Full-chain health prediction and energy efficiency optimization method for rail transit rotating equipment

Abstract

The invention relates to the technical field of equipment health prediction, in particular to a rail transit rotating equipment full-chain health prediction and energy efficiency optimization method, which comprises the following steps of: performing Markov chain conversion on a transfer proportion and recursion on a stage proportion on stage labeling, converting a service life occupation proportion into a service life residual proportion and performing weighted summary on the service life residual proportion and a stage penalty coefficient; a multi-unit state is updated according to time steps, cutting and writing-back are carried out on a phase difference threshold value and a power consumption boundary which are consistent in direction and identical in increment direction, Kalman filtering selects an updating path according to a residual mean square and outputs a posterior interval set, the interval boundary and sensitivity difference are linked to generate a marginal contribution sequence, and the marginal contribution sequence is merged and sorted; meanwhile, traversal mileage points are recorded to form a trigger list, gradient, curve, power under station distance index, retrogradation, amplitude limiting gear and energy consumption calculation are linked, a posterior interval slope penalty term is superposed to screen out a cross-border term, a working condition action index table is formed, chain-level degradation direction consistency is enhanced, and false triggering and missing triggering probabilities are converged.

Description

Technical Field

[0001] This invention relates to the field of equipment health prediction technology, and in particular to a method for full-chain health prediction and energy efficiency optimization of rotating equipment in rail transit. Background Technology

[0002] The field of equipment health prediction technology aims to build a computable model of equipment operating status. By establishing deterministic functions for multiple operating parameters, it can predict the performance degradation trajectory and calculate the remaining life index, making the operation and maintenance activities of equipment predictable. It can also build a correlation structure for the equipment execution state variables and identify abnormal change execution patterns, so that potential failure modes can be identified before actual failure. This generates decision-making quantities for scheduling maintenance plans, reducing sudden downtime events, and stabilizing operating efficiency.

[0003] The purpose of the whole-chain health prediction and energy efficiency optimization method for rotating equipment in rail transit is to establish degradation models, health index models, and energy consumption function models for multiple rotating units, including motor rotors, bearing assemblies, gear transmission groups, and braking actuators. By calculating the equipment performance degradation trajectory and energy usage distribution, the method aims to make maintenance plans, operation scheduling, and energy consumption regulation predictable and executable, reduce unplanned downtime during the equipment's operating cycle, improve the quantifiability of health status, reduce energy consumption per unit operating distance, and maintain the availability of the equipment group above a set threshold.

[0004] Existing technologies focus on mapping multiple operating parameters to deterministic functions, separating the implementation of state variable correlation structures and abnormal change pattern recognition. Chain-level multi-unit coupling relies heavily on empirical weights and static rules. When cross-unit symbol consistency constraints are lacking, phenomena such as traction load increases, braking power consumption fluctuations, and meshing impact phase drift may be interpreted by different functions, leading to contradictory degradation direction judgments under the same operating condition. This causes maintenance plan trigger point drift, and remaining life indicators often provide single-point values ​​rather than intervals. After operating condition switching and noise disturbances enter the calculation link, repeated boundary violations are likely to occur near the threshold, forcing frequent adjustments to maintenance schedules. When pattern recognition relies on discrete features and single-channel anomalies, it is difficult to separate overlapping scenarios of potential fault modes. For example, when bearing clearance slowly expands and rotor imbalance increases simultaneously, vibration energy and current fluctuations change at the same time. Deterministic functions tend to attribute changes to a single component, and spare parts preparation and maintenance window selection deviate from the dominant unit. Energy consumption adjustment is often implemented separately from health prediction, and power level and regeneration ratio adjustments lack life lower bound constraints. This leads to sudden changes in health indicators after energy consumption parameter adjustments, making it difficult to trace the source. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit, comprising the following steps:

[0007] S1: Based on the full-chain unit list of rotating equipment in rail transit, the axle box bearings, gear pairs, traction motor rotors and brake discs are addressed and entered into the table. The state components are paired with the observations item by item and the symbolic constraints are solidified to establish a constraint set.

[0008] S2: Based on the constraint set, organize the key performance index sequence along the mileage and cycle number, calculate the rate of change of adjacent points to summarize the decay rate parameter, label the degradation phase number according to the rate of change segmentation rule, use a Markov chain, convert the stage labeling to the transfer ratio and derive the stage proportion sequence, fuse the lifetime occupancy ratio and stage penalty coefficient and summarize to obtain the health index trajectory.

[0009] S3: Based on the health index trajectory, update the state of multiple units by time step and calculate the residual sequence. Perform consistent direction, same increment, phase difference threshold, power consumption boundary pruning and write-back item by item. Use Kalman filtering and select the update path according to the residual mean square comparison result to obtain the posterior interval set.

[0010] S4: Based on the posterior interval set, the lifetime difference of the perturbation state is used to generate a sensitivity table. The marginal contribution sequence is obtained by multiplying and adding the posterior mean. The couplings, gearboxes, axle boxes, and traction motors are merged and sorted, and the crossing mileage points are marked to obtain the trigger list.

[0011] S5: Based on the trigger list, select power, regeneration, and limiting range according to slope, curve, and station distance, calculate energy consumption, add multi-unit degradation posterior interval set slope penalty term, remove out-of-bounds terms, and obtain the working condition action index table.

[0012] As a further aspect of the present invention, the constraint set includes a unit identifier set, a state variable naming set, an observation naming set, a symbol constraint set, and a load level segmentation set; the health index trajectory includes a chain health index sequence, a unit health index sequence, a degradation phase number sequence, a stage proportion sequence, and a remaining life proportion sequence; the posterior interval includes a rotor imbalance interval, a bearing radial clearance interval, a tooth flank clearance drift interval, a braking friction attenuation interval, and a confidence boundary set; the trigger list includes a crossing mileage point set, a warning threshold set, a shutdown threshold set, a unit contribution ranking set, and a unit priority set; and the action index includes a slope level entry, a curve radius level entry, an inter-station distance level entry, a traction power gear entry, a regenerative braking ratio gear entry, and a torque limiting gear entry.

[0013] As a further aspect of the present invention, the specific steps for generating the constraint set are as follows:

[0014] Based on the full-chain unit list of rotating equipment in rail transit, the axle box bearings, gear pairs, traction motor rotors and brake discs are addressed and entered into the table. State component fields are written to each unit and matched with the observation fields item by item. Symbolic constraint values ​​are written according to the pairing relationship to generate a pairing constraint table.

[0015] Based on the pairing constraint table, a one-to-one correspondence check is performed on the state component field and the observation field, rows with symbolic constraint conflicts are deleted and missing rows are filled in, load level segment boundary values ​​are written and segment intervals are locked, and a constraint set is established.

[0016] As a further aspect of the present invention, the specific steps for generating the health index trajectory are as follows:

[0017] Based on the constraint set, the key performance indicator sequence is organized along the mileage and cycle number, grouped by unit number and sorted by time index, missing points and reverse points are deleted, continuous intervals are locked and written into the interval start and end indexes, and a performance sequence table is generated.

[0018] Based on the performance sequence table, the rate of change sequence is generated by the difference between adjacent points, the rate of change is aggregated according to the window length and the window statistics are output, the abrupt windows are removed according to the abnormal threshold and the continuous windows are retained, the decay rate parameters are summarized and written into the cell row, and a degradation labeling table is generated.

[0019] Based on the degradation label table, a Markov chain is used to calculate the transfer ratio from the stage label count and derive the stage proportion sequence. The lifespan occupancy ratio is converted into the lifespan remaining ratio. The stage penalty coefficient and the lifespan remaining ratio are added according to weight and written into the chain summary row to obtain the health index trajectory.

[0020] As a further aspect of the present invention, the Markov chain first defines the stage set as normal, mildly degraded, moderately degraded, and severely degraded. The stage label sequence is read sequentially along the mileage and cycle number. The occurrence frequency of stage pairs between adjacent time points is counted and written into the transition count matrix. The transition count matrix is ​​normalized row by row to obtain the transition ratio matrix. The current time point stage ratio vector is multiplied by the transition ratio matrix and iteratively updated according to the prediction step number. The stage ratio sequence at each prediction time point is output. The stage ratio sequence is converted into a stage penalty coefficient sequence according to a preset mapping table and written into the chain summary row.

[0021] As a further aspect of the present invention, the specific steps for generating the posterior interval set are as follows:

[0022] Based on the health index trajectory, the equivalent unbalance of the rotor, the equivalent radial clearance of the bearing, the tooth flank clearance drift, and the braking friction attenuation are written according to the time step. The traction current fluctuation, shaft end vibration energy, meshing impact index, and braking power consumption index are aligned according to the unit number. The observation difference is calculated and written into the residual row to generate the state residual table.

[0023] Based on the state residual table, the direction of the unbalance increment is determined, the radial clearance increment is determined, the phase difference of the tooth flank clearance drift and impact index is determined, the boundary of friction decay and power consumption is determined, the out-of-bounds items are trimmed and written back and the trimming flag is marked, and a constraint state table is generated.

[0024] Based on the constraint state table, Kalman filtering is used to calculate the residual mean square at each time step and compare it with the threshold. Above the threshold, multiple samples are written and quantiles are taken to form upper and lower bounds. Below the threshold, single-point updates are written and the upper and lower bounds are expanded according to the cumulative deviation. The upper and lower bound rows are summarized by unit number to obtain the posterior interval set.

[0025] As a further embodiment of the present invention, the Kalman filter first reads the state estimate of the previous time step, the covariance matrix of the previous time step, the observation vector of the current time step, the observation matrix, the state transition matrix, the process noise covariance matrix, and the observation noise covariance matrix from the constraint state table. It then calculates the predicted state value and the predicted covariance matrix, calculates the observation residual vector and the residual covariance matrix, calculates the Kalman gain matrix, adds the predicted state value to the Kalman gain matrix and the observation residual vector, and performs a combined operation on the Kalman gain matrix, the observation matrix, and the predicted covariance matrix to obtain the updated covariance matrix. Finally, it squares the difference between the updated state value and the observation vector, calculates the mean square of the residual, compares it with a threshold, selects the updated state value and a multi-point sample set based on the comparison result, calculates the upper and lower bounds of the state components according to the updated covariance matrix, and summarizes them by unit number to form a posterior interval set.

[0026] As a further aspect of the present invention, the specific steps for generating the trigger list are as follows:

[0027] Based on the posterior interval set, a perturbation value is taken between the upper and lower bounds of each state component interval, the difference in lifetime reduction before and after the perturbation is calculated and written into the cell row, the difference results are summarized according to the state component fields, and a sensitivity table is generated.

[0028] Based on the sensitivity table, the sensitivity values ​​are multiplied by the posterior mean item by item and summed to obtain the marginal contribution sequence. The sequence is aggregated and summed according to coupling, gearbox, axle box, and traction motor and arranged in descending order. The extrapolated sequence is compared with the lower threshold point by point and the crossing mileage points are recorded to obtain the trigger list.

[0029] As a further aspect of the present invention, the crossing mileage point is specifically defined as the position on the mileage axis where the extrapolation sequence changes from above the lower threshold to below the lower threshold. The extrapolation sequence is discretized into adjacent mileage points and corresponding values ​​according to a fixed mileage step size. When the value of a certain mileage point is above the lower threshold and the value of the next mileage point is below the lower threshold, and when the value of a certain mileage point is equal to the lower threshold and the value of the next mileage point is below the lower threshold, it is determined that there is a crossing mileage point between two adjacent mileage points. The crossing mileage point is determined by the ratio of the change in value between the mileage of the two adjacent mileage points. When there are multiple crossings, the first crossing mileage point is recorded, and subsequent crossing mileage points are not recorded. When no crossing occurs, the crossing mileage point is recorded as a null value.

[0030] As a further aspect of the present invention, the specific steps for generating the working condition action index table are as follows:

[0031] Based on the trigger list, the slope level, curve radius level, and station distance level are concatenated into a working condition index key. The traction power gear, regenerative braking ratio gear, and torque limiting gear are enumerated according to the working condition index key and written into the row to generate a working condition action candidate table.

[0032] Based on the aforementioned candidate table of working conditions and actions, the energy consumption per unit distance is calculated using the speed curve and traction power. The action score is obtained by adding the slope penalty term of the multi-unit degraded posterior interval set to the energy consumption value. Out-of-bounds action rows are deleted, and entries are output according to the working condition index key to obtain the working condition action index table.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] In this invention, the entire chain unit list drives the axle box bearing, gear pair, traction motor rotor, and brake disc with unified addressing. The state components and observations are paired item by item and written with symbol constraints, so that the traction current fluctuation, shaft end vibration energy, meshing impact index, braking power consumption index, imbalance, radial clearance, tooth side clearance drift, and friction attenuation form a verifiable correlation. After the key performance indicators are sorted by mileage and cycle number, the change rate of adjacent points is extracted and the degradation phase is marked.

[0035] In this invention, the stage label is converted into the transition ratio through Markov chain and the stage proportion is recursively deduced. The lifetime occupancy ratio is converted into the lifetime remaining proportion and weighted and summarized with the stage penalty coefficient to obtain a continuously extrapolated health index trajectory. The multi-unit state is updated according to time steps and pruning and writing back are performed on the direction consistency, incremental direction, phase difference threshold, and power consumption boundary.

[0036] In this invention, Kalman filtering selects the update path based on the residual mean square and outputs a posterior interval set. The interval boundary and sensitivity difference are linked to generate a marginal contribution sequence and merged and sorted. At the same time, the crossing mileage points are recorded to form a trigger list. Power, regeneration, amplitude limit and energy consumption calculations are linked under the slope, curve and station distance indexes and a posterior interval slope penalty term is superimposed to filter out out-of-bounds terms, forming an operating condition action index table. The consistency of chain-level degradation direction is enhanced, the probability of false triggering and missed triggering converges, the lifetime interval expression covers operating condition fluctuations, and the maintenance trigger window and dominant unit sorting have verifiable basis. The operating energy consumption and health constraints are output in the same table for easy scheduling implementation. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the main steps of the present invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] Example 1 Please see Figure 1 This invention provides a technical solution: a method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit, comprising the following steps:

[0040] S1: Based on the full-chain unit list of rotating equipment in rail transit, the axle box bearings, gear pairs, traction motor rotors and brake discs are addressed and entered into the table. The state components are paired with the observations item by item and the symbolic constraints are solidified to establish a constraint set.

[0041] S2: Based on the constraint set, organize the key performance index sequence along the mileage and cycle number, calculate the rate of change of adjacent points to summarize the decay rate parameter, label the degradation phase number according to the rate of change segmentation rule, use Markov chain, convert the stage label to the transfer ratio and derive the stage proportion sequence, fuse the lifetime occupancy ratio and stage penalty coefficient and summarize to obtain the health index trajectory.

[0042] S3: Based on the health index trajectory, update the state of multiple units by time step and calculate the residual sequence. Perform consistent direction, same increment, phase difference threshold, power consumption boundary pruning and write-back item by item. Use Kalman filtering and select the update path according to the residual mean square comparison result to obtain the posterior interval set.

[0043] S4: Based on the posterior interval set, the lifetime difference of the perturbation state is used to generate a sensitivity table. The marginal contribution sequence is obtained by multiplying and adding the posterior mean. The couplings, gearboxes, axle boxes, and traction motors are merged and sorted, and the crossing mileage points are marked to obtain the trigger list.

[0044] S5: Based on the trigger list, select power, regeneration, and limiting range according to slope, curve, and station distance, calculate energy consumption, add multi-unit degradation posterior interval set slope penalty term, remove out-of-bounds terms, and obtain the working condition action index table.

[0045] The constraint set includes a set of unit identifiers, a set of state variable names, a set of observation names, a set of symbolic constraints, and a set of load level segments. The health index trajectory includes a chain health index sequence, a unit health index sequence, a degradation phase number sequence, a stage proportion sequence, and a remaining life proportion sequence. The posterior interval includes a rotor imbalance interval, a bearing radial clearance interval, a tooth flank clearance drift interval, a braking friction attenuation interval, and a set of confidence boundaries. The trigger list includes a set of crossing mileage points, a set of warning thresholds, a set of shutdown thresholds, a set of unit contribution rankings, and a set of unit priorities. The action index includes entries for slope level, curve radius level, inter-station distance level, traction power gear, regenerative braking ratio gear, and torque limiting gear.

[0046] The specific steps for generating the constraint set are as follows:

[0047] Based on the full-chain unit list of rotating equipment in rail transit, the axle box bearings, gear pairs, traction motor rotors and brake discs are addressed and entered into the table. State component fields are written to each unit and matched with the observation fields item by item. Symbolic constraint values ​​are written according to the pairing relationship to generate a pairing constraint table.

[0048] Based on the pairing constraint table, a one-to-one correspondence check is performed on the state component field and the observation field, the row of symbol constraint conflict is deleted and the missing row is filled in, the load level segment boundary value is written and the segment interval is locked, and a constraint set is established.

[0049] Based on the complete chain unit list of rotating equipment in rail transit, a coding and addressing rule is adopted to address axle box bearings, gear pairs, traction motor rotors, and brake discs in the table. The coding and addressing rule takes 2 digits of the line number with a value range of 01 to 99, 3 digits of the vehicle number with a value range of 001 to 999, 1 digit of the bogie position with a value range of 1 to 2, and 1 digit of the unit category with a value range of 1 to 4, and the unit serial number is 2 digits with a value range of 01 to 99. The above fields are concatenated in order to form the unit number field, which is then assigned to each unit. The state component field is written and paired with the observation field item by item. The field name matching rule is adopted with complete consistency as the matching condition. The four state components, rotor equivalent unbalance, bearing equivalent radial clearance, tooth side clearance drift, and braking friction attenuation, are used to establish a one-to-one pairing row with the four observations, traction current fluctuation, shaft end vibration energy, meshing impact index, and braking power consumption index. Symbol constraint values ​​are written according to the pairing relationship. The symbol constraint values ​​are limited to three categories: positive, negative, and zero constraints. Positive and negative constraints are added or removed respectively. Same-direction and reverse zero constraints do not participate in the judgment. A pairing constraint table is generated.

[0050] Based on the pairing constraint table, a consistency check rule is used to perform a one-to-one correspondence check on the state component fields and the observation fields. The consistency check rule limits each state component field to only one pairing row and each observation field to only one pairing row. When duplicate pairing rows occur, they are retained in priority order, which is limited to power consumption related rows, shock related rows, vibration related rows, and current related rows. Conflicting symbol constraint rows are deleted and missing rows are filled in. The conflict row judgment condition is limited to the same pairing row having both positive and negative values. The missing row filling rule is limited to filling in according to the lexicographical order of the state component fields and setting the symbol constraint value to zero. The load level segment boundary values ​​are written and the segment intervals are locked. The load intensity values ​​are divided into four load segmentation rules: no-load interval (value range 0 to 25), light-load interval (value range 26 to 50), medium-load interval (value range 51 to 75), and heavy-load interval (value range 76 to 100). The lower and upper bounds are written for each segment and the lock flag field is set to 1. A constraint set is established.

[0051] The specific steps for generating a health index trajectory are as follows:

[0052] Based on the constraint set, the key performance indicator sequence is organized in order of mileage and number of cycles, grouped by unit number and sorted by time index, missing points and reverse points are deleted, continuous intervals are locked and written into the interval start and end indexes to generate a performance sequence table.

[0053] Based on the performance sequence table, the rate of change sequence is generated by the difference between adjacent points, the rate of change is aggregated according to the window length and the window statistics are output, the abrupt windows are removed according to the abnormal threshold and the continuous windows are retained, the decay rate parameters are summarized and written into the cell row, and a degradation labeling table is generated.

[0054] Based on the degradation label table, a Markov chain is used to convert the transfer ratio from the stage label count and derive the stage proportion sequence. The lifespan occupancy ratio is converted into the lifespan remaining ratio. The stage penalty coefficient and the lifespan remaining ratio are added according to weight and written into the chain summary row to obtain the health index trajectory.

[0055] Based on the constraint set, the key performance indicator sequence is organized along the mileage and cycle count order using sorting and merging rules. The sorting and merging rules take the unit number field, time index field, mileage field, and cycle count field as key values ​​and perform stable sorting in ascending order of unit number and time index. Missing points are deleted. Missing points are determined by key performance indicators that are null or non-numerical. Reversal points are deleted. Reversal points are determined by two adjacent rows of time index that decreases and two adjacent rows of mileage that decreases. Continuous interval locking is performed. Continuous interval determination is determined by two adjacent rows of time index that have a difference of no more than 60 and two adjacent rows of mileage that have a difference of no more than 5. Interval start and end indexes are written. The interval start and end indexes are taken from the first and last row time indices of each continuous interval to generate a performance sequence table.

[0056] Based on the performance sequence list, a third-order polynomial degradation curve fitting algorithm is used to generate decay rate parameters. The third-order polynomial degradation curve fitting algorithm takes mileage and number of cycles as independent variables, sets the power to 3, sets the regularization coefficient to 0.001, sets the upper limit of the number of iterations to 200, and sets the convergence threshold to 0.000001. After outputting the coefficient vector, the change rate sequence of adjacent points is calculated for each time index of the fitted curve. The change rate is aggregated by a sliding window aggregation rule with a window length of 50 and a step size of 10. The window statistics are taken as the median and interquartile range. An outlier threshold is removed. The outlier threshold is set when the absolute value of the median change rate of the window exceeds 0.02 and the interquartile range exceeds 0.05. Continuous windows are retained. The continuous window determination is that the number of retained windows is not less than 3 and the difference between the time indices of adjacent retained windows does not exceed 30. The decay rate parameters are summarized and written into the cell row to generate a degradation label table.

[0057] Based on the degradation label table, the Discrete Markov Chain (DMC) fault stage transition algorithm is used to convert the stage label count into the transition ratio and derive the stage proportion sequence. The DMC fault stage transition algorithm limits the stage set to four categories: normal, mild degradation, moderate degradation, and severe degradation. The transition count is taken as the number of occurrences of stage pairs between two adjacent time steps and written into a 4x4 count table. The transition ratio is obtained by normalizing the count table row by row and smoothing it by adding 1 with a smoothing coefficient of 1. The prediction step is set to 100 and the initial stage proportion is taken as the current stage's heat vector. The Linear Differential Lifetime Dissipation Accumulation Algorithm is used to convert the lifetime occupancy ratio into the lifetime remaining ratio. The Linear Differential Lifetime Dissipation Accumulation Algorithm takes the load intensity value range of 0 to 100 and takes the cycle number increment with a lifetime occupancy ratio step size of 0.001. The stage penalty coefficient and the lifetime remaining ratio are added according to weight and written into the chain summary row. The weight is set to a stage penalty coefficient of 0.6 and a lifetime remaining ratio of 0.4, resulting in the health index trajectory.

[0058] The Markov chain first limits the stage set to normal, mildly degraded, moderately degraded, and severely degraded. It reads the stage label sequence in order of mileage and cycle number, counts the occurrence of stage pairs between adjacent time points and writes them into the transition count matrix. It normalizes the transition count matrix row by row to obtain the transition ratio matrix. It performs matrix multiplication between the current time point stage ratio vector and the transition ratio matrix and iterates and updates it according to the prediction steps. It outputs the stage ratio sequence at each prediction time point. It converts the stage ratio sequence into the stage penalty coefficient sequence according to the preset mapping table and writes it into the chain summary row.

[0059] Markov chains, according to the formula:

[0060] in: Indicates time Improvement of the health index for rotating equipment in rail transit This represents the time step index of the Markov chain during the entire operation of rotating equipment in rail transit. This indicates the number of multiple degradation stages in the degradation process of rotating equipment in rail transit. Indicates the degradation stage index. Indicates time The equipment is in the degradation stage The percentage of each stage Indicates time The percentage of lifespan occupied by rotating equipment in rail transit. Indicates the degeneration stage The stage penalty coefficient, Indicates the degeneration stage The degradation sensitivity weight coefficient, Indicates time Correction factor for remaining lifetime uncertainty;

[0061] Execution process: First, based on the degradation label table, the transition counts between multiple degradation stages are statistically analyzed, and a Markov chain transition probability matrix is ​​constructed. Then, the state distribution is propagated stepwise according to the matrix to obtain the time step. In the degenerative stage stage proportion Then, the lifespan occupancy rate is calculated as the ratio of the equipment's cumulative operating time to its designed lifespan. And converted into the percentage of remaining lifespan (1- Next, an objective function is established based on the failure loss, maintenance cost, and performance degradation corresponding to each degradation stage, and the stage penalty coefficient is solved using least squares and gradient descent. Calculate the percentage of each degradation stage. The standard deviation over time is used to measure degradation sensitivity, and the standard deviation is normalized to generate degradation sensitivity weighting coefficients. Then at the moment Multiple lifetime trajectory predictions are performed based on the Markov chain state distribution, and the variance of the resulting remaining lifetime samples is normalized to generate a lifetime remaining uncertainty correction coefficient. Ultimately , , , and Substituting into the formula and summing, we obtain the full-chain health index of rotating rail transit equipment, which can simultaneously reflect the distribution of degradation stages, lifespan consumption, attenuation intensity, degradation sensitivity, and lifespan uncertainty.

[0062] The specific steps for generating the posterior interval set are as follows:

[0063] Based on the health index trajectory, the equivalent unbalance of the rotor, the equivalent radial clearance of the bearing, the tooth flank clearance drift, and the braking friction attenuation are written according to the time step. The traction current fluctuation, shaft end vibration energy, meshing impact index, and braking power consumption index are aligned according to the unit number. The observation difference is calculated and written into the residual row to generate the state residual table.

[0064] Based on the state residual table, the direction of the unbalance increment is determined, the radial clearance increment is determined, the phase difference of the tooth flank clearance drift and impact index is determined, the boundary of friction decay and power consumption is determined, the over-limit items are trimmed and written back and the trimming flag is marked, and the constraint state table is generated.

[0065] Based on the constraint state table, Kalman filtering is used to calculate the residual mean square at each time step and compare it with the threshold. Above the threshold, multiple samples are written and quantiles are taken to form the upper and lower bounds. Below the threshold, single-point updates are written and the upper and lower bounds are expanded according to the cumulative deviation. The upper and lower bound rows are summarized by cell number to obtain the posterior interval set.

[0066] Based on the health index trajectory, the equivalent unbalance of the rotor, the equivalent radial clearance of the bearing, the tooth flank clearance drift, and the braking friction attenuation are written according to the time step using the time step alignment interpolation rule. The time step alignment interpolation rule takes a sampling period of 0.1 seconds and unifies the time index to a millisecond-level integer and uses the unit number as the primary key. Observation alignment is performed, and the traction current fluctuation, shaft end vibration energy, meshing impact index, and braking power consumption index are matched with the unit number and time index. When missing, forward hold is used and the hold step is limited to no more than 5. Observation difference is calculated by subtracting the observation sliding mean from the current observation value. The sliding window length is 30 steps and the sliding mean update step size is 1 step. Residuals are written, and the field corresponding to the traction current difference of the unbalance is recorded as r1, the field corresponding to the vibration difference of the radial clearance is recorded as r2, the field corresponding to the impact difference of the tooth flank clearance drift is recorded as r3, and the field corresponding to the power consumption difference of the friction attenuation is recorded as r4. These are written to the same row to generate a state residual table.

[0067] Based on the state residual table, a sign consistency judgment rule is used to determine the direction consistency of the unbalance increment. This rule considers the unbalance increment and the traction current difference increment to have the same sign as the unbalance increment to pass, sets the judgment window length to 10 steps, and allows 2 reverse reversals. A monotonic same-direction judgment rule is used to determine the radial clearance increment to pass, setting the minimum increment threshold to 0.0005 and the number of consecutively satisfied steps to 5. Finally, a phase difference judgment rule is used to determine the tooth flank clearance drift and meshing impact indicators. The phase difference determination rule is that the difference between the zero-crossing positions of the two sequences is no more than 3 steps to pass, and the zero-crossing search window length is 60 steps. The boundary determination rule is used to perform boundary determination on friction attenuation and braking power consumption. The boundary determination rule is that the difference between braking power consumption is in the range of 0 to 120, the friction attenuation is in the range of 0 to 1, and the proportion of the number of times the two change in the same direction is not less than 0.7. The over-boundary item is truncated and written back. The over-boundary value is truncated to the corresponding upper and lower boundaries and written into the truncation flag bits c1 to c4 with values ​​of 0 and 1 and saved in the same column as the unit number and time index to generate a constraint state table.

[0068] Based on the constraint state table, Kalman filtering is used to calculate the residual mean square step by step and compare it with a threshold. The Kalman filter takes a 4x4 identity matrix for the state transition matrix and adds a 0.995 attenuation coefficient to the diagonal. The observation matrix is ​​also a 4x4 identity matrix. The initial state vector is the first row value of each state component. The initial covariance matrix has 1 on the diagonal and 0 on the off-diagonal. The process noise covariance matrix has a diagonal of 0.01 and is segmented according to load level, with the heavy load segment adjusted to 0.02. The observation noise covariance matrix has a diagonal of 0.05 and the vibration channel is adjusted to 0.08. The predicted state and the predicted state are calculated according to the prediction update order. Measure the covariance and calculate the Kalman gain, write back the updated state and updated covariance, and perform residual mean square calculation. The residual mean square calculation takes the squares of r1 to r4 and calculates the mean, and sets the threshold to 0.04. Then, perform update path selection. Above the threshold, write multiple samples and take quantiles to form upper and lower bounds, taking quantiles of 0.1 and 0.9 and setting the number of sample points to 200. Below the threshold, write single-point updates and expand the upper and lower bounds according to the cumulative deviation, taking the expansion step size of 0.005 and setting the cumulative window length to 20 steps. Summarize the upper and lower bound rows according to the cell number and write the lower bound fields u1 to u4 and the upper bound fields v1 to v4 to obtain the posterior interval set.

[0069] Kalman filtering first reads the state estimate, covariance matrix, observation vector, observation matrix, state transition matrix, process noise covariance matrix, and observation noise covariance matrix from the constraint state table at the previous time step. It then calculates the predicted state value and the prediction covariance matrix, the observation residual vector and the residual covariance matrix, and the Kalman gain matrix. The predicted state value is multiplied by the Kalman gain matrix and the observation residual vector, and the updated state value is obtained. The Kalman gain matrix is ​​combined with the observation matrix and the prediction covariance matrix to obtain the updated covariance matrix. The mean square of the difference between the updated state value and the observation vector is calculated to obtain the residual mean square, which is compared with a threshold. Based on the comparison results, updated state values ​​and a multi-point sample set are selected. The upper and lower bounds of the state components are calculated according to the updated covariance matrix, and the results are summarized by cell number to form a posterior interval set.

[0070] Markov chains, according to the formula:

[0071] in: Indicates time step Unit Number Improved estimates of observational residual covariance. This indicates the unit number used to distinguish different functional units within the entire chain of rotating equipment in rail transit. This represents the time step index corresponding to the device's operation process in Kalman filter sequence derivation. Indicates unit number State covariance amplification weighting coefficient, Indicates unit number The observation mapping matrix, Indicates time step The predicted covariance matrix, Indicates unit number The observation noise covariance matrix, Indicates time step The observation noise adjustment weighting coefficient, Indicates time step Unit Number The squared measure of energy efficiency deviation Indicates unit number Energy efficiency sensitive weighting coefficient, Indicates time step Unit Number The constraint violates the strength quantification value. Indicates unit number The constraint violates the penalty weight coefficient;

[0072] Execution process: Constructing time steps using a constraint state table and a Kalman filter prediction stage. Predictive covariance matrix Using unit number Observation mapping matrix The predicted covariance is mapped to the constrained observation space and formed into a term. By unit number State covariance amplification weighting coefficient Adjusting the proportion of influence in residual covariance to give greater weight to cells with high prediction uncertainty in the posterior construction, based on cell numbering. Observation noise covariance matrix Characterize the energy of the constraint measurement error, and in time steps Observation noise adjustment weighting coefficient Adjust the impact magnitude of the noise term according to the degree of operational disturbance, and then assign it to the unit number. Energy efficiency deviation squared measure Calculate the degree of deviation between the target energy efficiency and the actual energy efficiency of the equipment, and use an energy efficiency sensitivity weighting coefficient. The energy efficiency deviation is mapped to the residual covariance improvement term, so that cells with significant energy efficiency deviations are strengthened in the posterior interval update, and then the cells are numbered... At time step The constraint violates the strength quantification value Describe the degree of exceeding the limit, and use constraint violation penalty weighting coefficients. The impact of violations is enhanced, making constraint anomalies more significantly increase the residual estimates, and ultimately the weighted sum of the four terms constitutes the improved residual covariance. and with The process of driving residual mean square calculation, threshold judgment, multi-point sample quantile update, single-point cumulative deviation expansion, and upper and lower bound row construction forms the posterior interval set of the entire chain operation state of rotating equipment in rail transit.

[0073] The specific steps for generating the trigger list are as follows:

[0074] Based on the posterior interval set, the perturbation value is taken between the upper and lower bounds of each state component interval, the difference in lifetime reduction before and after the perturbation is calculated and written into the cell row, the difference results are summarized according to the state component fields, and a sensitivity table is generated.

[0075] Based on the sensitivity table, the sensitivity values ​​are multiplied by the posterior mean item by item and summed to obtain the marginal contribution sequence. The sequence is aggregated and summed according to coupling, gearbox, axle box, and traction motor and arranged in descending order. The extrapolated sequence is compared with the lower bound of the threshold point by point and the crossing mileage points are recorded to obtain the trigger list.

[0076] Based on the posterior interval set, a hierarchical equidistant perturbation sampling rule is adopted to take perturbation values ​​between the upper and lower boundaries of each state component interval. The hierarchical equidistant perturbation sampling rule sets the number of sampling points to 50, the number of sampling layers to 10, and the number of points in each layer to 5. The perturbation value is limited to equidistant points between the lower and upper boundaries of the interval, and a random offset is added to each layer. The offset value ranges from 1% to 3% of the interval span. The difference in lifespan reduction before and after the perturbation is calculated. The lifespan reduction is taken from the mileage corresponding to the first time the chain health index extrapolation sequence falls below the lower boundary of the shutdown threshold, and converted into the remaining mileage. The conversion ratio is set to 1, and the extrapolation step size is set to 1 kilometer. The difference is written to the unit row. The unit row key value is the unit number and time index. The state component name is written to the field name column, the perturbation value is written to the perturbation value column, and the difference result is written to the difference column. The difference results are summarized by state component field. The summary rule takes the average value, the absolute average value, and the maximum value of the difference of the same state component in the same unit and writes them into the summary column to generate a sensitivity table.

[0077] Based on the sensitivity table, a step-by-step multiplication-addition merging rule is used to multiply the sensitivity values ​​with the posterior mean item by item and accumulate them to obtain the marginal contribution sequence. The step-by-step multiplication-addition merging rule takes the posterior mean from the mean of the upper and lower bounds of the posterior interval set, aligns it by cell number and time index, fills missing items with the most recent previous value, and limits the filling steps to no more than 5. The sensitivity values ​​and posterior means are matched one-to-one according to the state component order, multiplied item by item, accumulated in the cell row, and written into the contribution column. Aggregation summation is then performed according to coupling, gearbox, axle box, and traction motor, and the results are sorted in descending order. The component category field is retrieved by the aggregation key, and the component category mapping table is fixed as follows: for couplings, the row containing rotor imbalance and torque-related rows are used; for gearboxes, the row containing tooth backlash drift is used; for axle boxes, the row containing radial clearance is used; and for traction motors, the row containing current fluctuation is used. The extrapolation sequence is compared point by point with the lower threshold and the crossing mileage points are recorded. The extrapolation sequence is the chain health index extrapolation sequence, and the comparison step size is set to 1 kilometer. The crossing judgment is changed from above the lower threshold to below the lower threshold. Only the earliest crossing is retained among multiple crossings, and the non-crossing is recorded as a null value, thus obtaining the trigger list.

[0078] Crossing mileage points are specifically defined as positions on the mileage axis where the extrapolated sequence changes from above to below the lower threshold. The extrapolated sequence is discretized into adjacent mileage points and their corresponding values ​​using a fixed mileage step size. When a mileage point's value is above the lower threshold and the next mileage point's value is below the lower threshold, or when a mileage point's value is equal to the lower threshold and the next mileage point's value is below the lower threshold, it is determined that there is a crossing mileage point between two adjacent mileage points. The crossing mileage point is determined by the proportion of the change in value between the mileage of the two adjacent mileage points. When there are multiple crossings, the first crossing mileage point is recorded, and subsequent crossing mileage points are not recorded. When no crossing occurs, the crossing mileage point is recorded as a null value.

[0079] The specific steps for generating the working condition action index table are as follows:

[0080] Based on the trigger list, the slope level, curve radius level, and station distance level are concatenated into the working condition index key. The traction power gear, regenerative braking ratio gear, and torque limiting gear are enumerated according to the working condition index key and written into the row to generate a working condition action candidate table.

[0081] Based on the candidate table of working conditions and actions, the energy consumption value per unit distance is calculated by converting the speed curve and traction power. The action score is obtained by adding the slope penalty term of the multi-unit degraded posterior interval set to the energy consumption value. The out-of-bounds action rows are deleted and the entries are output according to the working condition index key to obtain the working condition action index table.

[0082] Based on the trigger list, the slope grade, curve radius grade, and station distance grade are concatenated into a working condition index key using the working condition index coding rule. The working condition index coding rule sets the slope grade values ​​to 1 to 4, with intervals limited to 0 to 5, 6 to 15, 16 to 25, and 26 to 40; the curve radius grade values ​​to 1 to 4, with intervals limited to 200 to 400, 401 to 800, 801 to 1500, and 1501 to 5000; and the station distance grade values ​​to 1 to 4, with intervals limited to 0.5 to 1.5, 1.6 to 3.0, 3.1 to 5.0, and 5.1 to 10.0. The working condition index key is formatted as follows: 2 digits for slope grade, 2 digits for curve radius grade... The two-digit number and the two-digit number of the inter-station distance level are concatenated and padded with zeros. The traction power level, regenerative braking ratio level, and torque limiting level are enumerated according to the working condition index key using the Cartesian enumeration rule and written into the row. The traction power level values ​​are 0, 20, 40, 60, 80, and 100 and written into the field as a percentage. The regenerative braking ratio level values ​​are 0, 25, 50, 75, and 100 and written into the field as a percentage. The torque limiting level values ​​are 40, 55, 70, 85, and 100 and written into the field as a percentage. All level combinations are written into each working condition index key and written into the crossing mileage point field and unit priority field in the trigger list to generate a working condition action candidate table.

[0083] Based on the candidate table of working conditions and actions, a discrete energy consumption conversion rule is used to convert the speed curve and traction power into energy consumption per unit distance. The discrete energy consumption conversion rule sets the time step to 0.1 and takes the speed curve as the speed value for each time step, converting the speed unit to meters per second. The traction power is taken as the gear percentage multiplied by the rated power, with the rated power set to 300 and written in kilowatts. The energy increment is calculated by time step and written into the energy column, with the energy unit converted to kilowatt-hours. The energy increment is accumulated by the inter-station distance to obtain the interval energy consumption, and the interval energy consumption is divided by the inter-station distance to obtain the energy consumption per unit distance. A slope penalty superposition rule is used to add the slope penalty of the multi-unit degenerate posterior interval set to the energy consumption value to obtain the action score. The slope penalty superposition rule sets the slope... The rate value is derived from the extrapolated slope of the chain health index of the multi-unit degradation posterior interval and the slope with a quantile of 0.9 is taken. The slope weight is set to 0.7 and the energy consumption weight is set to 0.3 and written into the weight column. The action score is written into the score column by multiplying the slope penalty by the slope weight and the energy consumption value per unit distance by the energy consumption weight. The out-of-bounds elimination rule is used to delete out-of-bounds action rows and output entries according to the working condition index key. The out-of-bounds elimination rule sets that the traction power level must not exceed 100 and must not be less than 0, the regenerative braking ratio level must not exceed 100 and must not be less than 0, and the torque limit level must not exceed 100 and must not be less than 40. The action score is deleted for empty value rows, and the remaining rows are sorted in ascending order of score and the first 3 rows are written into the index entries to obtain the working condition action index table.

[0084] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit, characterized in that, Includes the following steps: S1: Based on the full-chain unit list of rotating equipment in rail transit, the axle box bearings, gear pairs, traction motor rotors and brake discs are addressed and entered into the table. The state components are paired with the observations item by item and the symbolic constraints are solidified to establish a constraint set. S2: Based on the constraint set, organize the key performance index sequence along the mileage and cycle number, calculate the rate of change of adjacent points to summarize the decay rate parameter, label the degradation phase number according to the rate of change segmentation rule, use a Markov chain, convert the stage labeling to the transfer ratio and derive the stage proportion sequence, fuse the lifetime occupancy ratio and stage penalty coefficient and summarize to obtain the health index trajectory. S3: Based on the health index trajectory, update the state of multiple units by time step and calculate the residual sequence. Perform consistent direction, same increment, phase difference threshold, power consumption boundary pruning and write-back item by item. Use Kalman filtering and select the update path according to the residual mean square comparison result to obtain the posterior interval set. S4: Based on the posterior interval set, the lifetime difference of the perturbation state is used to generate a sensitivity table. The marginal contribution sequence is obtained by multiplying and adding the posterior mean. The couplings, gearboxes, axle boxes, and traction motors are merged and sorted, and the crossing mileage points are marked to obtain the trigger list. S5: Based on the trigger list, select power, regeneration, and limiting range according to slope, curve, and station distance, calculate energy consumption, add multi-unit degradation posterior interval set slope penalty term, remove out-of-bounds terms, and obtain the working condition action index table.

2. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 1, characterized in that, The constraint set includes a set of unit identifiers, a set of state variable names, a set of observation names, a set of symbolic constraints, and a set of load level segments. The health index trajectory includes a chain health index sequence, a unit health index sequence, a degradation phase number sequence, a stage proportion sequence, and a remaining life proportion sequence. The posterior interval includes a rotor imbalance interval, a bearing radial clearance interval, a tooth flank clearance drift interval, a braking friction attenuation interval, and a set of confidence boundaries. The trigger list includes a set of crossing mileage points, a set of warning thresholds, a set of shutdown thresholds, a set of unit contribution rankings, and a set of unit priorities. The action index includes entries for slope level, curve radius level, inter-station distance level, traction power gear, regenerative braking ratio gear, and torque limiting gear.

3. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 1, characterized in that, The specific steps for generating the constraint set are as follows: Based on the full-chain unit list of rotating equipment in rail transit, the axle box bearings, gear pairs, traction motor rotors and brake discs are addressed and entered into the table. State component fields are written to each unit and matched with the observation fields item by item. Symbolic constraint values ​​are written according to the pairing relationship to generate a pairing constraint table. Based on the pairing constraint table, a one-to-one correspondence check is performed on the state component field and the observation field, rows with symbolic constraint conflicts are deleted and missing rows are filled in, load level segment boundary values ​​are written and segment intervals are locked, and a constraint set is established.

4. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 1, characterized in that, The specific steps for generating the health index trajectory are as follows: Based on the constraint set, the key performance indicator sequence is organized along the mileage and cycle number, grouped by unit number and sorted by time index, missing points and reverse points are deleted, continuous intervals are locked and written into the interval start and end indexes, and a performance sequence table is generated. Based on the performance sequence table, the rate of change sequence is generated by the difference between adjacent points, the rate of change is aggregated according to the window length and the window statistics are output, the abrupt windows are removed according to the abnormal threshold and the continuous windows are retained, the decay rate parameters are summarized and written into the cell row, and a degradation labeling table is generated. Based on the degradation label table, a Markov chain is used to calculate the transfer ratio from the stage label count and derive the stage proportion sequence. The lifespan occupancy ratio is converted into the lifespan remaining ratio. The stage penalty coefficient and the lifespan remaining ratio are added according to weight and written into the chain summary row to obtain the health index trajectory.

5. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 4, characterized in that, The Markov chain first limits the stage set to normal, mildly degraded, moderately degraded, and severely degraded. It reads the stage label sequence in order of mileage and cycle number, counts the occurrence of stage pairs between adjacent time points and writes them into the transition count matrix. It normalizes the transition count matrix row by row to obtain the transition ratio matrix. It performs matrix multiplication between the current time point stage ratio vector and the transition ratio matrix and iterates and updates it according to the prediction step number. It outputs the stage ratio sequence at each prediction time point. It converts the stage ratio sequence into a stage penalty coefficient sequence according to a preset mapping table and writes it into the chain summary row.

6. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 1, characterized in that, The specific steps for generating the posterior interval set are as follows: Based on the health index trajectory, the equivalent unbalance of the rotor, the equivalent radial clearance of the bearing, the tooth flank clearance drift, and the braking friction attenuation are written according to the time step. The traction current fluctuation, shaft end vibration energy, meshing impact index, and braking power consumption index are aligned according to the unit number. The observation difference is calculated and written into the residual row to generate the state residual table. Based on the state residual table, the direction of the unbalance increment is determined, the radial clearance increment is determined, the phase difference of the tooth flank clearance drift and impact index is determined, the boundary of friction decay and power consumption is determined, the out-of-bounds items are trimmed and written back and the trimming flag is marked, and a constraint state table is generated. Based on the constraint state table, Kalman filtering is used to calculate the residual mean square at each time step and compare it with the threshold. Above the threshold, multiple samples are written and quantiles are taken to form upper and lower bounds. Below the threshold, single-point updates are written and the upper and lower bounds are expanded according to the cumulative deviation. The upper and lower bound rows are summarized by unit number to obtain the posterior interval set.

7. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 6, characterized in that, The Kalman filter first reads the state estimate, covariance matrix, observation vector, observation matrix, state transition matrix, process noise covariance matrix, and observation noise covariance matrix from the constraint state table. It then calculates the predicted state value and the predicted covariance matrix, the observation residual vector and the residual covariance matrix, and the Kalman gain matrix. The predicted state value is multiplied by the Kalman gain matrix and the observation residual vector, and the updated state value is obtained. The Kalman gain matrix is ​​combined with the observation matrix and the predicted covariance matrix to obtain the updated covariance matrix. The mean square of the difference between the updated state value and the observation vector is calculated to obtain the residual mean square, which is compared with a threshold. Based on the comparison result, the updated state value and a multi-point sample set are selected. The upper and lower bounds of the state components are calculated according to the updated covariance matrix, and the results are summarized by unit number to form a posterior interval set.

8. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 1, characterized in that, The specific steps for generating the trigger list are as follows: Based on the posterior interval set, a perturbation value is taken between the upper and lower bounds of each state component interval, the difference in lifetime reduction before and after the perturbation is calculated and written into the cell row, the difference results are summarized according to the state component fields, and a sensitivity table is generated. Based on the sensitivity table, the sensitivity values ​​are multiplied by the posterior mean item by item and summed to obtain the marginal contribution sequence. The sequence is aggregated and summed according to coupling, gearbox, axle box, and traction motor and arranged in descending order. The extrapolated sequence is compared with the lower threshold point by point and the crossing mileage points are recorded to obtain the trigger list.

9. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 8, wherein the crossing mileage point is specifically the position of the extrapolation sequence changing from above the lower threshold to below the lower threshold on the mileage axis. The extrapolation sequence is discretized into adjacent mileage points and corresponding values ​​according to a fixed mileage step. When the value of a certain mileage point is above the lower threshold and the value of the next mileage point is below the lower threshold, and when the value of a certain mileage point is equal to the lower threshold and the value of the next mileage point is below the lower threshold, it is determined that there is a crossing mileage point between two adjacent mileage points. The crossing mileage point is determined by the ratio of the change in value between the mileage of the two adjacent mileage points. When there are multiple crossings, the first crossing mileage point is recorded, and subsequent crossing mileage points are not recorded. When no crossing occurs, the crossing mileage point is recorded as a null value.

10. The method for whole-chain health prediction and energy efficiency optimization of rotating equipment in rail transit according to claim 1, characterized in that, The specific steps for generating the working condition action index table are as follows: Based on the trigger list, the slope level, curve radius level, and station distance level are concatenated into a working condition index key. The traction power gear, regenerative braking ratio gear, and torque limiting gear are enumerated according to the working condition index key and written into the row to generate a working condition action candidate table. Based on the aforementioned candidate table of working conditions and actions, the energy consumption per unit distance is calculated using the speed curve and traction power. The action score is obtained by adding the slope penalty term of the multi-unit degraded posterior interval set to the energy consumption value. Out-of-bounds action rows are deleted, and entries are output according to the working condition index key to obtain the working condition action index table.

Images