solve the problems of data semantic mismatch, heterogeneous data features and information differences when performing data fusion.

[0070] The present invention proposes a non-insulated track circuit compensation capacitor fault diagnosis method based on phase space reconstruction and multi-source data fusion, which performs dimensionless processing on the compensation capacitor induction signal and the corresponding carrier frequency shift signal obtained by the high-speed comprehensive detection vehicle, thereby solving the problem that the measurement units and data levels of the two monitored signals are different when data fusion is performed.

[0071] The present invention proposes a method for diagnosing the fault of a non-insulated track circuit compensation capacitor based on multi-source data fusion of phase space reconstruction. Two variables that can reflect the working state of the track circuit compensation capacitor are fused through phase space reconstruction, and the state characteristics of the compensation capacitor are extracted based on the pseudo-phase diagram of the reconstructed phase space, so as to determine whether the compensation capacitor has a fault.

[0072] The present invention proposes a method for diagnosing the fault of the compensation capacitor of the non-insulated track circuit based on phase space reconstruction and multi-source data fusion. The method has been proved through verification experiments that the present invention can accurately determine the fault diagnosis of the compensation capacitor of the non-insulated track circuit, further promote the early warning and maintenance efficiency of the railway signal system, and bring theoretical and methodological innovations to railway safety and maintenance work. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 The main components and structures of the signal dynamic detection system;

[0074] Figure 2 Fault diagnosis process of compensation capacitor of non-insulated track circuit based on multi-source data fusion;

[0075] Figure 3 Comparison diagram of the induction signal and the frequency shift signal when there is no fault compensation capacitor in a certain section;

[0076] Figure 4 When there is no fault compensation capacitor in a certain section, the induction signal, frequency shift signal, and phase space trajectory diagram after multi-source information fusion;

[0077] Figure 5 The projection of the pseudo-phase diagram on the axis extracted from the phase space trajectory after multivariate information fusion when there is no fault compensation capacitor in a certain section.

[0078] Figure 6 Comparison of the induction signal and the frequency shift signal when a section contains a faulty compensation capacitor (C13 fault)

[0079] Figure 7 When a section contains a faulty compensation capacitor (C13 fault), the induction signal, frequency shift signal, and phase space trajectory diagram after multi-information fusion;

[0080] Figure 8 Pseudo-phase diagram extracted from the phase space trajectory after multi-source information fusion when a section contains a faulty compensation capacitor (C13 fault). DETAILED DESCRIPTION

[0081] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0082] The present invention is specifically implemented through the following technical solutions:

[0083] The signal dynamic detection system relies on two platforms, the electric service test vehicle and the high-speed comprehensive inspection vehicle. It can conduct comprehensive detection of the transmission characteristics, frequency characteristics, compensation capacitors, point transponders and operating status of locomotive signals of the track circuit under the dynamic operation conditions of the train. The signals collected by the dynamic detection system include the induced voltage signal of the track circuit, the induced signal of the compensation capacitor, the signal receiving code sequence, the unbalanced traction current, the transponder, the speed and other track circuit equipment information. Among them, the track circuit induced signal simultaneously collects the 50Hz industrial frequency signal of the section where it is located, as well as the frequency-shifted signals with carrier frequencies of 1700Hz, 2000Hz, 2300Hz and 2600Hz. The main components of the signal dynamic detection system are as follows: Figure 1 shown.

[0084] Among the signals collected by the signal dynamic detection system, the frequency shift signal contains the most track circuit information and is the data that best reflects the working status of the track circuit; the compensation capacitor sensing signal can also reflect whether the compensation capacitor has a fault. Therefore, the present invention selects the frequency shift signal and the compensation capacitor sensing signal as variables reflecting the working status of the track circuit compensation capacitor and as the data source for determining whether the compensation capacitor has a fault.

[0085] Phase space reconstruction (PSR), as an important tool for nonlinear time series analysis, has the ability to reconstruct the attractor of a dynamic system in a high-dimensional space. Through phase space reconstruction, an information space that reflects the dynamic characteristics of the time series can be obtained, which contains the dynamic behavior and structure of the system. In contrast, the physical meaning contained in the original actual sequence is blurred after phase space reconstruction. Therefore, the problem of multi-source data fusion can be regarded as a problem of multivariate phase space reconstruction. From this perspective, the measurement data of each sensor can be regarded as part of the system dynamics, and by integrating these data into a common phase space, the correlation and interaction between heterogeneous data can be captured. This method not only takes into account the dynamic characteristics of the data, but also comprehensively considers the information of multi-source data, thereby improving the interpretability and predictive ability of the fusion results.

[0086] According to the determined variables that can reflect the working state of the compensation capacitor, the delay time and embedding dimension are calculated one by one using phase space reconstruction. In order to ensure that the phase space of each variable can be fully expanded and the correlation between data is guaranteed without compressing the phase trajectory, the final delay time and embedding dimension are determined.

[0087] The variable monitoring data that can reflect the working state of the track circuit compensation capacitor is reconstructed in phase space to obtain a trajectory matrix of each variable.

[0088] The trajectory matrices of various variables are fused according to the adaptive weighted fusion method to complete the multi-source data fusion of various variables and obtain the fused trajectory matrix.

[0089] A pseudo phase diagram that can reflect the state characteristics of the system is extracted from the fused trajectory matrix, from which the state characteristics of the compensation capacitor are extracted.

[0090] Based on the state characteristics of the compensation capacitor extracted from the pseudo-phase diagram, the accuracy of the compensation capacitor fault diagnosis is verified.

[0091] The specific steps are as follows:

[0092] Step 1: Obtain data. The selected data that can reflect the working state variables of the track circuit compensation capacitor include: compensation capacitor induction signal, carrier frequency, and frequency shift signals with carrier frequencies of 1700Hz, 2000Hz, 2300Hz, and 2600Hz. The compensation capacitor induction signal is expressed as , according to the carrier frequency, the frequency shift signal corresponding to the carrier frequency is ,in is the signal length.

[0093] Step 2: Data preprocessing.

[0094] Step 2.1: Perform maximum-minimum normalization processing on the compensation capacitance sensing signal CC and the frequency shift signal IV respectively to obtain the normalized compensation capacitance sensing signal CC_norm and the frequency shift signal IV_norm, as shown in formula (1).

[0095]

[0096] Step 2.2: Calculate the normalized compensation capacitance sensing signal CC_norm i and the frequency shift signal IV_norm i After the isotropy processing is performed, the compensated capacitance sensing signal CC_T and the frequency shift signal IV_T after isotropy are obtained as shown in formula (2):

[0097]

[0098] Step 2.3: Wavelet denoising, using wavelet transform to denoise the assimilated compensation capacitor sensing signal CC_T and frequency-shifted signal IV_T. Specifically, the time series data is input into the wavelet transform denoising algorithm. This algorithm uses the BayesShrink method and soft thresholding at 3 levels of wavelet decomposition. The symmetric wavelet (sym7) is selected for decomposition and reconstruction, and the signal variance is recalibrated to ensure the stability and accuracy of the denoising effect. The denoised compensation capacitor sensing signal is obtained. With frequency shift signal .

[0099] Step 3: Phase space reconstruction, using the mutual information method and GP algorithm to calculate the pre-processed compensation capacitor sensing signal one by one With frequency shift signal The delay time and embedding dimension.

[0100] Step 3.1: Mutual information method to calculate delay time .

[0101] Define the signal to be analyzed as , where t represents the time index and N is the sequence length. In order to measure the delay of the sequence under different time delays, the mutual information metric MI(T) is introduced to represent the signal A sequence delayed by T time units from itself The mutual information between them is calculated as follows:

[0102]

[0103] in, and Respectively represent the sequence and sequence The value space. is delaying When the observed and The joint probability of . and yes and The marginal probability distribution of .

[0104] Mutual Information It reflects the reduction in uncertainty about one part of the sequence when the information about another part is known. When the peak value is reached, it means that The sharing between the historical information and future information of the sequence is maximized under the delay of The peak value can determine the optimal delay time :

[0105]

[0106] The peak value indicates that Moment, time series The past value of The values ​​after the time have the highest interdependence.

[0107] Step 3.2: Use the GP algorithm to find the embedding dimension m.

[0108] Setting distance threshold , and calculate the distances of all point pairs to obtain the distribution function of point pairs in phase space :

[0109]

[0110] in, is the total number of points in phase space, is the unit step function, Yes and The Euclidean distance between .

[0111] Embedding Dimension The relevant dimension It is estimated that is defined as:

[0112]

[0113] In practice, different Value calculation , and plotted on a log-log coordinate right Figure. and The slope of the linear fit is the correlation dimension , by choosing an appropriate slope, the embedding dimension can be estimated ,make sure To meet the requirements of phase space reconstruction.

[0114] Step 3.3: Determine the final delay time and embedding dimension. The compensated capacitance sensing signal is calculated based on the mutual information method. And the frequency shift signal corresponding to the carrier frequency is selected according to the carrier frequency: The delay time is , the embedding dimension calculated by the GP algorithm is In order to ensure that the phase space of each variable can be fully expanded and the correlation between data is guaranteed without compressing the phase trajectory, the maximum embedding dimension is selected according to formula (7): and minimum delay time as phase space reconstruction parameters.

[0115]

[0116] Step 3.4: According to the final determined delay time and embedding dimension, use the delay embedding method to respectively compensate the capacitance sensing signal and the frequency-shifted signal corresponding to the carrier frequency Reconstruct the phase space and obtain the trajectory matrix of the two variables and As shown in equation (8) and equation (9).

[0117]

[0118]

[0119] Mode and middle, is the embedding dimension, is the delay time, is the number of state vectors obtained by phase space reconstruction, .

[0120] Step 4: Fuse the trajectory matrices of each variable according to the adaptive weighted fusion algorithm. The algorithm will include several main steps: data segmentation, information entropy evaluation, fuzzy logic method adaptive weight allocation and weight fusion.

[0121] Step 4.1: Data partitioning. First, define the partitioning strategy. CC and Y IV Each column of is divided into blocks of size l. The part less than size l is also treated as an independent block.

[0122] Step 4.2: Information entropy evaluation. Assign weights to each block of each data source. By calculating the information entropy of each block, evaluate the entropy difference between different data sources to determine the "information value" of each block in different sources. The higher the information entropy value, the higher the uncertainty and complexity of the block. Define the information entropy function as :

[0123]

[0124] Where p(x) is the probability of element x appearing in the data block.

[0125] Step 4.3: Fuzzy logic adaptive weight assignment. Use fuzzy logic to adaptively assign weights to each block. First, a fuzzy logic system and associated fuzzy rule set are established to map information entropy to the weights of data blocks. A fuzzy membership function is used to assign a subordinate measure to each observed entropy level. The membership function is defined for each information complexity interval (low, medium, and high) by the formula The following three functions are used to reflect this:

[0126]

[0127] Secondly, the weight is calculated based on the information entropy of each data block and its membership under each membership function. In the present invention, a data block with low entropy is assigned a higher weight, while a data block with high entropy is assigned a lower weight, which means that the present invention prefers data blocks with high certainty. By multiplying its membership with a predetermined weight factor and summing them, the weight W of each data block can be obtained as shown in formula (12):

[0128]

[0129] Among them, α, β, and γ are the corresponding weight factors, α=0.5, β=0.3, and γ=0.2.

[0130] Step 4.4: Weighted fusion. Use step 4.3 to optimize the weight of each block and perform weighted fusion to obtain the final fused trajectory matrix Y. First, the weights of each block corresponding to each variable are normalized according to formula (13):

[0131]

[0132] Secondly, the weighted fusion method is used to fuse each block of each variable. The weighted fusion formula is as follows: As shown:

[0133]

[0134] in, and denote the j-th block of the i-th column of variables CC and IV, respectively, and They represent the weights of the j-th block in the ith column of variables CC and IV respectively.

[0135] Finally, the fused blocks are concatenated to obtain the final fused trajectory matrix Y.

[0136] Step 5: Feature extraction, the trajectory matrix Y after adaptive weighted fusion is used to extract the pseudo-phase diagram that can reflect the state characteristics of the system by using singular value decomposition, thereby extracting the state characteristics of the compensation capacitor. The pseudo-phase diagram is a projection of the state change curve of the original system in the embedded phase space, where there is a deterministic mapping relationship between the system state and the pseudo-phase diagram. Since the reconstructed phase space is usually multi-dimensional, it becomes a challenge to intuitively represent the multi-dimensional pseudo-phase diagram, which needs to be properly processed in order to extract the essential characteristics of the system.

[0137] Let the singular value decomposition of the fused trajectory matrix Y be as follows: As shown:

[0138]

[0139] in, is a diagonal matrix, as shown in the formula, , called matrix The singular values ​​of , the corresponding normalized and orthogonal eigenvectors are The study found that the trajectory approaches The motion of a dimensional hyperellipsoid, where The axes represent the directions of different principal axes, and the essential characteristics of the system are mainly reflected in the directions of the eigenvector axes corresponding to the larger eigenvalues. and Each column vector is orthogonal to each other, so we have . The column vectors of the matrix are the fused matrix In the base The projection on express No. Column vector. Based on the relationship between the pseudo phase diagram projection and the compensation capacitor position, the state characteristics of the compensation capacitor are extracted.

[0140] Step 6: Fault diagnosis, pseudo phase diagram in The projection on the axis shows high sensitivity to different fault levels and locations of the compensation capacitor. When a compensation capacitor fails, the projection amplitude at that location will decrease accordingly, and the severity of the fault is proportional to the proportion of the amplitude reduction. Based on this, a threshold can be set To determine whether the compensation capacitor is faulty. If , the compensation capacitor is faulty; otherwise, the compensation capacitor is normal.

[0141] Figure 1 The main components of the signal dynamic detection system are shown, among which the compensation capacitor detection subsystem mainly uses the electromagnetic induction principle to identify the state of the compensation capacitor along the line. The system consists of a transmitting antenna and a receiving antenna. The transmitting antenna is installed above the bottom of the vehicle perpendicular to the rail surface, while the receiving antenna is located above the middle of the two rails at the bottom of the vehicle. When the detection vehicle travels to the position of the track circuit compensation capacitor, the magnetic field strength in the electromagnetic loop formed by the vehicle wheelset, rails and compensation capacitor will change. According to Maxwell's electromagnetic field theory, in the closed loop formed by the rails and wheelset, the alternating electric field generated by the transmitting antenna will induce an alternating magnetic field. According to Faraday's law of electromagnetic induction, this alternating magnetic field will generate a corresponding induced electromotive force. The potential difference in the closed loop causes the current to flow, and the generated current will generate a corresponding magnetic field. The receiving antenna installed above the center line of the track can receive this magnetic field and convert it into an electrical signal to feed back to the system. Combined with computer detection technology, this system can detect the state of the compensation capacitor.

[0142] Figure 2 The flowchart of the fault diagnosis process of the compensation capacitor of the non-insulated track circuit based on multi-source data fusion is shown in the figure. The first step is to extract the variable induction signal and frequency shift signal that can reflect the working state of the track circuit compensation capacitor from the historical data of the monitoring system of the track high-speed detection vehicle; the second step is to normalize and homogenize the selected variable signals, and use the mutual information method and GP algorithm to respectively obtain the delay time and embedding dimension corresponding to each variable; the third step is to use formula (7) to obtain the final delay time and embedding dimension, and reconstruct the phase space of each variable signal to obtain the corresponding trajectory matrix; the fourth step is to fuse the two trajectory matrices with multi-source data to obtain the fused trajectory matrix; the fifth step is to obtain the pseudo-phase diagram projection from the fused trajectory matrix; the sixth step is to extract the compensation capacitor characteristics from the pseudo-phase diagram projection and make a fault judgment.

[0143] Figure 3The results of preprocessing the induction signal and frequency shift signal of a section X without fault compensation capacitor extracted from the historical data of the rail high-speed integrated inspection vehicle monitoring system are shown. In this example section, there are 16 fault-free compensation capacitors with a carrier frequency of 1700Hz.

[0144] Figure 4 The preprocessed induction signal, frequency-shifted signal, and phase space trajectory after multi-source information fusion corresponding to the fault-free compensation capacitor section X are shown. It can be seen from the phase space trajectory after multi-source information fusion that the trajectory fully integrates the trajectory characteristics of the induction signal and the frequency-shifted signal.

[0145] Figure 5 The pseudo phase diagram is extracted from the trajectory matrix after multi-source information fusion from the fault-free compensation capacitor section X. The projection on the axis. Figure 5 It can be seen that the pseudo-phase diagram is Axis and The projection on the axis shows a high sensitivity to different fault levels and locations of the compensation capacitors.

[0146] Figure 6 The results of preprocessing the induction signal and frequency shift signal of a section X extracted from the historical data of the rail high-speed integrated inspection vehicle monitoring system when the faulty compensation capacitor is present are shown. In this example section, there are 16 compensation capacitors with a carrier frequency of 1700Hz. The compensation capacitor C13 in this section is faulty.

[0147] Figure 7 The preprocessed induction signal, frequency-shifted signal, and phase space trajectory after multi-source information fusion corresponding to the fault compensation capacitor section X are shown. It can be seen from the phase space trajectory after multi-source information fusion that the trajectory fully integrates the trajectory characteristics of the induction signal and the frequency-shifted signal.

[0148] Figure 8 The pseudo phase diagram is extracted from the trajectory matrix after multi-source information fusion from the fault-free compensation capacitor section X. The projection on the axis. Figure 8 It can be seen that the pseudo-phase diagram is The projection on the axis shows a high sensitivity to the different fault levels and locations of the compensation capacitor. When the compensation capacitor C13 fails, the amplitude of the pseudo-phase diagram projection at the corresponding position is will become smaller, and the more serious the fault, the smaller the amplitude. Figure 5 By comparison, we can is the threshold for compensation capacitor fault judgment. , the compensation capacitor is faulty; otherwise, the compensation capacitor is normal.

[0149] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0150] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for diagnosing faults of compensation capacitors in non-insulated track circuits, characterized in that: The following steps are involved: Step 1: Acquire data. The selected data that can reflect the working state variables of the track circuit compensation capacitor include: compensation capacitor induction signal, carrier frequency, and frequency shift signals with carrier frequencies of 1700 Hz, 2000 Hz, 2300 Hz, and 2600 Hz; Among them, the compensation capacitor sensing signal is expressed as , according to the carrier frequency, the frequency shift signal corresponding to the carrier frequency is ,in is the signal length; Step 2: Data Preprocessing Step 2.1: Perform maximum-minimum normalization processing on the compensation capacitance sensing signal CC and the frequency shift signal IV respectively to obtain the normalized compensation capacitance sensing signal CC_norm and the frequency shift signal IV_norm, as shown in formula (1): Step 2.2: Calculate the normalized compensation capacitance sensing signal CC_norm i and the frequency shift signal IV_norm i The isotropic processing is performed to obtain the compensated capacitance sensing signal CC_T and the frequency shift signal IV_T after isotropic processing, which are specifically shown in formula (2): Step 2.3: wavelet denoising, using wavelet transform to perform denoising on the assimilated compensation capacitor sensing signal CC_T and the frequency shift signal IV_T; Specifically, the time series data is input into the wavelet transform denoising algorithm, and the BayesShrink method and soft thresholding are used at the three-layer wavelet decomposition level. The symmetric wavelet (sym7) is selected for decomposition and reconstruction, and the signal variance is recalibrated to ensure the stability and accuracy of the denoising effect, and the compensated capacitive sensing signal after denoising is obtained. With frequency shift signal ; Step 3: Phase space reconstruction, using the mutual information method and GP algorithm to calculate the pre-processed compensation capacitor sensing signal one by one With frequency shift signal The delay time and embedding dimension of Step 3.1: Mutual information method to calculate delay time Define the signal to be analyzed as , where t represents the time index, N is the sequence length, and in order to measure the delay of the sequence under different time delays, the mutual information metric MI(T) is introduced to represent the signal A sequence delayed by T time units from itself The mutual information between them is calculated as follows: in, and Respectively represent the sequence and sequence The value space, is delaying When the observed and The joint probability of and yes and The marginal probability distribution of Mutual Information It reflects the reduction in uncertainty about another part of the sequence when the information of one part is known. When the peak value is reached, it means that Under the delay of , the sharing between the historical information and future information of the sequence is the largest, so by finding The peak value of : The peak value indicates that Moment, time series The past value of The values ​​after the moment have the highest interdependence; Step 3.2: GP algorithm to find the embedding dimension m Setting distance threshold , and calculate the distances of all point pairs to obtain the distribution function of point pairs in phase space : in, is the total number of points in phase space, is the unit step function, Yes and The Euclidean distance between Embedding Dimension By correlation dimension It is estimated that is defined as: In practice, different Value calculation , and plotted on a log-log coordinate right The picture, in and The slope of the linear fit is the correlation dimension , by choosing an appropriate slope, we can estimate the embedding dimension ,make sure To meet the requirements of phase space reconstruction; Step 3.3: Determine the final delay time and embedding dimension The compensated capacitance sensing signal is calculated based on the mutual information method And the frequency shift signal corresponding to the carrier frequency is selected according to the carrier frequency: The delay time is , the embedding dimension calculated by the GP algorithm is In order to ensure that the phase space of each variable can be fully expanded and the correlation between data is guaranteed without compressing the phase trajectory, the maximum embedding dimension is selected according to formula (7): and minimum delay time as phase space reconstruction parameters; Step 3.4: According to the final determined delay time and embedding dimension, use the delay embedding method to respectively compensate the capacitance sensing signal and the frequency-shifted signal corresponding to the carrier frequency Reconstruct the phase space and obtain the trajectory matrix of the two variables and As shown in formula (8) and formula (9): Mode and middle, is the embedding dimension, is the delay time, is the number of state vectors obtained by phase space reconstruction, ; Step 4: Fuse the trajectory matrices of each variable according to the adaptive weighted fusion algorithm; Step 5: Feature extraction. After adaptive weighted fusion, the trajectory matrix Y is used to extract the pseudo-phase diagram that can reflect the state characteristics of the system by using singular value decomposition, thereby extracting the state characteristics of the compensation capacitor. The pseudo-phase diagram is a projection of the state change curve of the original system in the embedded phase space. There is a deterministic mapping relationship between the system state and the pseudo-phase diagram. Since the reconstructed phase space is usually multi-dimensional, it becomes a challenge to intuitively represent the multi-dimensional pseudo-phase diagram, which needs to be processed in order to extract the essential characteristics of the system. Let the singular value decomposition of the fused trajectory matrix Y be as follows: As shown: in, is a diagonal matrix, as shown in the formula, , called matrix Singular values ​​of , and satisfy , the corresponding normalized and orthogonal eigenvectors are ,matrix and Each column vector is orthogonal to each other, so we have , The column vectors of the matrix are the fused matrix In the base The projection on express No. Column vector, based on the relationship between the pseudo phase diagram projection and the compensation capacitor position, extracts the state characteristics of the compensation capacitor; Step 6: Fault diagnosis, pseudo phase diagram in The projection on the axis shows high sensitivity to different fault levels and locations of the compensation capacitor; when a fault occurs in the compensation capacitor, the projection amplitude at that location will decrease accordingly, and the severity of the fault is proportional to the proportion of the amplitude reduction. A threshold is set. To determine whether the compensation capacitor is faulty, If , the compensation capacitor is faulty; otherwise, the compensation capacitor is normal.

2. A method for diagnosing faults of compensation capacitors in a non-insulated track circuit according to claim 1, characterized in that: The adaptive weighted fusion algorithm includes several main steps: data segmentation, information entropy evaluation, fuzzy logic method adaptive weight allocation and weight fusion: Step 4.1: Data chunking First, define the block strategy; transform the trajectory matrix Y CC and Y IV Each column of is divided into blocks of size l; the part less than size l is also treated as an independent block; Step 4.2: Information entropy evaluation Assign weights to each block of each data source, calculate the information entropy of each block, evaluate the entropy difference between different data sources, and determine the "information value" of each block in different sources. The higher the information entropy value, the higher the uncertainty and complexity of the block. The information entropy function is defined as : Among them, p(x) is the probability of element x appearing in the data block; Step 4.3: Fuzzy logic adaptive weight assignment A fuzzy logic method is used to adaptively assign weights to each block. First, a fuzzy logic system and an associated set of fuzzy rules are established to map information entropy to the weights of data blocks. A fuzzy membership function is used to assign a subordinate measure to each observed entropy level. The membership function is defined according to each information complexity interval (low, medium, and high) and is expressed by the formula The following three functions are used to reflect this: Secondly, the weight is calculated based on the information entropy of each data block and its membership under each membership function; in the present invention, a data block with low entropy is assigned a higher weight, while a data block with high entropy is assigned a lower weight, which means that the present invention prefers data blocks with high certainty; by multiplying its membership with a predetermined weight factor and summing them, the weight W of each data block can be obtained as shown in formula (12): Among them, α, β, and γ are corresponding weight factors, α=0.5, β=0.3, and γ=0.2; Step 4.4: Weighted Fusion Use step 4.3 to optimize the weight of each block, and perform weighted fusion to obtain the final fused trajectory matrix Y; first, the weights of each block corresponding to each variable are normalized according to formula (13): Secondly, the weighted fusion method is used to fuse each block of each variable. The weighted fusion formula is as follows: As shown: in, and denote the j-th block of the i-th column of variables CC and IV, respectively, and denote the weights of the jth block in the ith column of variables CC and IV, respectively; Finally, the fused blocks are concatenated to obtain the final fused trajectory matrix Y.

Images