Subway platform door system health degree evaluation and prediction method based on operation and maintenance data

Abstract

The invention discloses a subway platform door system health degree evaluation and prediction method based on operation and maintenance data, and relates to the technical field of intelligent operation and maintenance of urban rail transit equipment, and the method comprises the steps of health degree evaluation system construction, feature optimization, time sequence prediction model design and verification, and model and time sequence analysis driven by fusion data. And dynamic evaluation and accurate prediction of the health state of the subway platform door system are realized. And a data driving method is closely combined with actual operation and maintenance requirements, so that the whole process optimization from health state evaluation to dynamic prediction is realized. Specifically, according to the patent scheme, firstly, a scientific and reasonable health degree evaluation system is constructed through a multiple linear regression model, the model is based on 14 key feature variables of a subway platform door system, SPSS software verification shows that the multiple correlation coefficient reaches up to 0.964, the decision coefficient reaches 0.929, and it is indicated that the model can accurately quantify the equipment health state; and a solid foundation is laid for subsequent prediction.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for urban rail transit equipment, and in particular to a method for evaluating and predicting the health status of subway platform door systems based on operation and maintenance data. Background Technology

[0002] Equipment health assessment and prediction technology, as a core component of fault prediction and health management (PHM) systems, has made significant progress in recent years in fields such as industry and aviation. However, its application in subway equipment, especially platform screen door systems, is still in its early stages.

[0003] Equipment health assessment aims to quantify the degree of equipment degradation by constructing health indicators (HIs). Existing methods are mainly divided into physical health indicators (PHIs) and virtual health indicators (VHIs).

[0004] Physically based health indicators (PHIs) rely on monitoring data of the physical characteristics of equipment, and commonly use statistical methods and signal processing techniques to extract features. This method of evaluating equipment health has clear physical meaning, but it depends on an accurate physical model. When applied to complex equipment, model construction is difficult and its adaptability is poor.

[0005] Virtual Health Indicators (VHIs) are constructed by fusing multi-source sensor data using data mining algorithms, resulting in health indicators without physical meaning. This method of evaluating device health is more suitable for data-rich scenarios, but it relies on high-quality data and has relatively weak interpretability.

[0006] Equipment health prediction aims to predict the future state of equipment using historical data, driving a shift in maintenance strategies from "planned maintenance" to "intelligent maintenance." Existing methods can be categorized into three types: model-driven, data-driven, and hybrid methods.

[0007] Model-driven approaches simulate equipment degradation processes using physical or mathematical models, such as ARMA models for predicting the remaining life of elevator door systems. These methods are highly reliable but require accurate mechanistic models and are difficult to generalize to complex equipment such as subway platform screen door systems.

[0008] Data-driven approaches utilize machine learning algorithms to learn degradation patterns from historical device data. In recent years, attention mechanisms have been introduced to enhance the model's ability to capture key time steps. Furthermore, ensemble learning algorithms such as XGBoost and LightGBM have been used for feature selection and prediction.

[0009] Hybrid approach: combines the advantages of model-driven and data-driven approaches, offering high accuracy, but also high complexity and implementation cost.

[0010] Despite advancements in equipment health assessment and prediction technologies across various fields, their application in subway platform screen door systems still exhibits the following significant shortcomings and deficiencies: ① Limitations in Application: Existing research mainly focuses on aviation and industrial equipment, with little attention paid to subway-specific equipment. Subway platform screen door systems are characterized by high-frequency, high-load operation, and their health degradation patterns differ significantly from those of industrial equipment. However, existing methods lack targeted models, leading to insufficient prediction accuracy. For example, traditional methods for constructing health indicators do not consider the interactive effects of equipment modules, making it difficult to capture their dynamic degradation characteristics.

[0011] ② Over-reliance on human experience and poor generalization: Traditional health assessment methods are highly subjective and cannot automatically extract deep features. Furthermore, physical model methods require parameter adjustments for specific equipment, resulting in weak generalization ability and difficulty adapting to the differences in platform screen door systems across different subway lines.

[0012] ③ Low model complexity and computational efficiency: Existing deep learning models have complex structures, requiring a large amount of data and computational resources for training. For example, while LSTM models can handle long sequences, they have a large number of parameters and high training costs; attention mechanisms, while enhancing the handling of long sequence dependencies, further increase model complexity. In actual subway operation and maintenance, the amount of data is limited and real-time prediction is required, making it difficult for existing models to balance accuracy and efficiency.

[0013] ④ Insufficient data quality and feature processing: Existing methods assume high data quality, but subway platform screen door system operation and maintenance data often suffers from incompleteness and high noise levels. Feature selection relies heavily on static methods and lacks dynamic evaluation of the importance of time-series features, leading to a decline in model generalization ability.

[0014] ⑤ Insufficient prediction accuracy and robustness: Existing models have poor adaptability to mutation data and high false positive and false negative rates.

[0015] ⑥ Lack of practical integration and application verification: Some studies remain at the level of data simulation and are not combined with the actual deployment of equipment, resulting in low practicality of research results.

[0016] Subway platform screen door systems are critical equipment for ensuring operational safety. They operate under high-frequency, high-load conditions for extended periods, and their health status directly impacts passenger safety and train punctuality. Traditional maintenance methods rely on manual experience, making it difficult to monitor equipment degradation patterns in real time. Furthermore, existing health prediction methods are mostly designed for industrial equipment and lack adaptation to the specific scenarios of subways.

[0017] Therefore, a method for evaluating and predicting the health of subway platform door systems based on operation and maintenance data is provided to address the above issues. Summary of the Invention

[0018] To address the aforementioned challenges, this invention provides a method for health evaluation and prediction of subway platform screen door systems based on operational data. This method includes the construction of a health evaluation system, feature optimization, and the design and verification of a time-series prediction model. By integrating a data-driven model with time-series analysis, it achieves dynamic evaluation and accurate prediction of the health status of the subway platform screen door system. By closely combining the data-driven approach with actual operational needs, it achieves end-to-end optimization from health status evaluation to dynamic prediction. Specifically, the patented solution first constructs a scientifically sound health evaluation system using a multiple linear regression model. This model is based on 14 key characteristic variables of the subway platform screen door system. Validation using SPSS software shows a high multiple correlation coefficient of 0.964 and a coefficient of determination of 0.929, indicating that the model can accurately quantify the health status of the equipment, laying a solid foundation for subsequent predictions.

[0019] To achieve the above objectives, this invention provides a method for health evaluation and prediction of subway platform door systems based on operation and maintenance data, including a health evaluation module and a health prediction module. The health evaluation module quantifies the current health status of the equipment through multi-dimensional feature variables, and the health prediction module uses a fusion model to capture the temporal degradation pattern. Specifically, it includes the following steps: S1: Collect raw operation and maintenance data, and use multiple linear regression analysis to filter feature variables related to the target variable to obtain the filtered dataset; S2: Enhance the filtered dataset obtained in S1 to obtain an expanded dataset; S3: Use the XGBoost algorithm to filter important features in the expanded dataset obtained in S2 to form the experimental dataset; S4: Normalize the data in the experimental dataset formed in S3, and divide the data into the training set according to the first 80% of the time series data and the test set according to the last 20% of the time series data. S5: Introduce an attention mechanism into GRU, input the divided training set and test set into the GRU-Attention model to form the XGBoost-GRU-Attention model for health prediction; S6: After the XGBoost-GRU-Attention model completes the health prediction, it performs inverse normalization on the data, compares the predicted values ​​with the true values, and outputs the evaluation results of the model's prediction performance.

[0020] Preferably, the operation and maintenance data of the subway platform door system is collected in S1. Each set of operation and maintenance data includes 14 characteristic parameters U1~U14. The subway platform door system includes a safety circuit conversion device (ATC) module, a unit controller (PEDC) module, and a signal conversion circuit (SIG) module, where U1~U5 belong to ATC, U6~U11 belong to PEDC, and U12~U14 belong to SIG.

[0021] Preferably, in step S1, the characteristic variables U1~U14 and the target variable "health of subway platform screen door system" are input into the statistical software SPSS for multiple linear regression analysis. The regression model is as follows: ; In the formula, The target variable is "health of the subway platform door system"; As a characteristic variable; The intercept; These are the regression coefficients; This is the error term; The significance of the model was verified by goodness-of-fit test, F-test, and T-test, and the formula for calculating the health of the subway platform door system was obtained: ; enter The time-feature value output is the health status of the subway platform door system at that time.

[0022] Preferably, in S2, linear interpolation is used for data augmentation, and the augmented dataset is the expanded dataset. The formula for linear interpolation is as follows: ; In the formula, and There are two known data points, and ; The independent variable that needs to be interpolated; For the corresponding estimated value, and It consists of two moments, before and after. It is always equal to 1. and The feature parameters are respectively in and The value at time is calculated by interpolating each characteristic parameter using the above formula.

[0023] Preferably, step S3 uses the XGBoost algorithm to screen important features. The feature importance metric is the built-in gain. The gain is calculated by summing the reduction in loss caused by the feature at all split nodes to evaluate the importance of the feature. The calculation formula is as follows: ; In the formula, and These are the sums of the first derivatives of all samples in the left and right child nodes, respectively. and These are the sums of the second derivatives of all samples in the left and right child nodes, respectively. This is a regularization parameter that controls the magnitude of node weights. This is a regularization parameter that controls the cost of the splitting operation. The XGBoost algorithm constructs a tree structure by iteratively splitting nodes. XGBoost calculates the gain of all split nodes and selects the feature with the largest gain as the splitting feature of the current node. The splitting process employs a binary partitioning strategy, if Then Node binary partitioning Nodes and Node; if Then for The nodes are divided.

[0024] Preferably, in S4, data normalization reduces the data to the [0,1] interval, eliminating the impact of magnitude differences in the data on the prediction results, as shown in the following formula: ; In the formula, This is the result after data normalization; This is a sample data point for the health status of subway platform door systems. and These represent the maximum and minimum values ​​of the health status sample data for the subway platform door system, respectively.

[0025] Preferably, in S5, an attention mechanism is introduced into the GRU to form the XGBoost-GRU-Attention model for predicting the health status of subway platform door systems; in the structure of the baseline model GRU, express The input vector at time step; Indicates resetting the gate vector; This indicates updating the gate vector; Indicates the candidate hidden state; express Output vector at time step; This represents the activation function. , , , The calculation formula is as follows: ; ; ; ; In the baseline model GRU, an attention mechanism is introduced to capture the relationship between any two positions in the time series. The attention mechanism performs a weighted summation of the GRU network input vectors and calculates the attention score using the Softmax function. The output vector in the GRU-Attention model is... The calculation formula is as follows: ; In the formula, This is the output vector of the GRU-Attention model; Attention score; This is the output vector of the GRU network. The calculation formula is as follows: ; In the formula, and These are the weighting coefficients; This is the bias coefficient; We want to input the well-defined training and test sets into the GRU-Attention model to generate the XGBoost-GRU-Attention model for predicting the health status of subway platform door systems.

[0026] Preferably, in S6, the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) are selected. 2 To evaluate the predictive performance of the model; MSE and RMSE are used to measure the average deviation between model predictions and actual values; MAE is used to measure the absolute error between predictions and actual values; MAPE is used to measure the relative error between predictions and actual values; R... 2 Used to measure a model's ability to explain data variation; the specific formula is as follows: ; ; ; ; ; In the formula, This indicates the true health status of the subway platform door system; This represents the predicted health status of the subway platform door system. This represents the average of the true values.

[0027] Therefore, the above-mentioned method for evaluating and predicting the health status of subway platform door systems based on operation and maintenance data, adopted in this invention, has the following beneficial effects: (1) This invention closely integrates the data-driven approach with actual operation and maintenance needs, and realizes full-process optimization from health status assessment to dynamic prediction.

[0028] (2) This invention constructs a scientific and reasonable health evaluation system through a multiple linear regression model. The model is based on 14 key characteristic variables of the subway platform door system. The SPSS software verification shows that the multiple correlation coefficient is as high as 0.964 and the determination coefficient is 0.929, indicating that the model can accurately quantify the health status of the equipment and lay a solid foundation for subsequent prediction.

[0029] (3) In the health prediction stage, this invention proposes the XGBoost-GRU-Attention model, which improves prediction accuracy and efficiency through multi-stage processing. The XGBoost algorithm sorts the importance of feature variables by gain calculation, sets the gain threshold to 0.0300, and automatically selects 7 key features, effectively reducing the complexity of the model.

[0030] (4) In this invention, the GRU network combined with the attention mechanism further enhances the ability to process time-series data. GRU solves the gradient problem of traditional RNN by updating and resetting the gate structure, while the attention mechanism enhances the sensitivity to capturing health change points by dynamically weighting key time steps.

[0031] (5) The predictive performance of this invention is significantly better than that of traditional models. Ablation experiments show that the mean squared error of the GRU model is reduced by about 15% after introducing XGBoost feature selection, and the addition of the attention mechanism further reduces the mean absolute error by 10%. The final model determination coefficient R 2 It remains stable above 0.96.

[0032] (6) This invention is highly practical. The health level classification and maintenance strategy are linked, supporting the transformation from "planned maintenance" to "intelligent maintenance". In actual tests, it can reduce sudden downtime by about 30%.

[0033] (7) This invention can reduce operation and maintenance costs by accurately predicting and optimizing maintenance cycles, while improving equipment reliability and passenger safety.

[0034] (8) The model proposed in this invention has strong generalization ability, and the feature selection and time series modeling framework can be adapted to other rail transit equipment, providing a technical reference for the intelligent operation and maintenance of subway equipment.

[0035] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0036] Figure 1 This is a flowchart of the method for evaluating and predicting the health of subway platform door systems based on operation and maintenance data in this invention; Figure 2 This is an organizational structure diagram of the subway platform door system in an embodiment of the present invention; Figure 3 This is a schematic diagram of the XGBoost algorithm in an embodiment of the present invention; Figure 4 This is a schematic diagram of the basic structure of the GRU in an embodiment of the present invention; Figure 5 This is a schematic diagram of the GRU-Attention model structure in an embodiment of the present invention; Figure 6 This is a graph showing the feature importance ranking results based on the XGBoost algorithm in an embodiment of the present invention; Figure 7 This is a comparison chart of the prediction results of different models in the comparative example of this invention. Detailed Implementation

[0037] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0038] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0039] The terms "comprising" or "including" as used in this invention mean that the element preceding the term encompasses the element listed after the term, and do not exclude the possibility of encompassing other elements. Terms such as "inner," "outer," "upper," and "lower" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In this invention, unless otherwise explicitly specified and limited, the term "attached" and similar terms should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can refer to a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication of two elements or the interaction relationship between two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0040] Example like Figure 1 As shown, the method for health evaluation and prediction of subway platform door systems based on operation and maintenance data includes a health evaluation module and a health prediction module. The health evaluation module quantifies the current health status of the equipment through multi-dimensional feature variables, and the health prediction module uses a fusion model to capture the temporal degradation pattern. Specifically, it includes the following steps: S1: Collect raw operation and maintenance data, and use multiple linear regression analysis to filter feature variables related to the target variable to obtain the filtered dataset; S1 collects operation and maintenance data of the subway platform screen door system, with each set of operation and maintenance data containing 14 feature parameters U1~U14; the subway platform screen door system includes a safety circuit switching device (ATC) module, a unit controller (PEDC) module, and a signal conversion circuit (SIG) module, where U1~U5 belong to ATC, U6~U11 belong to PEDC, and U12~U14 belong to SIG. The organizational structure of the subway platform screen door system is as follows: Figure 2 As shown.

[0041] In step S1, the characteristic variables U1~U14 and the target variable "health of subway platform screen door system" are input into the statistical software SPSS for multiple linear regression analysis. The regression model is as follows: ; In the formula, The target variable is "health of the subway platform door system"; As a characteristic variable; The intercept; These are the regression coefficients; This is the error term; The significance of the model was verified by goodness-of-fit test, F-test, and T-test, and the formula for calculating the health of the subway platform door system was obtained: ; enter The time-feature value output is the health status of the subway platform door system at that time.

[0042] Regression analysis was conducted using SPSS software on 14 characteristic variables and the target variable "health status of subway platform screen door system". The results showed that all characteristic variables were significantly correlated with health status, with a high multiple correlation coefficient of 0.964 and a determination coefficient of 0.929, indicating that the system can accurately quantify the health status of equipment and explain 92.9% of the health status variation. The evaluation system further divided the health status into four levels (healthy, relatively healthy, unhealthy, and faulty) and formulated corresponding maintenance strategies to achieve classified management of health status. Each level corresponds to different maintenance actions; for example, emergency repairs are triggered when the health status is below 60%, and preventative checks are performed when it is above 90%. The health level classification standards for the subway platform screen door system are shown in Table 1, and the corresponding maintenance strategies are shown in Table 2. This achieves closed-loop management from data to decision-making. The innovation of this method lies in combining the multi-module characteristics of subway equipment with statistical models, avoiding the problem of relying on manual experience in traditional methods, and providing a reliable basis for health management.

[0043] Table 1

[0044] Table 2

[0045] S2: Enhance the filtered dataset obtained in S1 to obtain the expanded dataset. Over time, the characteristic parameters of the subway platform door system will gradually deviate from their respective calibration values, and correspondingly, the health status will decline. Linear interpolation can better simulate the continuous downward trend of characteristic parameters over time, effectively expanding the sample size, i.e., the original dataset suitable for this study. The data augmentation here aims to increase the number of samples to better fit the time series trend. In S2, linear interpolation is used for data augmentation, and the augmented dataset is the expanded dataset. The formula for linear interpolation is as follows: ; In the formula, and There are two known data points, and ; The independent variable that needs to be interpolated; For the corresponding estimated value, and It consists of two moments, before and after. It is always equal to 1. and The feature parameters are respectively in and The value at time is calculated by interpolating each characteristic parameter using the above formula.

[0046] S3: Use the XGBoost algorithm to filter important features in the augmented dataset obtained in S2 to form the experimental dataset; Step S3 uses the XGBoost algorithm to filter important features. The feature importance metric is the built-in gain. The gain is evaluated by calculating the sum of the loss reductions brought by the feature at all split nodes. The calculation formula is as follows: ; In the formula, and These are the sums of the first derivatives of all samples in the left and right child nodes, respectively. and These are the sums of the second derivatives of all samples in the left and right child nodes, respectively. This is a regularization parameter that controls the magnitude of node weights. This is a regularization parameter that controls the cost of the splitting operation. The XGBoost algorithm constructs a tree structure by iteratively splitting nodes. XGBoost calculates the gain of all split nodes and selects the feature with the largest gain as the splitting feature of the current node. The splitting process employs a binary partitioning strategy, if Then Node binary partitioning Nodes and Node; if Then for The nodes are divided, and so on, with the splitting mechanism as follows: Figure 3 As shown; S4: As Figure 1 As shown, the data in the experimental dataset formed in S3 is normalized, and the first 80% of the data in the time series is divided into the training set, and the last 20% of the data in the time series is divided into the test set. In S4, data normalization reduces the data to the [0,1] interval, eliminating the impact of the magnitude difference of the data on the prediction results. The formula is as follows: ; In the formula, This is the result after data normalization; This is a sample data point for the health status of subway platform door systems. and These represent the maximum and minimum values ​​of the health status sample data for the subway platform door system, respectively.

[0047] S5: An attention mechanism is introduced into the GRU. The pre-defined training and test sets are input into the GRU-Attention model to form an XGBoost-GRU-Attention model for health prediction. In S5, an attention mechanism is introduced into the GRU to form an XGBoost-GRU-Attention model for predicting the health of the subway platform door system. Figure 4 As shown, the prediction stage employs a GRU network combined with an attention mechanism. GRU addresses the gradient problem of traditional RNNs by updating and resetting the gate structure, while the attention mechanism enhances the ability to capture long-sequence dependencies and abrupt changes through dynamic weighting of key time steps. Ablation experiments demonstrate that this fusion architecture significantly reduces prediction error, with a lower coefficient of determination R0. 2 The performance was improved to over 0.96, with particularly excellent performance in scenarios involving sudden changes in health. In the structure of the baseline model GRU, express The input vector at time step; Indicates resetting the gate vector; This indicates updating the gate vector; Indicates the candidate hidden state; express Output vector at time step; This represents the activation function. , , , The calculation formula is as follows: ; ; ; ; In the baseline model GRU, an attention mechanism is introduced to capture the relationship between any two positions in the time series. The attention mechanism performs a weighted summation of the GRU network input vector and calculates the attention score using the Softmax function. The GRU-Attention model structure is as follows: Figure 5 As shown, the output vector The calculation formula is as follows: ; In the formula, This is the output vector of the GRU-Attention model; Attention score; This is the output vector of the GRU network. The calculation formula is as follows: ; In the formula, and These are the weighting coefficients; This is the bias coefficient; We want to input the well-defined training and test sets into the GRU-Attention model to generate the XGBoost-GRU-Attention model for predicting the health status of subway platform door systems.

[0048] S6: After the XGBoost-GRU-Attention model completes the health prediction, it performs inverse normalization on the data, compares the predicted values ​​with the true values, and outputs the evaluation results of the model's prediction performance.

[0049] In S6, the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) are selected. 2 To evaluate the predictive performance of the model; MSE and RMSE are used to measure the average deviation between model predictions and actual values; MAE is used to measure the absolute error between predictions and actual values; MAPE is used to measure the relative error between predictions and actual values; R... 2 Used to measure a model's ability to explain data variation; the specific formula is as follows: ; ; ; ; ; In the formula, This indicates the true health status of the subway platform door system; This represents the predicted health status of the subway platform door system. This represents the average of the true values.

[0050] The XGBoost-GRU-Attention model improves prediction accuracy and efficiency through multi-stage processing. The XGBoost algorithm ranks feature variables by importance through gain calculation, setting a gain threshold of 0.0300 to automatically select seven key features, effectively reducing model complexity. Figure 6 As shown, the feature importance ranking results intuitively display the key influencing factors, avoiding interference from redundant information.

[0051] Comparative Example The GRU network, combined with an attention mechanism, further enhances its ability to process time-series data. GRU addresses the gradient problem of traditional RNNs by updating and resetting gate structures, while the attention mechanism, through dynamic weighting of key time steps, improves the sensitivity to capturing abrupt changes in health status. For example... Figure 7 As can be seen from the comparison of prediction results, the XGBoost-GRU-Attention model proposed in this invention has the highest degree of fit with the true value, especially in the period of sudden change in health status.

[0052] The results of XGBoost ablation experiments, the results of Attention mechanism ablation experiments, and the model performance comparison results are shown in Tables 3, 4, and 5, respectively, where "XGB" represents XGBoost.

[0053] Table 3

[0054] Table 4

[0055] Table 5

[0056] Ablation experiments and comparative experiments show that the XGBoost-GRU-Attention model outperforms the baseline model in all metrics, such as MAPE being less than 5% and R... 2The value exceeds 0.96. The protection points of the system lie in its comprehensive evaluation method, index calculation standards, and specific application scenarios in subway equipment prediction, ensuring the model's verifiability and comparability. The key technologies in this solution not only cover theoretical innovation but also emphasize practical application value. The core technologies to be protected include the specific implementation details of the above methods, parameter optimization strategies, and system integration schemes. These together constitute a complete and implementable technical system, providing crucial support for subway equipment health management.

[0057] Therefore, this invention adopts the aforementioned method for evaluating and predicting the health status of subway platform screen door systems based on operation and maintenance data. This method closely integrates data-driven approaches with actual operation and maintenance needs, achieving full-process optimization from health status evaluation to dynamic prediction. Specifically, the patented solution first constructs a scientifically sound health status evaluation system using a multiple linear regression model. This model is based on 14 key characteristic variables of the subway platform screen door system. SPSS software verification shows a high multiple correlation coefficient of 0.964 and a determination coefficient of 0.929, indicating that the model can accurately quantify the health status of the equipment, laying a solid foundation for subsequent predictions.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for health evaluation and prediction of subway platform door systems based on operation and maintenance data, characterized in that, It includes a health assessment module and a health prediction module. The health assessment module quantifies the current health status of the device through multi-dimensional feature variables, while the health prediction module uses a fusion model to capture the temporal degradation pattern, specifically including the following steps: S1: Collect raw operation and maintenance data, and use multiple linear regression analysis to filter feature variables related to the target variable to obtain the filtered dataset; S2: Enhance the filtered dataset obtained in S1 to obtain an expanded dataset; S3: Use the XGBoost algorithm to filter important features in the expanded dataset obtained in S2 to form the experimental dataset; S4: Normalize the data in the experimental dataset formed in S3, and divide the data into the training set according to the first 80% of the time series data and the test set according to the last 20% of the time series data. S5: Introduce an attention mechanism into GRU, input the divided training set and test set into the GRU-Attention model to form the XGBoost-GRU-Attention model for health prediction; S6: After the XGBoost-GRU-Attention model completes the health prediction, it performs inverse normalization on the data, compares the predicted values ​​with the true values, and outputs the evaluation results of the model's prediction performance.

2. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 1, characterized in that: In S1, the operation and maintenance data of the subway platform door system is collected. Each set of operation and maintenance data contains 14 characteristic parameters U1~U14. The subway platform door system includes a safety circuit conversion device ATC module, a unit controller PEDC module, and a signal conversion circuit SIG module. Among them, U1~U5 belong to ATC, U6~U11 belong to PEDC, and U12~U14 belong to SIG.

3. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 2, characterized in that: In step S1, the characteristic variables U1~U14 and the target variable "health of subway platform screen door system" are input into the statistical software SPSS for multiple linear regression analysis. The regression model is as follows: ； In the formula, The target variable is "health of the subway platform door system"; As a characteristic variable; The intercept; These are the regression coefficients; This is the error term; The significance of the model was verified by goodness-of-fit test, F-test, and T-test, and the formula for calculating the health of the subway platform door system was obtained: ； enter The time-feature value output is the health status of the subway platform door system at that time.

4. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 1, characterized in that: In S2, linear interpolation is used for data augmentation. The augmented dataset is called the expanded dataset. The formula for linear interpolation is as follows: ； In the formula, and There are two known data points, and ; The independent variable that needs to be interpolated; For the corresponding estimated value, and It consists of two moments, before and after. It is always equal to 1. and The feature parameters are respectively in and The value at time is calculated by interpolating each characteristic parameter using the above formula.

5. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 1, characterized in that: Step S3 uses the XGBoost algorithm to select important features. The feature importance metric is the built-in gain. The gain is calculated by summing the reduction in loss caused by the feature at all split nodes to evaluate the importance of the feature. The calculation formula is as follows: ； In the formula, and These are the sums of the first derivatives of all samples in the left and right child nodes, respectively. and These are the sums of the second derivatives of all samples in the left and right child nodes, respectively. This is a regularization parameter that controls the magnitude of node weights. This is a regularization parameter that controls the cost of the splitting operation. The XGBoost algorithm constructs a tree structure by iteratively splitting nodes. XGBoost calculates the gain of all split nodes and selects the feature with the largest gain as the splitting feature of the current node. The splitting process employs a binary partitioning strategy, if Then Node binary partitioning Nodes and Node; if Then for The nodes are divided.

6. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 1, characterized in that: In S4, data normalization reduces the data to the [0,1] interval, eliminating the impact of magnitude differences in data on the prediction results. The formula is as follows: ； In the formula, This is the result after data normalization; This is a sample data point for the health status of subway platform door systems. and These represent the maximum and minimum values ​​of the health status sample data for the subway platform door system, respectively.

7. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 1, characterized in that: In S5, an attention mechanism is introduced into GRU to form the XGBoost-GRU-Attention model for predicting the health status of subway platform door systems. In the structure of the baseline model GRU, express The input vector at time step; Indicates resetting the gate vector; This indicates updating the gate vector; Indicates the candidate hidden state; express Output vector at time step; This represents the activation function. , , , The calculation formula is as follows: ； ； ； ； In the baseline model GRU, an attention mechanism is introduced to capture the relationship between any two positions in the time series. The attention mechanism performs a weighted summation of the GRU network input vectors and calculates the attention score using the Softmax function. The output vector in the GRU-Attention model is... The calculation formula is as follows: ； In the formula, This is the output vector of the GRU-Attention model; Attention score; This is the output vector of the GRU network. The calculation formula is as follows: ； In the formula, and These are the weighting coefficients; This is the bias coefficient; We want to input the well-defined training and test sets into the GRU-Attention model to generate the XGBoost-GRU-Attention model for predicting the health status of subway platform door systems.

8. The method for health evaluation and prediction of subway platform door system based on operation and maintenance data as described in claim 1, characterized in that: In S6, the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) are selected. 2 To evaluate the predictive performance of the model; MSE and RMSE are used to measure the average deviation between model predictions and actual values; MAE is used to measure the absolute error between predictions and actual values; MAPE is used to measure the relative error between predictions and actual values; R... 2 Used to measure a model's ability to explain data variation; the specific formula is as follows: ； ； ； ； ； In the formula, This indicates the true health status of the subway platform door system; This represents the predicted health status of the subway platform door system. This represents the average of the true values.

Images