Railway dynamic environment multi-source data fusion intelligent operation and maintenance management method and system

Abstract

The application is specifically a railway dynamic environment multi-source data fusion intelligent operation and maintenance management method and system, relating to the technical field of railway machine room power environment monitoring, comprising: a multi-source data acquisition preprocessing module; a heterogeneous data space-time alignment module; a multi-dimensional feature conflict resolution module; an operation state evaluation and early warning module; and an intelligent linkage execution management and control module.In the application, the adaptive noise reduction method based on singular value decomposition can filter out various random noises and interference, and at the same time, data completion is carried out in a high-order interpolation mode, thereby guaranteeing the integrity and continuity of the time sequence; a three-dimensional space distance matrix considering the influence of environmental factors is constructed, reflecting the physical space relationship between different sensors, and combining the cubic spline upsampling and cross-correlation phase compensation technology, the alignment of multi-source heterogeneous data in the time and space dimensions is realized.

Description

Technical Field

[0001] This invention relates to the field of railway power environment monitoring technology, and in particular to a method and system for intelligent operation and maintenance management and control of railway power environment multi-source data fusion. Background Technology

[0002] With the rapid expansion of my country's high-speed railway network and the continuous improvement of railway intelligence, the number and scale of railway communication and signal equipment rooms have continued to grow, and the types of equipment in the equipment rooms have become increasingly diverse, which has put forward higher requirements for power environment monitoring and operation and maintenance management.

[0003] Currently, environmental monitoring systems have been widely used in various railway computer rooms, enabling real-time data collection and basic monitoring of the operating status of various equipment such as temperature and humidity, smoke detectors, water immersion, UPS power supplies, precision air conditioners, power distribution cabinets, and battery banks, providing important data support for the daily operation and maintenance management of computer rooms.

[0004] The data collected by the environmental monitoring system has significant multi-source heterogeneous characteristics. The data sources cover multiple dimensions such as environmental monitoring, electrical parameters, and equipment status. The sampling frequency of different types of data varies greatly, ranging from seconds to minutes, and the sensors are spatially distributed in the computer room.

[0005] In order to comprehensively utilize multi-source data information and improve the overall monitoring capability of the system, multi-source data fusion technology has become an important research direction in the field of environmental monitoring.

[0006] The industry has developed a variety of data fusion methods, which have improved the efficiency of data utilization and the reliability of monitoring results to a certain extent.

[0007] However, due to the physical spatial differences and signal propagation delays between different sensors, how to achieve accurate alignment of multi-source heterogeneous data in time and space, and how to reasonably handle the information differences between different evidence bodies, are still problems that need to be continuously optimized in the practical application of current multi-source data fusion technology.

[0008] In terms of system status assessment and operation and maintenance management, existing environmental monitoring systems generally adopt threshold-based alarm mechanisms. This mechanism is technically mature, simple to implement, and can effectively identify obvious equipment failures and parameter anomalies. However, with the increasing demands for the reliability of computer rooms in railway transportation, the traditional single threshold alarm method is no longer sufficient to comprehensively reflect the overall operating status of the system, nor can it detect potential faults in advance.

[0009] Meanwhile, in order to shorten fault response time, reduce delays caused by manual intervention, and improve the level of automated operation and maintenance of the system, it is necessary to further study more comprehensive and accurate system status assessment methods and more intelligent and efficient linkage control strategies. Summary of the Invention

[0010] The purpose of this invention is to provide a method and system for intelligent operation and maintenance management of railway dynamic environment multi-source data fusion in order to solve the above-mentioned problems.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A railway dynamic environment multi-source data fusion intelligent operation and maintenance management system includes:

[0013] The multi-source data acquisition and preprocessing module is used to connect to the environmental monitoring equipment via a physical interface to parse the underlying data stream, filter out noise, and output a continuous time series.

[0014] The heterogeneous data spatiotemporal alignment module is used to construct a three-dimensional spatial distance matrix for the computer room, upsample low-frequency data, compensate for physical phase delay by combining discrete cross-correlation function, and output a standard multidimensional data matrix.

[0015] The multidimensional feature conflict resolution module is used to generate initial basic probability assignments based on the physical state space identification framework and spatial distance matrix, calculate the Juselm distance and conflict coefficient to generate dynamic discount coefficients, and use the Dempster combination rule to fuse multi-source evidence to output state feature vectors.

[0016] The operational status assessment and early warning module is used to reconstruct the state feature vector into a high-dimensional phase space trajectory, calculate the discrete Friesian distance and the maximum Lyapunov exponent of the real-time trajectory deviation from the baseline trajectory, fuse them to generate a comprehensive degradation index and output early warning instructions;

[0017] The intelligent linkage execution control module is used to schedule early warning commands in a timely manner, calculate control increments using dynamic matrix control algorithms, perform electrical switching by combining phase-locked loop and zero-crossing prediction, and perform state feedback compensation through Luneburg observer.

[0018] Preferably, in the multi-source data acquisition and preprocessing module, noise filtering includes:

[0019] A Hankel matrix is ​​constructed by combining a one-dimensional discrete time series with a window length, where the window length is greater than the number of sampling points corresponding to the highest frequency period of the signal. Singular value decomposition of the Hankel matrix yields a left singular vector orthogonal matrix, a right singular vector orthogonal matrix, and a singular value diagonal matrix.

[0020] Calculate the difference components of adjacent singular values, and find the index corresponding to the maximum peak in the difference sequence as the truncation order to adaptively determine the signal-noise boundary;

[0021] The singular values ​​before the truncation order are retained and the remaining singular values ​​are set to zero to obtain a new diagonal matrix. The Hankel matrix is ​​then reconstructed, and the anti-diagonal averaging method is used to restore the reconstructed Hankel matrix to the denoised time series.

[0022] Preferably, the process of outputting the continuous time series includes:

[0023] A time sliding window is set and continuously advanced over time. Local maximum and local minimum values ​​are found within the time sliding window. Combined with a minimum positive real constant, the time series after noise reduction is dynamically normalized to eliminate the difference in the dimensions of physical quantities of different sensors.

[0024] When missing data is detected, extract the valid data points before and after the missing data point, and calculate the zero-order difference quotient to the higher-order difference quotient in sequence.

[0025] A Newton interpolation polynomial is constructed using the quotients of each order. The missing time points are substituted into the Newton interpolation polynomial to calculate the normalized data after interpolation, ensuring the continuity of the higher-order derivatives of the interpolation results.

[0026] Preferably, the heterogeneous data spatiotemporal alignment module constructs a three-dimensional spatial distance matrix for the computer room, including:

[0027] A three-dimensional Cartesian coordinate system for the computer room is established with a fixed corner of the computer room as the origin, and the spatial coordinate vector of the sensor is obtained.

[0028] use Calculate the first The sensor and the first Correcting the spatial distance between the sensors, among which For spatial coordinates, Let be the obstacle damping constant. The actual propagation path of environmental parameters is corrected by the obstacle damping constant to determine the number of physical obstacles through which the connection between sensors passes.

[0029] A symmetric spatial distance matrix is ​​constructed based on the corrected spatial distances between all sensors.

[0030] Preferably, the step of upsampling low-frequency data, compensating for physical phase delay using a discrete cross-correlation function, and outputting a standard multidimensional data matrix includes:

[0031] A cubic basis spline basis function is defined using a recursive formula. A tridiagonal linear equation system is constructed by combining boundary conditions and the control points are solved by the chasing method. A continuous-time reconstruction function is constructed by linearly combining the control points and the cubic basis spline basis function. The upsampled environmental data sequence is obtained by substituting the high-frequency electrical data timestamp sequence into the continuous-time reconstruction function.

[0032] Calculate the discrete cross-correlation function between the high-frequency electrical data sequence and the upsampled environmental data sequence, and iterate through the time delay steps to find the delay step corresponding to the maximum value of the discrete cross-correlation function as the actual physical time step;

[0033] The upsampled environmental data sequence is shifted forward by the corresponding delay steps to compensate for the time axis shift. The missing data points at the end of the shift are then filled in using Newton's difference quotient polynomial extrapolation, and a standard multidimensional data matrix is ​​output.

[0034] Preferably, the initial basic probability assignment based on the physical state space identification framework and spatial distance matrix includes:

[0035] Define a set of frameworks for identifying normal, abnormal, and fault states, and set the expected central value and standard deviation for each individual state based on historical benchmark data.

[0036] use Calculate the measured value Belonging to each single state Gaussian membership ,in The expected central value, Standard deviation;

[0037] use Calculate the spatial decay weight ,in The spatial attenuation constant, To correct the spatial distance, the Gaussian membership degree is corrected using spatial decay weights. The initial basic probability of the evidence body for a single proposition is assigned, and the initial basic probability of the evidence body for the entire set of uncertainties is also assigned.

[0038] Preferably, the step of calculating the Juselm distance and conflict coefficient to generate a dynamic discount coefficient, and fusing multi-source evidence using the Dempster combination rule to output a state feature vector, includes:

[0039] Calculate the conflict coefficient between any two pieces of evidence and the Juselm distance used to quantify the geometric distance in the vector space, using... Calculate the similarity, where The conflict coefficient, For Juselm distance;

[0040] The support of the evidence body is calculated and normalized to obtain the relative credibility weight. The initial basic probability is dynamically discounted using the relative credibility weight to obtain the corrected evidence body. The weights of severely conflicting and abnormal data are transferred to the entire set of uncertainties.

[0041] The modified evidence body is iteratively fused pairwise using the Dempster combination rule and the normalized conflict constant. Under the condition that the maximum probability difference is greater than the decision threshold and the uncertainty is less than the uncertainty threshold, the state feature vector is output based on the maximum confidence principle.

[0042] Preferably, the operational status assessment and early warning module specifically includes:

[0043] The mutual information function of the state feature vector is calculated and the time corresponding to the first local minimum of the mutual information function is found as the delay time. The proportion of false nearest neighbors is counted and the dimension corresponding to the first time the proportion of false nearest neighbors drops to near zero is found as the embedding dimension. The state feature vector is expanded into a high-dimensional phase space trajectory through the delayed coordinate method.

[0044] Construct a time-monotonically increasing coupling sequence between the baseline health trajectory and the real-time running trajectory, and calculate the maximum Euclidean distance between corresponding point pairs under the coupling sequence;

[0045] A distance matrix is ​​constructed and the recursive calculation rules for matrix elements are set. A dynamic programming algorithm is used to solve for the minimum maximum Euclidean distance among all possible coupling sequences, which is used as the discrete Friesian distance for quantifying geometric shape deviation.

[0046] In the reconstructed high-dimensional phase space, the nearest neighbor of the phase point is found, and the time interval is restricted to be greater than the average orbital period to exclude adjacent points on the same orbit, so as to calculate the average logarithmic distance between adjacent trajectories with the number of evolution steps;

[0047] The least squares method was used to perform linear regression fitting on the initial linearly rising region of the logarithmic distance-mean curve, and the maximum Lyapunov exponent was calculated by dividing the slope of the fitted line by the sampling time interval.

[0048] The maximum Lyapunov exponent is mapped, and the discrete Friesian distance and the mapped maximum Lyapunov exponent are normalized and weighted to generate a comprehensive degradation index. The index is then compared with the preset progressive warning threshold to output the corresponding graded warning command.

[0049] Preferably, the process of calculating the control increment using a dynamic matrix control algorithm and performing electrical switching in combination with a phase-locked loop and zero-crossing prediction is as follows:

[0050] Extract the urgency, transient power consumption, and physical loss assessment values ​​of the early warning command task to construct a multi-objective cost function for time-series scheduling;

[0051] A dynamic control matrix is ​​constructed using the response sequence under a unit step control signal. A quadratic objective function is constructed, which includes an error penalty weight matrix and a control increment penalty weight matrix. The control increment is calculated by taking the partial derivative of the quadratic objective function.

[0052] The backup power supply voltage signal is orthogonally virtualized, and the Parker transform is applied to calculate the orthogonal voltage component input to the proportional-integral regulator. The internal phase angle is updated to lock the backup power supply phase. The next natural zero-crossing time of the main voltage is calculated to send a trigger pulse to perform electrical switching in advance of the inherent delay time of the relay mechanical action.

[0053] A Luenberger observer is constructed to estimate the physical state vector in real time. The deviation between the estimated state and the desired target state is calculated to generate a compensating control quantity, which is then superimposed on the control increment.

[0054] A method for intelligent operation and maintenance management of railway dynamic environment data fusion includes:

[0055] The system connects to the environmental equipment to analyze the underlying data stream, uses an adaptive singular value decomposition algorithm to filter out noise, and outputs a continuous time series through dynamic sliding window normalization and Newton's difference quotient polynomial interpolation.

[0056] Construct a three-dimensional spatial distance matrix for the computer room, use cubic basis spline interpolation to upsample low-frequency data, combine discrete cross-correlation function to compensate for physical phase delay, and output a standard multidimensional data matrix;

[0057] Based on the identification framework and spatial distance matrix, the initial basic probability is assigned, the Juselm distance and conflict coefficient are calculated to generate dynamic discount coefficients, and the evidence is fused using the Dempster combination rule to output the state feature vector.

[0058] Using the Tukens embedding theorem, the state feature vector is reconstructed into a high-dimensional phase space trajectory. The discrete Friesian distance and the maximum Lyapunov exponent from the baseline trajectory are calculated, and a comprehensive degradation index is generated by fusing them and an early warning command is output.

[0059] The early warning commands are scheduled based on a multi-objective cost function, the control increment is calculated using a dynamic matrix control algorithm, electrical switching is performed by combining a phase-locked loop and zero-crossing prediction, and state feedback compensation is performed through a Luneburg observer.

[0060] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0061] 1. This invention employs an adaptive noise reduction method based on singular value decomposition, which effectively filters out various random noises and interferences. Simultaneously, it uses high-order interpolation for data completion, ensuring the integrity and continuity of the time series. Furthermore, it constructs a three-dimensional spatial distance matrix that considers environmental factors, reflecting the physical spatial relationships between different sensors. Combined with cubic spline upsampling and cross-correlation phase compensation techniques, it achieves alignment of multi-source heterogeneous data in both time and space dimensions.

[0062] 2. This invention transforms one-dimensional state characteristics into high-dimensional phase space trajectories through phase space reconstruction technology, comprehensively quantifying the system's operating state from two dimensions: geometric morphology and system chaotic characteristics. This can comprehensively reflect the overall health status of the computer room's power environment and help to detect potential faults in advance. In terms of linkage control, model predictive control algorithms are used to calculate control increments, and phase locking and zero-crossing prediction technologies are combined to perform electrical switching, which helps to reduce voltage fluctuations and current surges during the switching process. At the same time, real-time feedback compensation is performed through a state observer, which improves control accuracy and system response speed, enabling rapid fault isolation and system recovery when a fault occurs. Attached Figure Description

[0063] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0064] Figure 1 This is a system structure diagram of the present invention;

[0065] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0066] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0067] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein. Example 1

[0068] Its specific implementation method is combined with the appendix Figure 1 and attached Figure 2 Please provide a detailed explanation.

[0069] In this embodiment, it includes:

[0070] The multi-source data acquisition and preprocessing module is used to connect to the environmental equipment through a physical interface to parse the underlying data stream, use an adaptive singular value decomposition algorithm to filter out noise, and output a continuous time series through dynamic sliding window normalization and Newton's difference quotient polynomial interpolation.

[0071] The core function of this module is to perform real-time acquisition of underlying physical signals, protocol parsing, adaptive noise reduction, dynamic normalization, and missing value interpolation for complex power and environmental equipment in railway equipment rooms and communication base stations.

[0072] The railway dynamic environment is characterized by strong electromagnetic interference, high-frequency train vibration, and extreme temperature and humidity changes. Therefore, this module constructs a highly robust data cleaning and preprocessing mechanism at the pure mathematical and signal processing levels.

[0073] The railway dynamic environment system connects to a wide variety of devices, including uninterruptible power supplies, battery banks, AC / DC power distribution cabinets, precision air conditioners, temperature and humidity sensors, water immersion sensors, and access control equipment. This module connects to these devices through the physical interfaces of the edge computing gateway (including serial communication interfaces, controller area network bus interfaces, and Ethernet interfaces).

[0074] To address the heterogeneous communication protocols used by different devices, a multi-protocol parsing engine has been built inside the module.

[0075] Taking a serial communication protocol as an example, the module reads the data stream according to a fixed frame format and extracts the device address code, function code, data area, and checksum. When reading the data area, the module reassembles the continuous byte stream according to big-endian or little-endian mode based on a pre-configured register mapping table, converting it into actual physical quantity values.

[0076] To ensure the accuracy of data transmission, the module performs a cyclic redundancy check on each frame of data. Let the received message polynomial be... The generator polynomial is The module computes within a finite domain. Divide by remainder ,like If the value is zero, the data frame is considered valid; otherwise, the data frame is marked as invalid and discarded, thus eliminating garbled data caused by communication line interference at the data source.

[0077] In a railway environment, the strong electromagnetic pulses and mechanical vibrations generated when a train passes by will cause a large amount of high-frequency transient noise to be superimposed on the collected voltage, current and vibration signals.

[0078] In order to filter out noise without losing the signal's abrupt change characteristics, this module adopts an adaptive singular value decomposition denoising algorithm based on phase space reconstruction.

[0079] Suppose that the one-dimensional discrete-time sequence of a certain physical quantity obtained after protocol parsing is: ,in This represents the total length of the sequence.

[0080] Choose a suitable window length (usually taken) The value is greater than the number of sampling points corresponding to the highest frequency period of the signal, and satisfies... ), construct a dimension as Hankel matrix ,in .

[0081] Hankel matrix The construction method is as follows:

[0082] The first row of the matrix is arrive The second row of elements is arrive And so on, the first row element is arrive ;

[0083] Hankel matrix Singular value decomposition is performed, and the formula is expressed as: ,in, It is a dimension The left singular vector orthogonal matrix; It is a dimension The right singular vector orthogonal matrix; Representation matrix transpose; It is a dimension A diagonal matrix, whose elements are on the main diagonal. That is, it is a singular value, and satisfies ( for and (the smaller value in the range).

[0084] In noisy signals, larger singular values ​​typically correspond to the energy of the true signal, while smaller singular values ​​correspond to the energy of the noise. To adaptively determine the boundary between signal and noise, this module calculates the difference spectrum of the singular values.

[0085] Definition of the first The difference of the singular values ​​is ,in The range of values ​​is arrive Finding difference sequences The index corresponding to the maximum peak value in The index This is the truncation order of the singular values ​​of the effective signal.

[0086] Before keeping A singular value, The first in the matrix To the All singular values ​​are set to zero, resulting in a new diagonal matrix. ;

[0087] The Hankel matrix can be reconstructed using the truncated singular value matrix, as shown in the formula: The reconstructed matrix The elements in are denoted as ,in row index ( ), For column indexes ( ).

[0088] In order to make the matrix Restored to one-dimensional denoised time series This module uses the anti-diagonal averaging method.

[0089] The specific calculation rules are as follows:

[0090] when hour, ;when hour, ;when hour, .

[0091] Through the above calculations, the module can effectively separate and remove broadband white noise and impulse interference from the dynamic environment data.

[0092] Since the physical quantities collected by different sensors in the railway dynamic environment system have different dimensions and orders of magnitude (for example, voltage may be as high as hundreds of volts, while current may be only a few amperes, and temperature may be in the tens of degrees Celsius), in order to eliminate the influence of the difference in dimensions on the subsequent multidimensional state space mapping, the noise-reduced data must be normalized.

[0093] Considering the significant seasonal variations and diurnal temperature differences in the railway environment, fixed global maximum and minimum values ​​would cause the normalized data to lose its local fluctuation characteristics. Therefore, this module adopts a dynamic sliding window normalization method.

[0094] Set a length of A time-sliding window. For the current moment... Noise reduction data points The module is in the interval Finding local maxima and local minimum .

[0095] Normalized value of the current data point The calculation formula is: ,in, It is a very small positive real constant (for example, taking the value of ). Its function is to prevent mathematical anomalies such as a denominator of zero when the data in the window is completely identical.

[0096] As time goes on, the sliding window continues to move forward. and Dynamic updates ensure that the normalized data is updated correctly. Always distributed in Within the specified range, it can sensitively reflect minor local changes in the operating status of the equipment.

[0097] In actual operation, due to brief power outages of the sensors or momentary congestion of the communication link, data gaps are inevitable in the acquired time series. To ensure the continuity of the time series during subsequent phase space reconstruction, this module performs mathematical interpolation on the missing data.

[0098] Considering the continuity and smoothness of changes in dynamic physical quantities (such as temperature and voltage), this module uses Newton's difference quotient polynomial interpolation instead of simple linear interpolation.

[0099] Suppose that time is detected in the time series. If data is missing, the module extracts the data before and after the missing point. There are 1 valid data points, denoted as _____. .

[0100] First, calculate the quotients of each order of difference:

[0101] Define the zeroth order difference quotient as ;

[0102] The first-order difference quotient is defined as: ;

[0103] The second-order difference quotient is defined as: ;

[0104] And so on, the... The step difference quotient is defined as: ;

[0105] Using the calculated difference quotients of each order, a Newton interpolation polynomial is constructed. : ;

[0106] Missing moments Substituting into the polynomial above, the calculated value is... This refers to the normalized data after interpolation.

[0107] This method can make full use of the changing trends of the data around the missing point, ensure the continuity of the higher-order derivatives of the interpolation results, and provide a high-quality data foundation for subsequent dynamic analysis.

[0108] The heterogeneous data spatiotemporal alignment module is connected to the multi-source data acquisition and preprocessing module. It is used to construct a three-dimensional spatial distance matrix of the computer room, upsample low-frequency data using cubic base spline (B-spline) interpolation, and compensate for physical phase delay by combining discrete cross-correlation function to output a standard multidimensional data matrix.

[0109] After data preprocessing, the core challenge facing the system is the heterogeneity of multi-source data. Electrical data (such as three-phase voltage and current) in railway dynamic environment systems usually require high-frequency sampling (e.g., thousands of times per second) to capture transient power quality issues; while environmental data (such as computer room temperature, humidity, and concentration of harmful gases) change slowly and are usually sampled at low frequency (e.g., once per minute).

[0110] The different physical installation locations of various sensors within the computer room result in spatial differences and time delays in their perception of the same fault event. This module aims to unify these heterogeneous data with different sampling frequencies and phases under a strictly aligned spatiotemporal reference through pure mathematical transformations and spatial mapping, thus removing obstacles to the construction of a multidimensional state space.

[0111] To quantify the spatial correlation between different sensors, this module first establishes a three-dimensional Cartesian coordinate system for the railway machine room.

[0112] Using a fixed corner of the computer room as the origin, the length, width, and height of the computer room are defined as follows: axis, axis, axis.

[0113] Each environmental device and sensor in the computer room is abstracted as a node in this three-dimensional space, let the first node be... The spatial coordinate vector of each sensor is .

[0114] Considering the physical obstacles such as server racks and cable trays in the computer room, the traditional Euclidean straight-line distance cannot accurately reflect the actual propagation path of environmental parameters (such as hot airflow and water immersion).

[0115] This module introduces a corrected spatial distance calculation method that combines Manhattan distance with obstacle penalty coefficient;

[0116] Definition of the first The sensor and the first Correction space distance between sensors for: ;

[0117] in, The preset obstacle damping constant; For nodes With nodes The number of physical obstacles that the connecting lines pass through.

[0118] Based on the above formula, the module calculates the corrected spatial distance between all pairs of sensor nodes, constructing a dimension of Symmetric spatial distance matrix ( (This represents the total number of sensors). This matrix provides a geometric basis for subsequent analysis of the correlation between sensor data from different locations.

[0119] In order to align low-frequency environmental data with high-frequency electrical data on the time axis, the low-frequency data must be upsampled.

[0120] Simple zero-order preservation or linear interpolation can introduce non-differentiable inflection points into the data sequence, disrupting the smoothness of the state-space trajectory.

[0121] This module uses the cubic B-spline interpolation algorithm to reconstruct the continuous-time function of low-frequency signals at a purely mathematical level.

[0122] Suppose a low-frequency sensor in a time sequence The normalized data sequence collected above is .

[0123] To construct a B-spline curve, first define a node vector. .

[0124] B-spline basis functions Defined using the Cox-deBoor recursive formula, where Let be the order of the spline, taken here. (i.e., cubic B-spline).

[0125] The zeroth-order basis function is defined as:

[0126] when hour, ;otherwise ;

[0127] The recursive formula for higher-order basis functions is: ;

[0128] In the above formula, if the denominator is zero, then the term as a whole is defined as zero.

[0129] Using the aforementioned basis functions, the continuous-time reconstruction function for low-frequency data This can be represented as a control point. Linear combination with basis functions: ;

[0130] To solve for control points The module will use known low-frequency data points Substituting the above equations and combining them with boundary conditions (such as natural boundary conditions, i.e., the second derivatives at both ends of the curve are zero), we can construct a tridiagonal linear system of equations.

[0131] By solving this system of linear equations using the chasing method, all control points can be obtained. ;

[0132] Obtain continuous-time function Then, the module uses the timestamp sequence of the high-frequency electrical data. ,Will Substitute in sequence The calculations are performed to obtain the upsampled environmental data sequence. .

[0133] At this point, a one-to-one correspondence was achieved between low-frequency data and high-frequency data in terms of sampling frequency and number of timestamps.

[0134] Although the data is aligned in timestamps, there is an inherent causal delay between different parameters in the physical world.

[0135] For example, a sudden increase in the load current (high-frequency electrical data) of a UPS device in a computer room will cause the device to generate more heat. However, due to the thermodynamic inertia of the air, the temperature at the air conditioner return vent (low-frequency environmental data) will take some time to show an upward trend.

[0136] If current and temperature at the same timestamp are directly combined in the state space, the dynamic causal relationship of the system cannot be truly reflected.

[0137] This module uses a discrete cross-correlation function to calculate and compensate for this physical phase delay. Let the high-frequency electrical data sequence be... The upsampled environmental data sequence is The sequence length is .

[0138] Define the discrete cross-correlation function for these two sequences. for: ;

[0139] in, The time delay is the number of steps, and its value ranges from 1 to 2. , The maximum possible delay steps are preset based on the physical characteristics of the computer room.

[0140] Module traversal of all Calculate the corresponding cross-correlation function value. Searching for Delay steps to reach the maximum value : ;Should This represents the actual physical time steps by which environmental data lags behind electrical data.

[0141] The module performs time-axis shift compensation on the environmental data sequence, specifically as follows:

[0142] Environmental data sequence Move forward as a whole Each step, i.e., the new environmental data sequence ;

[0143] For the missing end after translation The data points are supplemented by extrapolation using the aforementioned Newton's difference quotient polynomial.

[0144] Through the construction of the spatial distance matrix, cross-band upsampling of cubic B-splines, and phase delay compensation of the cross-correlation function, the heterogeneous data spatiotemporal alignment module successfully transformed the originally chaotic, frequency-varying, and phase-misaligned multi-source dynamic environment data into a standard multidimensional data matrix that is strictly synchronized in the time dimension and has a clear topological relationship in the spatial dimension.

[0145] Each row in this matrix represents the state of all dynamic physical quantities at a unified time section, providing a solid and accurate mathematical foundation for the subsequent construction of high-dimensional state-space trajectories.

[0146] The multidimensional feature conflict resolution module is connected to the heterogeneous data spatiotemporal alignment module. It is used to generate initial basic probability assignments based on the physical state space identification framework and spatial distance matrix, calculate the Jousselme distance and conflict coefficient to generate dynamic discount coefficients, and use the Dempster combination rule to fuse multi-source evidence to output state feature vectors.

[0147] In railway dynamic environment systems, although multi-source data that has undergone spatiotemporal alignment has achieved synchronization in terms of timestamps and spatial topology, serious conflicts often occur in the representations of the same physical event by different sensors due to sensor aging, sudden changes in local environment (such as a short circuit in a cabinet causing instantaneous high temperature, while the temperature of the global air conditioning return air vent remains normal), or strong electromagnetic interference.

[0148] Traditional logical AND / OR operations or simple weighted average methods cannot handle such heterogeneous data with high conflict and high uncertainty.

[0149] This module abandons conventional rule matching and machine learning black-box models, introduces and deeply improves DS evidence theory, and achieves conflict resolution and highly reliable fusion of multidimensional features under a strict probabilistic framework by constructing dynamic discount coefficients and multidimensional conflict measurement functions.

[0150] To apply evidence theory, a complete and mutually exclusive identification framework must first be established. For the operating status of railway dynamic environment equipment, this module defines the set of identification frameworks as follows: .

[0151] in, This indicates that the equipment is operating normally. This indicates that the equipment is in an early stage of deterioration or a minor abnormal condition. This indicates a serious equipment malfunction or environmental limitations.

[0152] The power set of the identification frame is denoted as It contains all possible combinations of subsets, for example This indicates a state between normal and slightly abnormal, reflecting the uncertainty in state determination.

[0153] For each physical quantity (such as voltage, current, and temperature of a node) in the spatiotemporally aligned multidimensional data matrix, it is regarded as an independent information source (i.e., evidence).

[0154] Assume there is a total One piece of evidence, denoted as The task of this module is to assign a basic probability value, i.e., a quality function, to each piece of evidence. ,in for Any subset of.

[0155] The mass function must satisfy two fundamental conditions: the mass function of the empty set is zero, i.e. The sum of the quality functions of all subsets is 1, that is... .

[0156] To transform continuous physical quantity values ​​into mass functions, this module employs a Gaussian membership function incorporating the spatial distance matrix. For the... Each piece of evidence (sensor) has a normalized measurement value at its current moment. .

[0157] Based on historical benchmark data, a normal state is pre-set. Abnormal state Fault status The corresponding expected center values ​​are respectively , and their corresponding standard deviations are respectively .

[0158] Calculate the measured value Belonging to each single state ( Initial membership degree : ;

[0159] Considering the physical location of sensors in the three-dimensional space of the computer room, the closer the sensor is to the core equipment, the higher the reliability of its data.

[0160] Introducing the spatial distance matrix constructed in the previous module ;

[0161] Let the first The spatial correction distance between each sensor and the current monitoring target device is: Define spatial decay weights. for: ;

[0162] in, This is a preset spatial attenuation constant used to control the degree of influence of distance on reliability.

[0163] Combined with spatial weights, calculate the first A body of evidence against a single proposition Basic probability assignment : ;

[0164] For the universal set representing uncertainty (i.e., the specific state is unknown), its basic probability is assigned. The calculation is as follows: ;

[0165] Through the above mapping, the module transforms multi-source heterogeneous physical quantities into a unified set of initial evidence bodies under the identification framework.

[0166] In the classic DS evidence theory, when two pieces of evidence are highly conflicting (for example, sensor A strongly supports normality, while sensor B strongly supports malfunction), direct fusion will lead to the "Zadeh paradox," resulting in a conclusion that defies common sense.

[0167] To address this issue, this module performs conflict measurement on all evidence before fusion and generates dynamic discount coefficients.

[0168] Calculate any two pieces of evidence and Conflict coefficient between : ;

[0169] in, and These are subsets of the power set of the identification framework. The larger the value, the greater the probability that the two pieces of evidence point to mutually exclusive propositions, and the more serious the conflict.

[0170] The conflict coefficient alone cannot fully measure the differences between pieces of evidence; this module further introduces the Jousselme distance. To quantify the geometric distance of the evidence body in the vector space.

[0171] Define a dimension as Positive definite matrix Its elements Defined as: ;

[0172] in, Represents a set and The cardinality (number of elements) of the intersection. The cardinality of the union is represented. The formula for calculating the Jousselme distance is: ;

[0173] in, It is a column vector containing the differences in the quality function over all subsets; It is its transpose vector.

[0174] By combining the conflict coefficient and Jousselme distance, the body of evidence is defined. and similarity between : ;

[0175] Build a The similarity matrix;

[0176] For the For each piece of evidence, the sum of its similarity to all other pieces of evidence is called the support of that piece of evidence. : ;

[0177] Normalizing the support, we obtain the first... Relative credibility weight of each piece of evidence : ;

[0178] Using relative credibility weights Assigning values ​​to the original basic probabilities Dynamic discounting is performed to obtain the revised evidence. :

[0179] For any proper subset , ;

[0180] For the complete series , .

[0181] Through this step, anomalous data that severely conflicts with other sensors (such as isolated data generated by faulty sensors) will be assigned extremely low weights, and their support for a specific proposition will be transferred to the uncertain set. This eliminates the interference of malicious or distorted data on the system at the mathematical level.

[0182] After completing the dynamic discount correction for all pieces of evidence, this module uses Dempster's combination rule to apply the correction to the pieces of evidence. We will integrate them one by one.

[0183] To improve computational efficiency, a pairwise fusion iterative approach is adopted:

[0184] Suppose that the current has already merged with the previous One piece of evidence, to obtain comprehensive evidence. Now, let's compare it with the first... A revised piece of evidence By integrating the evidence, new comprehensive evidence can be obtained. .

[0185] For any non-empty subset in the identification framework The formula for calculating the fused mass function is as follows: ;

[0186] in, The normalized conflict constant is calculated using the following formula: ;

[0187] go through After several iterations of calculation, the final global quality function, which integrates information from all sensors, is obtained. .

[0188] The module uses decision logic based on the maximum confidence principle to output the final feature state:

[0189] If the following three conditions are met:

[0190] Condition one: ;

[0191] Condition two: (in This is the second most probable state. (This is a preset decision threshold, for example, 0.15).

[0192] Condition three: (in (This is the maximum allowable uncertainty threshold, for example, 0.1).

[0193] Then determine the current system state after multi-dimensional feature fusion as follows: If the above conditions are not met, the system retains the uncertain state and triggers a reassessment in the next sampling period. Through rigorous calculations in this module, the system successfully extracts multi-source heterogeneous data containing conflicts, noise, and uncertainties into high-confidence state feature vectors, providing a clean and reliable input source for subsequent phase space trajectory mapping.

[0194] The operational status assessment and early warning module is connected to the multi-dimensional feature conflict resolution module. It is used to reconstruct the state feature vector into a high-dimensional phase space trajectory using the Takens embedding theorem, calculate the discrete Fréchet distance and the maximum Lyapunov exponent of the real-time trajectory deviation from the baseline trajectory, fuse them to generate a comprehensive degradation index and output early warning instructions.

[0195] After acquiring high-quality multidimensional feature data that has undergone spatiotemporal alignment and conflict resolution, traditional environmental monitoring systems typically use fixed upper and lower thresholds to trigger alarms. However, the early degradation of railway environmental equipment (such as UPS inverters, precision air conditioning compressors, and battery packs) is often a slow, nonlinear dynamic evolution process, with parameter changes being extremely small in the early stages, completely masked within the normal fluctuation threshold range.

[0196] To address this industry pain point, this module abandons the threshold comparison method and, based on nonlinear dynamics and chaos theory, maps multidimensional time series to a high-dimensional phase space. By calculating the geometric distance and dynamic divergence rate of the phase space trajectory, it achieves accurate quantification and early warning of early hidden degradation of equipment.

[0197] Phase space reconstruction is a core mathematical tool for revealing the hidden dynamic characteristics of nonlinear systems. Let's assume that after processing by the preceding modules, the extracted one-dimensional discrete-time series representing the core operating state of the equipment is... ,in , This represents the total length of the sequence.

[0198] According to Takens' embedding theorem, this one-dimensional sequence can be expanded into a [formula / structure] using the delayed coordinate method. A phase space vector of dimension .

[0199] Reconstructed phase space matrix It can be represented as: ;

[0200] Among them, each phase point vector Defined as: ;

[0201] In the formula, For delay time; For the embedding dimension; This represents the total number of phase points in phase space.

[0202] To ensure that the reconstructed phase space is truly equivalent to the topology of the original system, the delay time must be scientifically determined. and embedding dimension .

[0203] This module uses mutual information to determine the delay time. Define a sequence and delayed sequence Mutual information function between for: ;

[0204] in, and These are the marginal probability densities of the sequence at the corresponding time points; This represents the joint probability density. The module calculates the above probability density using histogram estimation and continuously adds... The value of .

[0205] Finding mutual information functions The first time a local minimum is reached, the corresponding The value is used as the optimal delay time.

[0206] The information redundancy among the components of the phase space vector is minimal, and their independence is strongest.

[0207] Secondly, this module uses the spurious nearest neighbor method to determine the embedding dimension. In dimension 1 In the phase space, find the phase point Euclidean nearest neighbor The distance between the two is denoted as .

[0208] When the embedding dimension increases to At that time, the distance between the two points becomes Define the rate of change of distance. for: ;

[0209] if Greater than the preset threshold (The value is usually between 10 and 15), then it is determined. yes The spurious nearest neighbors are those points that are close in low-dimensional space only because of the folding caused by projection, rather than being truly dynamically adjacent.

[0210] The module calculates the proportion of false nearest neighbors among all phase points. This varies with the dimension. As the proportion of false nearest neighbors increases, this ratio will gradually decrease. When the proportion of false nearest neighbors first drops to near zero (e.g., less than 5%), the corresponding... The value is the optimal embedding dimension. .

[0211] After successfully constructing a high-dimensional phase space, the operating state of the device over a period of time is represented by a continuous trajectory in the phase space.

[0212] This module, under the condition of a brand-new or recently overhauled healthy device, collects a sufficiently long data segment to reconstruct a baseline health trajectory, denoted as... .

[0213] During real-time operation, the real-time running trajectory is reconstructed using the data within the current sliding window, denoted as... .

[0214] To quantify real-time trajectories Deviation from baseline trajectory To determine the geometric degree, this module uses the discrete Fréchet distance.

[0215] Unlike the traditional Euclidean distance, the Fréchet distance takes into account the order and continuity of the trajectory points;

[0216] Define a coupling sequence It is composed of trajectories and index pairs The sequence formed satisfies , And for all The index increment can only be 1 / 2. , or .

[0217] This ensures that time is monotonically increasing when comparing trajectories;

[0218] For a specific coupling sequence Calculate the maximum Euclidean distance between all corresponding pairs of points in this sequence: ;

[0219] Discrete Fréchet distance Defined as the minimum of the above maximum distance among all possible coupling sequences: ;

[0220] In practical calculations, this module uses a dynamic programming algorithm to solve the problem. Construct a dimension as Distance matrix .

[0221] The recursive formula for calculating matrix elements is as follows: ;

[0222] for , ;

[0223] for , ;

[0224] for and : ;

[0225] Final calculation That is, the discrete Fréchet distance. .

[0226] The larger the distance value, the more severely the current operating trajectory of the equipment deviates from its healthy state in terms of geometry, which directly reflects the wear and tear of the equipment's physical structure or the drift of its parameters.

[0227] In addition to geometric deviations, equipment degradation can also lead to the disruption of the system's internal dynamic stability, manifested as increased sensitivity to minute disturbances.

[0228] This module quantifies this chaotic divergence property by calculating the maximum Lyapunov exponent of the phase space trajectory.

[0229] The MLE is calculated using a small data volume method; in the reconstructed phase space, for each phase point... Find its nearest neighbor in space. Furthermore, the time interval between the two is limited to be greater than the average orbital period in order to exclude adjacent points on the same orbit.

[0230] The initial distance is denoted as ;

[0231] Over time (evolutionary steps) These two initially close points will evolve along their respective trajectories, and the distance between them will become... .

[0232] According to chaotic dynamics theory, the divergence of distances follows an exponential law: ,in This is the maximum Lyapunov index. This represents the sampling time interval.

[0233] Taking the natural logarithm of both sides of the above equation, we get: ;

[0234] To obtain the global value The module covers all phase points. In evolutionary steps To find the average of the logarithmic distances, define the function. : ;

[0235] draw With evolution steps A changing curve;

[0236] In the initial linearly rising region of the curve, a least-squares method is used for linear regression fitting. The slope of the fitted line is divided by the sampling time interval. This is the maximum Lyapunov exponent we are looking for. .

[0237] if This indicates that the system is in a stable periodic or convergent state and the equipment is operating well.

[0238] if This indicates that the system exhibits chaotic characteristics, with adjacent trajectories diverging rapidly, and The larger the value, the more severe the nonlinear degradation inside the equipment, and the higher the probability of sudden failure.

[0239] To provide maintenance personnel with intuitive control criteria, this module includes the geometric deviation index (Freche distance). ) and dynamic divergence index (maximum Lyapunov exponent) ) Perform dimensionless fusion to construct a comprehensive degradation index.

[0240] First of all and Perform normalization processing;

[0241] In historical statistics The maximum possible value is Then the normalized geometric deviation .

[0242] for If it is less than or equal to 0, then the normalized value is... If the value is greater than 0, the Sigmoid function is used for mapping. ,in This is the adjustment coefficient.

[0243] The formula for calculating the overall degradation index is: ;

[0244] in, and The preset weighting coefficients are used, and they satisfy the following conditions: .

[0245] The module internally sets three progressively higher warning thresholds: (Attention threshold) (Warning threshold) and (Dangerous threshold).

[0246] when At this time, the system determines that the device is in a healthy state and does not trigger any action;

[0247] when When the system determines that the equipment is in the early stage of degradation, it generates a "Level 1 Hidden Degradation Early Warning" command and records the degradation trend curve.

[0248] when When the system determines that the equipment has obvious abnormalities, it generates a "Level 2 Intervention Warning" command and extracts the main physical quantity features that caused the trajectory deviation.

[0249] when When the system determines that the equipment is on the verge of a serious malfunction, it immediately generates a "Level 3 Emergency Interruption Warning" command.

[0250] The aforementioned warning instructions and their corresponding CDI values ​​will be used as standardized control signals and directly output to the next-level intelligent linkage execution and control module.

[0251] By employing this computational method based purely on high-dimensional state-space geometry and dynamic evolution, this module successfully overcomes the lag of traditional threshold alarms, achieving advanced quantification and accurate early warning of hidden faults in railway dynamic environment equipment.

[0252] The intelligent linkage execution control module is connected to the operation status assessment and early warning module. It is used to schedule the early warning instructions based on a multi-objective cost function, calculate the control increment using a dynamic matrix control algorithm, perform electrical switching by combining phase-locked loop and zero-crossing prediction, and perform state feedback compensation through a Luneburg observer.

[0253] After the operational status assessment and early warning module outputs the Comprehensive Deterioration Index (CDI) and corresponding graded early warning instructions, the system needs to convert these purely digital diagnostic results into real electromechanical control actions in the physical world.

[0254] The railway dynamic environment system includes a variety of actuators such as high-power uninterruptible power supply (UPS), precision air conditioner, fresh air fan, fire extinguishing cylinder solenoid valve, and high-voltage DC relay.

[0255] The core task of this module is to calculate precise control increments and action sequences based on multi-objective collaborative optimization and deterministic control theory, without relying on human intervention, so as to achieve millisecond-level precise control and physical state correction of dynamic and environmental equipment.

[0256] When the system receives multiple warning commands at the same time (for example, a cabinet temperature exceeds the limit, triggering an air conditioning cooling command, and at the same time, the UPS inverter inside the cabinet shows a deterioration trend, triggering a main and backup power switching command), if they are executed out of order, it may cause an excessive instantaneous surge in the power grid or control logic conflicts.

[0257] This module first constructs a multi-objective cost function to mathematically sort and schedule all control tasks to be executed.

[0258] Suppose the system receives at the current time A set of control tasks to be executed For each task Extract its three core attribute parameters:

[0259] Mission urgency (Directly mapped from the Comprehensive Degradation Index (CDI) of the previous module), transient power consumption required for task execution. And the assessment value of physical losses that may result from delayed task execution. .

[0260] Definition of the first Each task is delayed. The dynamic scheduling cost function under : ;

[0261] in, These are the preset dimensionless weighting coefficients; This is an exponential growth factor that increases the urgency level over time. This represents the maximum remaining available transient power margin in the computer room. To prevent extremely small time constants with a denominator of zero.

[0262] The module searches for the global total cost function by traversing all possible combinations of task execution sequences. The execution sequence that reaches the minimum value.

[0263] in For the first in the sequence The actual waiting time of each task. By solving this discrete combinatorial optimization problem, the system generates a control action timing table strictly arranged according to the time axis, ensuring that high-risk degradation trends are blocked first, without exceeding the total power load of the computer room.

[0264] For environmental parameters such as temperature and humidity in computer rooms, which have large inertia and pure time delay characteristics, traditional proportional-integral-derivative (PID) control is prone to overshoot and oscillation. This module adopts a dynamic matrix control algorithm based on a step response model to calculate the precise control increments for the frequency of the precision air conditioning compressor and the speed of the fan.

[0265] The room temperature response sequence of the precision air conditioner under a unit step control signal was obtained through offline testing, denoted as... ,in To model the time-domain length, a dimension of [missing information] is constructed using this response sequence. Dynamic control matrix ( To predict the time domain, To control the time domain, and ):

[0266] matrix The first column is ;

[0267] matrix The second column is ;

[0268] And so on, the... Listed as .

[0269] Let the current time be The system expects to be available in the future. Within a certain step, the room temperature can smoothly transition to the target set value. .

[0270] Let the initial temperature prediction vector at the current moment be... ;

[0271] Define the quadratic objective function of the control system : ;

[0272] in, For the future The control increment vector (i.e., the adjustment amount of the air conditioning frequency) within a step size.

[0273] This is the error penalty weight matrix;

[0274] The incremental penalty weight matrix is ​​used to control the drastic movements of the control mechanism.

[0275] To make the objective function Minimize, for Taking the partial derivative and setting it equal to zero, we obtain the analytical solution for the control increment: ;

[0276] In actual operation, the module only extracts the calculation results. The first element in As the actual control output at the current moment, it is sent to the inverter controller of the precision air conditioner.

[0277] In the next sampling cycle, the system re-acquires the actual temperature, performs feedback correction, and performs rolling optimization. This pure algebraic matrix operation effectively overcomes the hysteresis effect in environmental control, achieving a stable temperature drop without overshoot.

[0278] When the system determines that a certain mains power input or UPS main circuit has serious deterioration (such as excessive voltage harmonic distortion rate or phase space trajectory divergence) and needs to switch to backup power, directly disconnecting the relay will generate a huge electric arc at the contact point, and may even cause the backend IT equipment to lose power and restart.

[0279] This module achieves seamless switching of electrical circuits through a purely mathematical phase-locked loop algorithm and AC zero-crossing prediction.

[0280] Let the main voltage signal be The backup power supply voltage signal is .

[0281] The module performs orthogonal virtual construction on the backup power supply voltage signal, generating mutually orthogonal signals. shaft and Axis signals: (through the analysis of) (Obtained by performing a quarter-cycle delay or differential calculation).

[0282] Subsequently, the Park transformation is applied to the stationary... Coordinate system rotation to synchronous rotation In the coordinate system.

[0283] Let the phase angle of the output of the internal oscillator of the phase-locked loop be... Transformation matrix Defined as: ;

[0284] Calculated Axis voltage components : When the phase-locked loop locks the backup power supply phase... The value approaches zero.

[0285] Will The error signal is input into a proportional-integral (PI) controller to calculate the adjustment amount of the angular frequency. : ;

[0286] in, and These are the proportional and integral gain constants, respectively. Update the internal phase angle. ,in The rated angular frequency (e.g., 2π×50rad / s).

[0287] Through the above calculations, the system can accurately determine the phase of the backup power supply in real time. When a switching command is received, the system does not act immediately, but instead calculates the main circuit voltage. The next natural midnight time .

[0288] The calculation formula is: ;

[0289] The system in The system issues trigger pulses to disconnect the main circuit and close the backup circuit one instant before the inherent mechanical delay time of the relay action (e.g., 15 milliseconds). Through this precise calculation based on trigonometric functions and coordinate transformations, the system ensures that the relay contacts complete physical opening and closing at the instant when the voltage or current is zero, fundamentally eliminating arcing and guaranteeing the absolute continuity of the power supply to the dynamic ring.

[0290] After the control command is issued, the actual physical action of the actuator may deviate from the theoretical calculation value due to mechanical wear, communication delay or external disturbance.

[0291] To ensure the accurate implementation of control measures, this module constructs a physical state feedback and compensation mechanism based on the Luneburger observer.

[0292] The controlled dynamic and environmental equipment (such as fans and pumps) is abstracted into a linear continuous-time state-space model: ;

[0293] in, This refers to the internal physical state vector of the equipment (such as rotor angular velocity, valve opening). The system output control command vector; This is the output vector actually measured by the sensor; These are the system matrix, input matrix, and output matrix, respectively.

[0294] Due to internal state Often, it is impossible to measure everything directly. The module constructs a Luneburger observer to estimate the state vector in real time. .

[0295] The dynamic equation of the observer is defined as: ;

[0296] in, The output predicted by the observer; This is the observer gain matrix. Error term. Multiply by the gain matrix It is then used as a correction term to continuously adjust the state estimate.

[0297] By employing the pole placement method, a reasonable gain matrix can be designed. This makes the error dynamic equation All eigenvalues ​​are located in the left half of the complex plane, thus ensuring the estimation error. It decays exponentially to zero over time.

[0298] The system calculates and estimates the state in real time. With the desired target state The deviation between the two values. If the deviation exceeds the set tolerance threshold, the system will automatically calculate a compensation control variable. ( (This is the feedback gain matrix), and it is superimposed on the original control command. middle.

[0299] This rigorous observation and compensation mechanism based on state-space equations enables the system to accurately adjust the dynamic environment equipment to a safe and stable operating state even in complex physical environments with external disturbances and mechanical aging. Example 2

[0300] Please see Figure 2 A method for intelligent operation and maintenance management of railway dynamic environment multi-source data fusion includes the following parts:

[0301] The system connects to the environmental equipment to analyze the underlying data stream, uses an adaptive singular value decomposition algorithm to filter out noise, and outputs a continuous time series through dynamic sliding window normalization and Newton's difference quotient polynomial interpolation.

[0302] Construct a three-dimensional spatial distance matrix for the computer room, use cubic basis spline interpolation to upsample low-frequency data, combine discrete cross-correlation function to compensate for physical phase delay, and output a standard multidimensional data matrix;

[0303] Based on the identification framework and spatial distance matrix, the initial basic probability is assigned, the Juselm distance and conflict coefficient are calculated to generate dynamic discount coefficients, and the evidence is fused using the Dempster combination rule to output the state feature vector.

[0304] Using the Tukens embedding theorem, the state feature vector is reconstructed into a high-dimensional phase space trajectory. The discrete Friesian distance and the maximum Lyapunov exponent from the baseline trajectory are calculated, and a comprehensive degradation index is generated by fusing them and an early warning command is output.

[0305] The early warning commands are scheduled based on a multi-objective cost function, the control increment is calculated using a dynamic matrix control algorithm, electrical switching is performed by combining a phase-locked loop and zero-crossing prediction, and state feedback compensation is performed through a Luneburg observer.

[0306] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0307] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0308] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0309] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0310] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A railway dynamic environment multi-source data fusion intelligent operation and maintenance management system, characterized in that, include: The multi-source data acquisition and preprocessing module is used to connect to the environmental monitoring equipment via a physical interface to parse the underlying data stream, filter out noise, and output a continuous time series. The heterogeneous data spatiotemporal alignment module is used to construct a three-dimensional spatial distance matrix for the computer room, upsample low-frequency data, compensate for physical phase delay by combining discrete cross-correlation function, and output a standard multidimensional data matrix. The multidimensional feature conflict resolution module is used to generate initial basic probability assignments based on the physical state space identification framework and spatial distance matrix, calculate the Juselm distance and conflict coefficient to generate dynamic discount coefficients, and use the Dempster combination rule to fuse multi-source evidence to output state feature vectors. The operational status assessment and early warning module is used to reconstruct the state feature vector into a high-dimensional phase space trajectory, calculate the discrete Friesian distance and the maximum Lyapunov exponent of the real-time trajectory deviation from the baseline trajectory, fuse them to generate a comprehensive degradation index and output early warning instructions; The intelligent linkage execution control module is used to schedule early warning commands in a timely manner, calculate control increments using dynamic matrix control algorithms, perform electrical switching by combining phase-locked loop and zero-crossing prediction, and perform state feedback compensation through Luneburg observer.

2. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, In the multi-source data acquisition and preprocessing module, noise filtering includes: A Hankel matrix is ​​constructed by combining a one-dimensional discrete time series with a window length, where the window length is greater than the number of sampling points corresponding to the highest frequency period of the signal. Singular value decomposition of the Hankel matrix yields a left singular vector orthogonal matrix, a right singular vector orthogonal matrix, and a singular value diagonal matrix. Calculate the difference components of adjacent singular values, and find the index corresponding to the maximum peak in the difference sequence as the truncation order to adaptively determine the signal-noise boundary; The singular values ​​before the truncation order are retained and the remaining singular values ​​are set to zero to obtain a new diagonal matrix. The Hankel matrix is ​​then reconstructed, and the anti-diagonal averaging method is used to restore the reconstructed Hankel matrix to the denoised time series.

3. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, The process of outputting a continuous time series includes: A time sliding window is set and continuously advanced over time. Local maximum and local minimum values ​​are found within the time sliding window. Combined with a minimum positive real constant, the time series after noise reduction is dynamically normalized to eliminate the difference in the dimensions of physical quantities of different sensors. When missing data is detected, extract the valid data points before and after the missing data point, and calculate the zero-order difference quotient to the higher-order difference quotient in sequence. A Newton interpolation polynomial is constructed using the quotients of each order. The missing time points are substituted into the Newton interpolation polynomial to calculate the normalized data after interpolation, ensuring the continuity of the higher-order derivatives of the interpolation results.

4. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, The heterogeneous data spatiotemporal alignment module constructs a three-dimensional spatial distance matrix for the computer room, including: A three-dimensional Cartesian coordinate system for the computer room is established with a fixed corner of the computer room as the origin, and the spatial coordinate vector of the sensor is obtained. use ， Calculate the first The sensor and the first Correcting the spatial distance between the sensors, among which For spatial coordinates, Let be the obstacle damping constant. The actual propagation path of environmental parameters is corrected by the obstacle damping constant to determine the number of physical obstacles through which the connection between sensors passes. A symmetric spatial distance matrix is ​​constructed based on the corrected spatial distances between all sensors.

5. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, Low-frequency data is upsampled, and a discrete cross-correlation function is used to compensate for the physical phase delay, outputting a standard multidimensional data matrix, including: A cubic basis spline basis function is defined using a recursive formula. A tridiagonal linear equation system is constructed by combining boundary conditions and the control points are solved by the chasing method. A continuous-time reconstruction function is constructed by linearly combining the control points and the cubic basis spline basis function. The upsampled environmental data sequence is obtained by substituting the high-frequency electrical data timestamp sequence into the continuous-time reconstruction function. Calculate the discrete cross-correlation function between the high-frequency electrical data sequence and the upsampled environmental data sequence, and iterate through the time delay steps to find the delay step corresponding to the maximum value of the discrete cross-correlation function as the actual physical time step; The upsampled environmental data sequence is shifted forward by the corresponding delay steps to compensate for the time axis shift. The missing data points at the end of the shift are then filled in using Newton's difference quotient polynomial extrapolation, and a standard multidimensional data matrix is ​​output.

6. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, The initial basic probability assignment based on the physical state space identification framework and spatial distance matrix includes: Define a set of frameworks for identifying normal, abnormal, and fault states, and set the expected central value and standard deviation for each individual state based on historical benchmark data. use Calculate the measured value Belonging to each single state Gaussian membership ,in The expected central value, Standard deviation; use Calculate the spatial decay weight ,in The spatial attenuation constant is To correct the spatial distance, the Gaussian membership degree is corrected using spatial decay weights. The initial basic probability of the evidence body for a single proposition is assigned, and the initial basic probability of the evidence body for the entire set of uncertainties is also assigned.

7. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, Calculate the Juselm distance and conflict coefficient to generate a dynamic discount coefficient, and fuse multi-source evidence using the Dempster combination rule to output a state feature vector, including: Calculate the conflict coefficient between any two pieces of evidence and the Juselm distance used to quantify the geometric distance in the vector space, using... ， Calculate similarity, where The conflict coefficient, For Juselm distance; The support of the evidence body is calculated and normalized to obtain the relative credibility weight. The initial basic probability is dynamically discounted using the relative credibility weight to obtain the corrected evidence body. The weights of severely conflicting and abnormal data are transferred to the entire set of uncertainties. The modified evidence body is iteratively fused pairwise using the Dempster combination rule and the normalized conflict constant. Under the condition that the maximum probability difference is greater than the decision threshold and the uncertainty is less than the uncertainty threshold, the state feature vector is output based on the maximum confidence principle.

8. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, The operational status assessment and early warning module specifically includes: The mutual information function of the state feature vector is calculated and the time corresponding to the first local minimum of the mutual information function is found as the delay time. The proportion of false nearest neighbors is counted and the dimension corresponding to the first time the proportion of false nearest neighbors drops to near zero is found as the embedding dimension. The state feature vector is expanded into a high-dimensional phase space trajectory through the delayed coordinate method. Construct a time-monotonically increasing coupling sequence between the baseline health trajectory and the real-time running trajectory, and calculate the maximum Euclidean distance between corresponding point pairs under the coupling sequence; A distance matrix is ​​constructed and the recursive calculation rules for matrix elements are set. A dynamic programming algorithm is used to solve for the minimum maximum Euclidean distance among all possible coupling sequences, which is used as the discrete Friesian distance for quantifying geometric shape deviation. In the reconstructed high-dimensional phase space, the nearest neighbor of the phase point is found, and the time interval is restricted to be greater than the average orbital period to exclude adjacent points on the same orbit, so as to calculate the average logarithmic distance between adjacent trajectories with the number of evolution steps; The least squares method was used to perform linear regression fitting on the initial linearly rising region of the logarithmic distance-mean curve, and the maximum Lyapunov exponent was calculated by dividing the slope of the fitted line by the sampling time interval. The maximum Lyapunov exponent is mapped, and the discrete Friesian distance and the mapped maximum Lyapunov exponent are normalized and weighted to generate a comprehensive degradation index. The index is then compared with the preset progressive warning threshold to output the corresponding graded warning command.

9. The intelligent operation and maintenance management system for railway dynamic and environmental multi-source data fusion according to claim 1, characterized in that, The process of calculating the control increment using a dynamic matrix control algorithm, combined with phase-locked loop and zero-crossing prediction to perform electrical switching is as follows: Extract the urgency, transient power consumption, and physical loss assessment values ​​of the early warning command task to construct a multi-objective cost function for time-series scheduling; A dynamic control matrix is ​​constructed using the response sequence under a unit step control signal. A quadratic objective function is constructed, which includes an error penalty weight matrix and a control increment penalty weight matrix. The control increment is calculated by taking the partial derivative of the quadratic objective function. The backup power supply voltage signal is orthogonally virtualized, and the Park transform is applied to calculate the orthogonal voltage component input to the proportional-integral regulator. The internal phase angle is updated to lock the backup power supply phase. The next natural zero-crossing time of the main voltage is calculated to send a trigger pulse to perform electrical switching in advance of the inherent delay time of the relay mechanical action. A Luenberger observer is constructed to estimate the physical state vector in real time. The deviation between the estimated state and the desired target state is calculated to generate a compensating control quantity, which is then superimposed on the control increment.

10. A method for intelligent operation and maintenance management of railway dynamic and environmental data fusion, comprising a railway dynamic and environmental data fusion intelligent operation and maintenance management system according to any one of claims 1-9, characterized in that, include: The system connects to the environmental equipment to analyze the underlying data stream, uses an adaptive singular value decomposition algorithm to filter out noise, and outputs a continuous time series through dynamic sliding window normalization and Newton's difference quotient polynomial interpolation. Construct a three-dimensional spatial distance matrix for the computer room, use cubic basis spline interpolation to upsample low-frequency data, combine discrete cross-correlation function to compensate for physical phase delay, and output a standard multidimensional data matrix; Based on the identification framework and spatial distance matrix, the initial basic probability is assigned, the Juselm distance and conflict coefficient are calculated to generate dynamic discount coefficients, and the evidence is fused using the Dempster combination rule to output the state feature vector. The state feature vector is reconstructed into a high-dimensional phase space trajectory using the Tukens embedding theorem. The discrete Friesian distance and the maximum Lyapunov exponent of deviation from the baseline trajectory are calculated and fused to generate a comprehensive degradation index and output an early warning command. The early warning commands are scheduled based on a multi-objective cost function, the control increment is calculated using a dynamic matrix control algorithm, electrical switching is performed by combining a phase-locked loop and zero-crossing prediction, and state feedback compensation is performed by a Luneburg observer.

Images