Urban rail equipment health degree intelligent diagnosis method and system based on cloud platform, and medium

Abstract

The invention relates to the technical field of rail transit, and discloses an intelligent diagnosis method and system for the health degree of urban rail equipment based on a cloud platform and a medium, the intelligent diagnosis method for the health degree of the urban rail equipment based on the cloud platform comprises the following steps: S1, multi-source data collection and edge preprocessing: collecting operation data of the urban rail equipment through a sensor, performing time sequence alignment and abnormal value filtering on the data to generate preprocessed data; s2, triggering type data uploading: based on the preprocessed data, judging whether an uploading condition is met or not through a statistical rule or a model prediction result, and uploading the data meeting the condition to a cloud platform; and S3, cloud model training and health degree prediction: constructing an equipment health state prediction model through a transfer learning algorithm by using the uploaded data. According to the invention, through deep learning modeling, edge calculation deployment and a multi-factor early warning mechanism, efficient evaluation, stable early warning and low-delay response of the equipment state are realized.

Description

Technical Field

[0001] The present invention relates to the field of rail transit technology, and in particular to a cloud platform-based intelligent diagnosis method, system and medium for the health of urban rail equipment. Background Art

[0002] In urban rail transit systems, there are many types of equipment and a complex operating environment. Once a key subsystem fails, it can affect the operation of a single line or even cause a large-scale shutdown accident. Therefore, identifying equipment health risks in advance and providing dynamic warnings are the core links to ensure the stable operation of the entire network.

[0003] At present, some intelligent systems have been applied to equipment status perception tasks. This type of technical solution usually deploys diagnostic engines centrally based on cloud platforms to classify and model the collected operating data. The system can output health score results with a certain accuracy to assist in judging potential risks of equipment, and it does play a positive role in improving the efficiency of manual judgment and reducing human misjudgment. At the same time, by mining historical data to establish a prediction model, it has a certain effect in discovering long-term degradation trends, especially in areas where known failure modes are relatively stable, and the system shows a certain degree of reliability.

[0004] Although existing technologies do play a positive role in improving the efficiency of manual judgment and reducing human misjudgment, they still have some shortcomings. First, existing technologies as a whole rely on unified modeling and processing on the cloud, lack real-time judgment capabilities on the edge, and often encounter problems such as "untimely processing of data backlogs" when faced with concurrent streams of multi-site and heterogeneous data. Some technical scoring models have poor generalization, and once the working conditions suddenly change, the results are easily distorted. In addition, existing technologies generally understand health scores at a single static value, ignoring the significance of trend changes and score fluctuations, which can easily lead to warnings being triggered only after the status has obviously deteriorated. Finally, some technologies are relatively mechanical in the design of alarm strategies, only setting fixed thresholds, and there is no level buffer mechanism. Once equipment is close to the boundary, it is frequently upgraded or downgraded, which can easily mislead operation and maintenance personnel and cause confusion in resource scheduling. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a cloud platform-based intelligent diagnosis method, system and medium for the health of urban rail equipment, which solves the problems in the prior art of non-real-time equipment health monitoring, inaccurate prediction and unstable early warning mechanism.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent diagnosis method for the health of urban rail equipment based on a cloud platform, comprising the following steps: S1. Multi-source data collection and edge preprocessing: collect the operation data of urban rail equipment through sensors, perform time series alignment and outlier filtering on the data, and generate preprocessed data; S2. Triggered data upload: Based on the preprocessed data, determine whether the upload conditions are met through statistical rules or model prediction results, and upload the data that meets the conditions to the cloud platform; S3. Cloud model training and health prediction: Utilize the uploaded data to construct a device health status prediction model through transfer learning algorithms, and dynamically update the model parameters based on federated learning to adapt to changes in the device status; S4. Dynamic health scoring and decision-making: Dynamically calculate the health score based on the output results of the health status prediction model, combined with the device operating conditions and environmental parameters, and generate operation and maintenance decision instructions; The steps S1 to S4 achieve closed-loop control through the collaborative interaction between the cloud platform and the edge computing nodes.

[0007] Preferably, the multi-source data collection and edge preprocessing include: Collect the vibration signals of the door guide rail through a three-axis vibration sensor; Collect the temperature data of the traction motor winding through a PT100 temperature sensor; Collect the current data of the drive circuit through a Hall current sensor; Align the time-series data of vibration, temperature, and current signals through the dynamic time warping algorithm; Filter out outliers from the data based on the sliding window statistical rule.

[0008] Preferably, the multi-source data collection and edge preprocessing further include: Generate synthetic vibration signals through a generative adversarial network to expand the training data; The generative adversarial network includes the adversarial training process of the generator and the discriminator; Collect environmental data through a temperature and humidity sensor, and introduce rainfall and wind speed parameters as environmental factors.

[0009] Preferably, the upload conditions include: The data exceeds the statistical threshold range based on the historical operation data distribution, and the statistical threshold range is dynamically calculated through the mean and standard deviation within the sliding window; The abnormal probability output by the lightweight model deployed on the edge computing node exceeds the dynamically adjusted threshold issued by the cloud platform, and the dynamically adjusted threshold is configured according to the device type and operation period; The environmental parameters exceed the safe operation threshold defined by the device manufacturer, and the safe operation threshold is associated with the device model and operating condition parameters.

[0010] Preferably, the transfer learning algorithm includes: Freeze some network layers based on the pre-trained deep neural network model; Align the feature distributions of general device data and urban rail transit device data through an adversarial domain adaptation algorithm; Aggregate the local model parameters of multiple edge nodes based on the federated learning algorithm to generate a global model.

[0011] Preferably, the dynamic health score and decision-making include: Calculate the health score based on the remaining life prediction value and the real-time failure probability; Dynamically adjust the health score threshold according to the device operation period; Generate a maintenance work order when the health score is lower than the threshold.

[0012] Preferably, the dynamic health score and decision-making further include: Introduce the device historical maintenance record as a scoring weight adjustment item; Balance the failure risk and maintenance cost based on the multi-objective optimization algorithm.

[0013] Preferably, the collaborative interaction includes: The edge node performs data preprocessing and real-time anomaly detection, and the cloud platform performs model training and global decision-making; The edge node communicates with the cloud platform through the 5G time-sensitive network protocol with low latency; The cloud platform sends the model parameters to the edge node, and the edge node periodically reports the device status metadata.

[0014] The present invention also provides an intelligent diagnosis system for the health of urban rail transit devices based on a cloud platform, including: Data acquisition module: Deploy vibration sensors, temperature sensors and current sensors to collect device operation data; Edge computing module: Perform data cleaning, feature extraction and anomaly detection based on an embedded hardware platform; Cloud analysis module: Construct a health status prediction model through transfer learning and federated learning, and dynamically update the parameters; Decision feedback module: Generate a heat map of the health score and a maintenance work order, and push them to the mobile terminal; Collaborative interaction module: Realize task allocation, parameter synchronization and communication control between the edge node and the cloud platform; The system realizes real-time monitoring and dynamic optimization of the device health status through a closed-loop feedback mechanism.

[0015] The present invention further provides a storage medium, on which a computer program is stored, and when the computer program is executed, it runs the foregoing intelligent diagnosis method for the health of urban rail transit devices based on a cloud platform.

[0016] The present invention provides an intelligent diagnosis method, system and medium for the health of urban rail transit devices based on a cloud platform. It has the following beneficial effects: 1. The present invention adopts a technical solution that integrates a deep learning health modeling algorithm with an edge computing framework, achieving the technical effects of high-precision prediction of the health status of urban rail equipment and low-latency monitoring response. Compared with the existing solutions that rely on centralized modeling and have obvious operation bottlenecks, it solves the problems of poor response timeliness, weak model generalization, and lagging operation and maintenance instructions in multi-site concurrent monitoring.

[0017] 2. The present invention realizes the multi-dimensional dynamic determination of the risk state of equipment by constructing a warning decision-making mechanism that combines trend recognition, fluctuation discrimination, and hysteresis classification. The technical effects are manifested as smoother changes in equipment status, more stable warning levels, and a significant reduction in false alarm rates. Compared with the traditional method based on static thresholds or single-score judgment, it solves the problems of frequent triggering of alarms by critical fluctuations, drastic jumps in levels, and waste of operation and maintenance resources.

[0018] 3. The present invention introduces the optimization of the data processing link and the compression and deployment of the pre-trained scoring model. The health scoring and warning determination can be completed at the edge end, greatly compressing the length of the data backhaul and processing chain, achieving the effects of significantly improved data processing efficiency, reduced consumption of computing resources, and more stable system operation. Compared with the existing structures that rely heavily on cloud computing, have high transmission costs, and poor fault tolerance, it effectively overcomes the technical shortcomings of difficult on-site deployment and poor system robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is the method flow chart of the present invention; Figure 2 is the system architecture diagram of the present invention; Figure 3 is the storage medium control diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the specification of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Please refer to the attached Figure 1 , the embodiments of the present invention provide an intelligent diagnosis method, system, and medium for the health degree of urban rail equipment based on a cloud platform, including the following steps: S1. Multi-source data collection and edge preprocessing: Collect the operation data of urban rail equipment through sensors, perform time series alignment and outlier filtering on the data, and generate preprocessed data; As the starting link of the health diagnosis process, S1 mainly realizes the acquisition of real-time data on the operating status of the equipment and local preliminary data processing. This step is directly related to the effect and accuracy of subsequent data uploading, model training, and health assessment, and is therefore a fundamental support module in the entire technical solution.

[0022] Under normal circumstances, due to the wide variety of urban rail transit equipment, there are significant differences in the operating parameters, state characteristics, and environmental conditions of different equipment. Therefore, the use of multi-source data fusion can more comprehensively reflect the actual working status of the equipment. At the same time, to ensure the availability of the data, after data collection in this step, data synchronization processing and outlier removal are also required to further improve the efficiency of subsequent processing and the reliability of the results.

[0023] In some embodiments, to ensure the synchronization and consistency of multi-source heterogeneous data, the present invention also designs a specific data alignment mechanism, and combines certain data cleaning and enhancement means to solve problems such as data time differences, data noise, and data scarcity.

[0024] In this embodiment, multi-source data collection is achieved by deploying various types of sensors at key positions of urban rail transit equipment.

[0025] Specifically: Vibration signals during equipment operation can be collected through a three-axis vibration sensor to characterize the mechanical state of the equipment, and the vibration signal is denoted as .

[0026] As an option, a PT100 temperature sensor can be installed on the traction motor winding to detect the temperature change of the winding in real time, and the temperature signal is denoted as .

[0027] In a possible implementation, current change data of the drive circuit can be obtained through a Hall current sensor to reflect the electrical working state of the equipment, and the current signal is denoted as .

[0028] To reflect the impact of the environment on the equipment, a temperature and humidity sensor can also be configured to collect the ambient temperature and the ambient humidity , and at the same time, combined with a rain sensor and a wind speed sensor to obtain the rainfall and the wind speed , jointly constituting the environmental impact parameter set .

[0029] Due to problems such as asynchronous, response delay, or non-linear change in multi-source data collection, time series alignment is performed on all collected data in this embodiment. Specifically: In the process of aligning the device operation data, the dynamic time warping (DTW) algorithm is preferably adopted, which has been disclosed in the foregoing content and will not be elaborated here. It should be noted that this algorithm is applicable to the time alignment processing of non-equal-length, multi-time-delay, and multi-modal data to ensure that various types of data have corresponding relationships within the same time window.

[0030] After completing the time synchronization, this embodiment performs anomaly detection and outlier filtering on the original data to eliminate occasional abnormal interference.

[0031] In order to achieve efficient and reliable anomaly detection, this embodiment further adopts a weighted sliding window statistical rule to identify outliers in the data. The specific anomaly detection method is as follows: When the following conditions are met, it is determined as an outlier and filtered: ; Where: is the original signal to be processed, which can be data such as vibration, temperature, current, etc. in this embodiment; is the signal after outlier filtering; is the time point The weighted average value of the data within the sliding window near it is calculated as: ; is the weighted standard deviation within the corresponding sliding window, and the calculation method is: ; is the length of the sliding window, which is generally set according to the device characteristics, and the common value range is 20 to 100; is the weight of the The weight of the is the anomaly determination threshold coefficient, and the common value range is 2 to 4. The specific value is adjusted according to the actual operation characteristics of the device;

[0032] This method can effectively handle abnormal phenomena such as mutations and short-time pulses in the signal, and further improve the data quality.

[0033] In another possible implementation, in order to solve the problem of insufficient operation data of some devices, this embodiment can also perform data augmentation on the vibration signal and use a generative adversarial network (GAN) to synthesize and expand the vibration signal. The overall principle, generation and discrimination mechanism of GAN have been disclosed in the foregoing content and will not be elaborated here.

[0034] It should be noted that in specific implementation, a generator and a discriminator with a convolutional structure can be adopted to better capture the time-domain and frequency-domain features of vibration signals. Meanwhile, Fourier transform or energy spectrum detection can be performed on the generated data to ensure the physical rationality and feature authenticity of the generated data.

[0035] In addition, in some embodiments, in order to further improve the processing ability at the edge, the present invention can design a dynamic configuration mechanism in the edge computing module to automatically adjust according to different device types and different operating conditions: The sampling rate of the collected signal; The length of the sliding window ; The anomaly detection threshold ; The number of training epochs and batch size of the GAN model; The start / stop and parameter configuration of different sensors.

[0036] S1 can implement a more flexible and intelligent data acquisition and preprocessing process, ensuring that the data after this step meets the technical requirements of high quality, high consistency, and high availability, and providing a reliable data basis for subsequent steps such as trigger-based data upload, cloud modeling, and health prediction.

[0037] S2. Trigger-based data upload: Based on the preprocessed data, determine whether the upload conditions are met through statistical rules or model prediction results, and upload the data that meets the conditions to the cloud platform; Step S2 is a key link to further realize the data interaction control between the edge node and the cloud platform on the basis of multi-source data acquisition and edge preprocessing in step S1. Generally, due to the complex environment, large number of devices, and diverse data types at the operation site of urban rail transit equipment, if all device data are uploaded to the cloud platform in real time, it will not only easily cause waste of network bandwidth resources, but also may increase data processing latency. Therefore, the core of the implementation of step S2 lies in designing a reasonable and targeted trigger-based data upload mechanism to achieve optimized control and intelligent scheduling of data upload.

[0038] As an option, in the design of the data upload timing and conditions in this step, not only the abnormal features of the data itself are considered, but also the prediction results of the edge model and the operating environment features of the device are comprehensively considered, and decisions on data upload control are made based on multi-dimensional information.

[0039] In a possible implementation manner, the implementation logic of the trigger-based data upload mechanism mainly includes anomaly detection trigger based on statistical features, trigger based on edge model prediction results, and trigger based on environmental parameter threshold judgment, and multiple trigger conditions can be used independently or in combination to improve the rationality and accuracy of data upload.

[0040] In this embodiment, the first type of trigger condition is a trigger mechanism based on a statistical threshold.

[0041] Specifically, for the key device operation data preprocessed in step S1 in the edge node, the mean and standard deviation within a sliding window are used for dynamic threshold judgment. If the current monitoring data significantly exceeds the historical statistical feature range, it is regarded as a potential anomaly and needs to be uploaded.

[0042] This judgment rule can be expressed as: ; Where: represents the monitoring data value at the current time point which can be the vibration signal, temperature signal or current signal of the device; represents the sliding weighted mean centered at the time point with a window length of as described in step S1; represents the weighted standard deviation of the data within the corresponding sliding window, as described in step S1; represents the dynamic threshold adjustment coefficient, which is used to control the sensitivity of anomaly triggering, and the specific value is adjusted according to the device characteristics, generally in the range of 1.5 to 4.

[0043] As an option, the sliding window length can be dynamically adjusted according to the stability of the device operation. Generally, the more stable the device operation, the longer the window length can be set, and vice versa, it can be appropriately shortened.

[0044] In some embodiments, different threshold parameters can be set for different types of signals respectively. For example, the anomaly threshold for the vibration signal is relatively high, while the anomaly threshold for the current signal is relatively low to adapt to the fluctuation characteristics of various signals.

[0045] In this embodiment, the second type of trigger condition is a trigger mechanism based on the prediction result of the edge model.

[0046] Specifically, for the key device operation data preprocessed in step S1, a lightweight anomaly detection model can be integrated inside the edge node to output the anomaly probability of the current device state in real time. If the anomaly probability predicted by the model exceeds the dynamic threshold configured or issued by the cloud platform, data upload can be triggered.

[0047] The judgment condition can be expressed as: ; Where: represents the edge anomaly detection model at the current time point for the input data The predicted abnormal probability value; Indicates the model prediction abnormal probability threshold dynamically configured or issued by the cloud platform, which can be dynamically adjusted according to factors such as device type, health score, historical abnormal data, etc.

[0048] As an option, the edge model can adopt algorithms such as logistic regression, lightweight neural network, and anomaly detection decision tree to balance the prediction effect and computational efficiency.

[0049] Generally, in order to avoid frequent uploads caused by short-term sudden abnormal probability jitters, Smoothing processing or confidence interval evaluation within a short time window can be performed to improve the stability of judgment.

[0050] In this embodiment, the third type of trigger condition is a trigger mechanism based on environmental parameter overrun.

[0051] Specifically, during the operation of the device, if the external environmental parameters of the device exceed the safe operation threshold defined by the device manufacturer, the upload of relevant environmental data and device operation data can be triggered.

[0052] This judgment condition can be expressed as: ; Where: Represents a certain environmental parameter value monitored at time point It can be temperature, humidity, rainfall, or wind speed, etc.; Represents the maximum safe operation threshold allowed for this environmental parameter, which is generally determined according to the design standards of the device manufacturer or industry specifications.

[0053] As an option, different types of environmental parameters can be configured with different safe operation thresholds. For example, the threshold of the temperature parameter may be between 45°C and 55°C, the humidity threshold may be between 85%RH and 95%RH, and the rainfall and wind speed thresholds can be adjusted according to the geographical environment where the device is located.

[0054] In a possible implementation, when multiple environmental parameters are simultaneously close to the threshold and the health score of the device is low, high-priority data upload can be directly triggered to ensure the safety of the device.

[0055] In some embodiments, in order to improve the intelligence and flexibility of the trigger mechanism, the present invention can also design a combined trigger strategy to combine and judge the above three types of trigger conditions to improve the accuracy of data upload.

[0056] As an option, data upload can be triggered when any one type of condition is met. In another possible implementation, the conditional combination logic can be configured according to the actual scenario, such as uploading only when two or three types of conditions are met, further reducing the upload frequency.

[0057] Generally, in order to achieve more efficient data upload, in this embodiment, metadata packets can be attached during the data upload process, specifically including: The category of conditions triggering the upload (statistical threshold, model prediction, or environmental overrun); The unique identification information of the device; All the original time-series data at the time of triggering the upload; The environmental parameter snapshot at the time of triggering the upload; The probability value of anomaly prediction and related model output results; The data upload timestamp and the location of the edge node.

[0058] Through metadata upload, after receiving the data, the cloud platform can quickly restore the device status and locate problems, providing data support for subsequent model training, health analysis, and operation and maintenance decision-making.

[0059] S3. Cloud model training and health prediction: Using the uploaded data, construct a device health status prediction model through transfer learning algorithms, and dynamically update the model parameters based on federated learning to adapt to changes in the device status; In step S2, the edge node completed three types of trigger judgments based on statistical features, model prediction results, and environmental parameters, and uploaded the data that met the conditions to the cloud platform. This step is based on this, using the powerful computing and analysis capabilities of the cloud platform to deeply process and analyze the uploaded data, so as to identify whether the device is currently in an abnormal state and further quantify its health level.

[0060] Generally, the data triggering the upload contains signals in a local time period of abnormal device operation. Such signals often show strong non-linear fluctuations, pattern mutations, or mixed operating condition responses in a short time. In order to accurately identify its abnormal type and severity, it is necessary to construct a discriminant model with adaptive feature extraction capabilities and sensitive to sequence data structures, and achieve a robust assessment of the health level through multi-dimensional embedding, metric learning, and confidence mechanisms.

[0061] As an option, the cloud can deploy an anomaly classification network based on deep representation learning, combined with weak supervision annotation or historical label databases to identify anomaly categories, improving the model interpretability and controllability of device operation and maintenance.

[0062] In this embodiment, to handle the non-stationarity and modal drift problems of the uploaded signals, the system preferentially processes the time-series data Perform time-frequency feature enhancement processing.

[0063] Specifically, the short-time Fourier transform is used to map the signal from the time domain to the time-frequency joint domain, and a two-dimensional feature tensor is constructed: ; Where: is the input one-dimensional time series signal, and the value at the time point ; is the window function, and the center position is , usually a Hamming window or a Gaussian window; is the frequency variable, with the unit of Hz; is the sliding position of the time window; is the time-frequency spectrum matrix of the transformed signal and serves as the input for the subsequent model.

[0064] This operation helps to extract the local time-domain excitation responses under different frequency bands and can provide a richer feature representation for the subsequent model.

[0065] After the feature extraction module, the system models the state difference between the uploaded sample and the historical normal samples based on the sparse embedding and distance metric strategy, and then realizes the quantitative recognition of abnormal events.

[0066] In this embodiment, the following abnormal scoring function is defined to measure the degree of deviation of the sample from the normal state: ; Where: is the abnormal score of the uploaded sample , and the larger the value, the more it deviates from the normal state; is the feature embedding representation of the input sample obtained through the deep coding network; is the center point (mean vector) of all historical normal samples in the embedding space; is the L2 norm, which is used to calculate the Euclidean distance.

[0067] Generally, when the abnormal score exceeds the empirical threshold, the system determines it as an abnormal event and attempts to perform class mapping and alarm marking.

[0068] To further enhance the reliability of the model, the system introduces an adaptive confidence estimation mechanism. This mechanism dynamically adjusts the confidence level of abnormal judgment according to the neighborhood distribution density of the sample in the feature space.

[0069] In a possible implementation, the following local confidence scoring function is defined: ; Where: is the local confidence score of the sample, and the larger the value, the more reliable the discrimination; represents the historical samples that are the closest to the sample in the embedding space; represents the deep embedding vector of the sample; is the number of neighbors, generally taking values between 5 and 30, and is automatically selected according to the sample density.

[0070] This mechanism helps to enhance the stability of the model judgment at the fuzzy boundary of the data, avoiding misidentification or missed reports.

[0071] After the abnormal event recognition is completed, the system further quantitatively evaluates the health of the current device.

[0072] In this embodiment, the health score function is designed based on the normalized abnormal distance and the weight penalty mechanism, and the specific expression is as follows: ; Where: is the health score of the device at the time point ; is the abnormal score of the current sample, from the aforementioned formula; is the reference distance value, taking the upper quartile of the maximum abnormal score in the historical normal state; is the health decay coefficient, used to adjust the score decay speed; is a very small positive number, used to prevent the denominator from being 0.

[0073] This function ensures that: when the sample is slightly abnormal, the health score drops slowly, and when there is a significant deviation, it drops rapidly, meeting the engineering safety requirements.

[0074] In some embodiments, to improve the robustness of the model, the system can also introduce a working condition sensitive adjustment mechanism for the device dimension.

[0075] Specifically, for some devices that exhibit "non-abnormal fluctuations" when working in high-load, high-temperature, or high-humidity environments, the system introduces a working condition label vector to perform secondary adjustment on the health result: ; Where: is the final health score; is the working condition correction factor, obtained through table lookup or an MLP network according to the represented operating state; is the current working condition identification vector, such as a combined label like "high load + night + high humidity".

[0076] ​This mechanism can effectively reduce the interference of environmental factors on the health score and improve the practicality and credibility of the scoring system.

[0077] S4. Dynamic health score and decision-making: According to the output results of the health status prediction model, combined with the operating conditions of the equipment and environmental parameters, dynamically calculate the health score and generate operation and maintenance decision instructions; In step S3, the system has completed the quantification of the equipment's health at the current moment, including the mapping mechanism based on the anomaly score, the confidence adjustment strategy, and the health score compensation under the working conditions. This health value provides a quantitative basis for subsequent operation and maintenance scheduling and risk control.

[0078] Generally, in the urban rail transit system, the equipment is widely distributed and the working environment varies greatly. Relying solely on the single-point health score to judge its operation risk has certain limitations. Especially when the equipment is in a marginal state or the score fluctuates near the critical value, without a perfect trend evaluation mechanism and early warning buffer strategy, it may lead to false alarms, missed alarms, or even mis-triggered maintenance.

[0079] Therefore, the main task of this step is to build a multi-dimensional, traceable, and forward-looking early warning decision-making mechanism based on the health score, and realize the automatic classification of the equipment status through hierarchical classification, providing a basis for subsequent maintenance decision-making and resource allocation.

[0080] In this embodiment, the system first performs trend modeling on the time series of the equipment health degree to depict the change rate and fluctuation form of the health evolution trajectory.

[0081] Specifically, for the health score sequence of a certain equipment within the time window , …, , the system defines the average decay rate as follows: ; Where: is the average health degree decay rate at the current time point , with the unit of the score decrease value per time step; is the length of the trend analysis window (unit: time step or minute); is the health score at the time point ; is the score at the next moment; The above difference represents the health decay amount per step, and taking the average reflects the trend intensity.

[0082] As an option, can be dynamically set according to the equipment operation frequency. For example, equipment with low-frequency operation uses a longer window (such as 1 hour), and high-frequency equipment uses a shorter window (such as 15 minutes) to improve the adaptability of trend judgment.

[0083] Based on the identified trend, the system introduces a fluctuation term to construct a more robust early warning scoring function. To identify the instability of the device state, the local fluctuation factor is defined as follows: ; Where: is the standard deviation of the fluctuation of the health score at time point ; is the length of the time window; is the health score at the th historical time point; is the mean value of the health scores within the time period.

[0084] This indicator reflects the stability of the scores in the recent period, and the larger the value, the more unstable it is.

[0085] This fluctuation term is used to judge whether there are frequent jumps in the health scores and can be used as an auxiliary factor for early warning grading.

[0086] Combining the health score, the decay rate, and the fluctuation factor, the system defines the comprehensive early warning scoring function as follows: ; Where: is the comprehensive early warning score at time point , and the larger the numerical value, the closer it is to the high risk; is the current health score; is the current decay rate, from the above formula; is the standard deviation of the score fluctuation; , , are the weighting coefficients, satisfying + + = 1, which can be set through offline training or expert rules.

[0087] In some embodiments, to improve the prediction lead time, the system will perform short-term prediction on and use an AR model or an LSTM sequence network to identify in advance the devices that may trigger an early warning in the future.

[0088] In this embodiment, the system defines multiple levels of early warning grades based on the value of . Different from the traditional static threshold judgment mechanism, the early warning grade determination here can integrate the trend smoothing function and the jump threshold combination to avoid false alarms caused by frequent critical fluctuations.

[0089] To prevent the device score from frequently switching grades just when it crosses a certain grade critical point, the system introduces a hysteresis mechanism: When a device enters the level-II early warning from normal operation, its early warning score shall be greater than ; When it returns from level-II to normal, it shall be less than , and .

[0090] Such a hysteresis strategy can avoid unnecessary alarms caused by drastic jumps in scores near the boundary.

[0091] In practical applications, the system implements corresponding management measures for different early warning levels. For example: The severe level triggers the automatic generation of operation and maintenance work orders; The moderate level enters the operation and maintenance pre-arrangement plan; The mild level only records the status and enters the health observation pool.

[0092] In some embodiments, the average health score of devices in the same area or the same subsystem can also be combined to implement a regional coordination adjustment mechanism. That is, when multiple devices show a concentrated downward trend in health, the system can perform a regional risk assessment in advance and raise the early warning levels of individual devices to achieve overall safety redundancy.

[0093] Steps S1 to S4 achieve closed-loop control through the collaborative interaction between the cloud platform and the edge computing nodes.

[0094] The intelligent diagnosis system for the health of urban rail transit equipment based on the cloud platform described below can be referred to in correspondence with the intelligent diagnosis method for the health of urban rail transit equipment based on the cloud platform described above.

[0095] Please refer to the appendix Figure 2 , the present invention also provides an intelligent diagnosis system for the health of urban rail transit equipment based on the cloud platform, including: Data acquisition module: Deploy vibration sensors, temperature sensors and current sensors to collect equipment operation data; Edge computing module: Perform data cleaning, feature extraction and anomaly detection based on an embedded hardware platform; Cloud analysis module: Build a health status prediction model through transfer learning and federated learning and dynamically update parameters; Decision feedback module: Generate a heat map of health scores and maintenance work orders and push them to the mobile terminal; Collaborative interaction module: Achieve task allocation, parameter synchronization and communication control between the edge node and the cloud platform; The system realizes real-time monitoring and dynamic optimization of the equipment health status through a closed-loop feedback mechanism.

[0096] The system in this embodiment can be used to execute the above method embodiment, and its principle and technical effects are similar, which will not be elaborated here.

[0097] A storage medium described below can be correspondingly referred to in relation to the intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform described above.

[0098] Please refer to the appendix Figure 3 The present invention further provides a storage medium having a computer program stored thereon, and when the computer program is executed, the foregoing intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform is run.

[0099] The storage medium of this embodiment can be used to implement the above method embodiment, and the principles and technical effects are similar, and will not be elaborated here.

[0100] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform, characterized in that It includes the following steps: S1. Multi-source data collection and edge preprocessing: Collect the operation data of urban rail transit equipment through sensors, perform time series alignment and outlier filtering on the data, and generate preprocessed data; S2. Triggered data upload: Based on the preprocessed data, judge whether the upload condition is met through statistical rules or model prediction results, and upload the data that meets the conditions to the cloud platform; S3. Cloud model training and health prediction: Use the uploaded data to construct a device health status prediction model through transfer learning algorithms, and dynamically update model parameters based on federated learning to adapt to changes in device status; S4. Dynamic health score and decision-making: According to the output results of the health status prediction model, dynamically calculate the health score in combination with the device operation conditions and environmental parameters, and generate operation and maintenance decision instructions; The steps S1 to S4 achieve closed-loop control through the collaborative interaction between the cloud platform and the edge computing nodes.

2. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 1, wherein The multi-source data collection and edge preprocessing include: Collect the vibration signals of the door guide rail through a three-axis vibration sensor; Collect the temperature data of the traction motor winding through a PT100 temperature sensor; Collect the current data of the drive circuit through a Hall current sensor; Align the time series data of vibration, temperature, and current signals through the dynamic time warping algorithm; Filter outliers from the data based on the sliding window statistical rule.

3. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 2, characterized in that The multi-source data collection and edge preprocessing also include: Generate synthetic vibration signals through a generative adversarial network to expand the training data; The generative adversarial network includes the adversarial training process of the generator and the discriminator; Collect environmental data through a temperature and humidity sensor, and introduce rainfall and wind speed parameters as environmental factors.

4. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 1, wherein, The upload conditions include: The data exceeds the statistical threshold range based on the historical operation data distribution, and the statistical threshold range is dynamically calculated through the mean and standard deviation within the sliding window; The abnormal probability output by the lightweight model deployed on the edge computing node exceeds the dynamically adjusted threshold issued by the cloud platform, and the dynamically adjusted threshold is configured according to the device type and operation period; The environmental parameters exceed the safe operation threshold defined by the device manufacturer, and the safe operation threshold is associated with the device model and operating condition parameters.

5. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 1, wherein The transfer learning algorithms include: Freeze some network layers based on a pre-trained deep neural network model; Align the feature distributions of general device data and urban rail transit equipment data through an adversarial domain adaptation algorithm; Aggregate the local model parameters of multiple edge nodes based on the federated learning algorithm to generate a global model.

6. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 1, wherein The dynamic health score and decision-making include: Calculate the health score based on the remaining life prediction value and the real-time failure probability; Dynamically adjust the health score threshold according to the device operation period; Generate a maintenance work order when the health score is lower than the threshold.

7. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 1, characterized in that The dynamic health score and decision-making further include: Introduce the device historical maintenance record as a scoring weight adjustment item; Balance the failure risk and maintenance cost based on the multi-objective optimization algorithm.

8. The intelligent diagnosis method for the health of urban rail transit equipment based on a cloud platform according to claim 1, wherein The collaborative interaction includes: The edge node performs data preprocessing and real-time anomaly detection, and the cloud platform performs model training and global decision-making; The edge node communicates with the cloud platform through the 5G time-sensitive network protocol with low latency; The cloud platform sends model parameters to the edge nodes, and the edge nodes periodically report device status metadata.

9. An intelligent diagnosis system for the health of urban rail transit equipment based on a cloud platform, characterized in that, Using the intelligent diagnosis method for the health of urban rail transit equipment based on the cloud platform according to any one of claims 1-8, comprising: Data acquisition module: Deploy vibration sensors, temperature sensors and current sensors for collecting equipment operation data; Edge computing module: Perform data cleaning, feature extraction and anomaly detection based on an embedded hardware platform; Cloud analysis module: Construct a health status prediction model through transfer learning and federated learning and dynamically update the parameters; Decision feedback module: Generate a heat map of health score and maintenance work orders and push them to the mobile terminal; Collaborative interaction module: Realize task assignment, parameter synchronization and communication control between the edge nodes and the cloud platform; The system realizes real-time monitoring and dynamic optimization of the device health status through a closed-loop feedback mechanism.

10. A storage medium, on which a computer program is stored, characterized in that, When the computer program is executed, it runs the intelligent diagnosis method for the health of urban rail transit equipment based on the cloud platform according to any one of claims 1-8.

Images