Research on a Network Diagnosis Method for a Train Control Center and Interlocking Integrated System Based on a Fuzzy Broad Learning System Model

Abstract

In high-speed railway signaling systems, the network structure of the Train Control Center and Inter-locking Integrated System (TIS) is highly complex, with a large number of interfaces, numerous redundant channels, and forwarding components such as switches. These factors result in challenges such as insufficient accuracy, low efficiency, and poor real-time performance in terms of network monitoring and fault diagnosis. As the scale of railway yards continues to expand, these issues are becoming increasingly prominent. To address these challenges, this paper proposes a network fault propagation model based on the Fuzzy Broad Learning System (FBLS). By leveraging nonlinear transformations and feature mapping techniques, FBLS can efficiently extract and analyze network fault features, even with a relatively small amount of data. Experimental results show that the FBLS-based diagnostic model achieves higher accuracy and faster response speed in fault identification and propagation path analysis compared to traditional graph theory and fuzzy reasoning methods. Further comparisons with existing methods validate the advantages of FBLS in handling multi-source data, improving noise tolerance, and adapting to large-scale railway yard network systems, demonstrating its broad application prospects in railway signaling systems.

1. Introduction

Train Control Centers (TCCs) and the Computer Based Interlocking (CBI) play a crucial role in ensuring train safety [1]. The Train Control and Interlocking Integrated System (TIS) merges TCC and CBI, effectively solving the interface issues caused by their independent configuration; it has thus become an important direction for the future development of China’s high-speed rail signal system [2]. As the scale of railway stations continues to expand and the complexity of equipment increases, network failures have become a key factor affecting system stability. In the high-risk environment of high-speed rail, network failures could lead to dispatch errors, delays, or even major safety incidents. Therefore, accurately and rapidly diagnosing the propagation path of network failures has become a core issue in ensuring the stable operation of such systems.

Traditional fault diagnosis methods primarily rely on technologies such as graph theory, fuzzy reasoning, and rule-based systems. In this regard, Xu et al. [3], based on research on network monitoring requirements, proposed a hybrid reasoning method that combines graph theory models and fuzzy reasoning. This method is aimed at reasoning and analyzing typical switch faults and channel faults. These methods are effective in small-scale networks; however, when dealing with large-scale complex networks, they face issues such as low computational efficiency, poor accuracy, and a strong dependency on expert databases.

In recent years, machine learning and deep learning have made progress in the field of fault recognition. For example, Zuo et al. [4] proposed a complex power grid cascading fault diagnosis method based on semi-supervised machine learning, successfully improving the accuracy and efficiency of diagnoses. Lei et al. [5] reviewed fault diagnosis techniques for switch machines based on deep learning, showcasing the application potential of deep learning in rail transit systems. In addition, Liu et al. [6] conducted a study on a fault diagnosis method for turnout switch equipment based on the random forest algorithm (RFA). This method utilizes the powerful learning capabilities of the random forest algorithm to efficiently and accurately identify turnout faults, thereby improving the reliability of fault diagnoses. Zhong [7] proposed a turnout fault classification algorithm that combines support vector machines (SVM) with a hybrid classification strategy. Experimental results show that the SVM-based fault diagnosis algorithm demonstrates excellent classification performance for typical turnout fault patterns. However, these methods have relatively poor interpretability, lacking clear explanations of the model’s decision-making process, which limits their effectiveness in practical applications.

To address these issues, the Neuro-Fuzzy Model has gradually gained attention. This model combines the classification ability of neural networks with the reasoning advantages of fuzzy logic, thereby providing good interpretability. For example, Wu et al. [8] proposed an initial fault diagnosis method for T-S fuzzy systems and applied it to high-speed rail traction equipment. Liu et al. [9], based on the Takagi-Sugeno fuzzy model, proposed fault estimation and signal compensation methods, applying them to wind turbine systems. Bai [10] applied the Fuzzy ELM (FELM) algorithm to an intelligent fault diagnosis method for urban rail transit signaling equipment systems. However, traditional neuro-fuzzy systems [11] still face challenges in terms of automatically determining their structure, which leads to complexities in the training process, large model sizes, and difficulty in interpretation [12].

The Broad Learning Structure [13], due to its lack of multi-layer connections and the absence of the need for gradient descent to update weights, offers faster computation speed. The Fuzzy Broad Learning System is a novel type of fuzzy neural network [14]. By introducing a set of first-order T–S fuzzy systems to replace the feature nodes on the basis of broad learning, it improves the classification ability while maintaining computational efficiency. For example, Li et al. [15] proposed a photovoltaic power generation forecasting method based on the fuzzy broad learning model and demonstrated its superior performance. Wang et al. [16] proposed a mechanical equipment fault diagnosis model based on the TSK fuzzy broad learning system, showing the method’s effectiveness and high accuracy in fault diagnosis.

Currently, there is relatively little research on the application of the Fuzzy Broad Learning System (FBLS) in the railway sector. In this context, this study proposes a network fault diagnosis method for railway signal systems based on FBLS, applying it to analyze network fault diagnoses and fault propagation path analyses in the TIS system. FBLS, with its extended network structure and nonlinear transformations, can achieve efficient learning and fault diagnosis with fewer training data, thereby improving computational efficiency, accuracy, and interpretability, addressing the limitations of existing models. The specific contributions of this paper are as follows:

• A network fault diagnosis method based on FBLS is proposed, improving the accuracy and real-time performance of diagnosis.
• Interpretable Linguistic Fuzzy Rules (ILFR) are embedded into the enhanced nodes, enabling the interpretability and readability of FBLS.
• The method resolves redundancy and cross-fault issues in complex networks, improving fault propagation analysis.

The structure of this paper is as follows: Section 2 describes the basic principles of FBLS in detail. Section 3 introduces the network fault propagation mechanism of the TIS system and elaborates on the application of FBLS in network fault analyses. Section 4 presents the experimental design based on FBLS and displays the experimental results. Section 5 summarizes the findings of this study and suggests directions for future research.

2. Fuzzy Broad Learning System

2.1. Overview of the FBLS Model

The Fuzzy Broad Learning System (FBLS) combines the Random Vector Functional Link Neural Network (RVFLNN) and pseudoinverse theory, offering an efficient and progressive learning method through a multi-layered neural-fuzzy inference architecture [14]. The basic architecture of FBLS is shown in Figure 1. Unlike the standard Broad Learning System (BLS) [13], FBLS uses multiple first-order Takagi-Sugeno (TS) fuzzy subsystems in the feature node layer, employing the k-means algorithm to group input data and determine the number of fuzzy subsystem rules. This approach allows for the extraction of more fault feature information. In the fuzzy subsystem layer, input data are processed through fuzzy rules to generate multiple fuzzy subsystems. The intermediate values produced by these fuzzy subsystems are then directly passed to the enhanced node layer for nonlinear transformation, and the final output is generated by the combined outputs of the enhanced node layer and the fuzzy subsystems.

2.2. Key Components and Working Principles of FBLS

• Construction of Fuzzy Rules

The Fuzzy Broad Learning System (FBLS) establishes a nonlinear mapping relationship between input features and system outputs through fuzzy rules. The input feature matrix is defined as $X=(x_1,x_2,\dots ,x_N{)}^T\in {\mathbb{R}}^{N\times M},$ where each sample is represented as $x_s=(x_{s1},x_{s2},\dots ,x_{sM}),s=1,2,\dots ,N.$ The fuzzy rules follow the Takagi-Sugeno (TS) model, and their mathematical expression is given by:

$$z_{sk}^i=f_k^i(x_{s1},x_{s2},\dots ,x_{sM})=\sum _{t=1}^M {\alpha }_{kt}^i{x}_{st}$$

(1)

where ${\alpha }_{kt}^i$ are coefficients initialized randomly from a uniform distribution over $[0,1]$ and later determined by the pseudo-inverse method.

The activation strength of the $k$-th fuzzy rule in the $i$-th fuzzy subsystem is computed as:

$${\tau }_{sk}^i={\sum }_{t=1}^M {\mu }_{kt}^i(x_{st})$$

(2)

and the normalized weighted activation strength for each fuzzy rule is represented as:

$${\omega }_{sk}^i={\displaystyle \frac{{\tau }_{sk}^i}{\sum _{k=1}^{K_i} {\tau }_{sk}^i}}$$

(3)

The membership function (${\mu }_{kt}^i\left(x\right)$) uses a Gaussian function and is defined as:

$${\mu }_{kt}^i(x)=e^{-{\left({\displaystyle \frac{x-c_{kt}^i}{{\sigma }_{kt}^i}}\right)}^2}$$

(4)

where $c_{kt}^i$ and ${\sigma }_{kt}^i$ are the center and width (standard deviation) of the fuzzy set, respectively.

2.

Output of the Fuzzy System

The output of each fuzzy subsystem is formed by the weighted combination of the fuzzy rule results, as illustrated in Figure 2. For the $s$-th training sample, the intermediate output vector of the $i$-th fuzzy subsystem is:

$$Z_{si}=({\omega }_{s1}^i{z}_{s1}^i,{\omega }_{s2}^i{z}_{s2}^i,\dots ,{\omega }_{s{K}_i}^i{z}_{s{K}_i}^i)$$

(5)

Expanding this to all training samples, the output matrix of the $i$-th fuzzy subsystem is:

$$Z_i=(Z_{1i},Z_{2i},\dots ,Z_{Ni}{)}^T\in {\mathbb{R}}^{N\times K_i},i=1,2,\dots ,n$$

(6)

To ensure consistency in notation, the intermediate output matrix of all $n$ fuzzy subsystems is represented as:

$$Z^n=(Z_1,Z_2,\dots ,Z_n)\in {\mathbb{R}}^{N\times (K_1+K_2+\cdots +K_n)}$$

(7)

where $Z^n$ contains the core features extracted by the fuzzy rule system, providing input for subsequent processing by the enhancement nodes.

3.

Feature Transformation by Enhancement Nodes

To enhance the nonlinear representation capability of the system, FBLS introduces enhancement nodes to perform nonlinear transformations on the output matrix $Z^n$. The output matrix of the enhancement layer is:

$$H^m=\left(H_1,H_2,\dots ,H_m\right)\in {\mathbb{R}}^{N\times \left(L_1+L_2+\cdots +L_m\right)}$$

(8)

$H_j={\mathsf{\xi }}_{\mathrm{j}}({\mathrm{Z}}^{\mathrm{n}}{\mathrm{W}}_{hj}+{\mathsf{\beta }}_{hj})\in {\mathbb{R}}^{\mathrm{N}\times {\mathrm{L}}_{\mathrm{j}}}$, and: $W_h^j$: Weight matrix connecting the fuzzy subsystem outputs to the enhancement nodes; ${\beta }_h^j$: Bias terms.

The weights and biases are randomly generated from the interval $[0,1]$ to provide mutually independent enhancement node groups.

4.

System Output and Weight Optimization

The output of FBLS is given by:

$$\hat{Y}=\left[\begin{array}{c}{Z}^n,H^m\end{array}\right]W_F=N_{\mathrm{out}}{W}_F$$

(9)

$\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} W_F$ is the weight matrix connecting all fuzzy subsystems and enhancement nodes to the output layer.

The calculation formula for $W_F$ is:

$$W_F={\left(\lambda I+N_{\mathrm{out}}^T{N}_{\mathrm{out}}\right)}^{-1}{N}_{\mathrm{out}}^T\hat{Y}$$

(10)

2.3. Embedding Interpretable Linguistic Fuzzy Rules (ILFR)

In complex network fault diagnoses, the interpretability of the model is a key factor in achieving efficient operations and enhancing user trust. Traditional neural networks are often considered “black-box” models, making it difficult to clearly demonstrate their internal reasoning logic. The introduction of Interpretable Linguistic Fuzzy Rules (ILFR) addresses this issue. The goal is to express the relationship between inputs and outputs clearly in natural language, thereby improving the system’s transparency and readability.

• Generation of fuzzy rules

The fuzzy subsystem layer calculates the membership degree of input variables to generate a fuzzy rule base. Each rule describes the fuzzy relationship between input and output variables.

For example, for input variables $x_1,x_2,\dots ,x_M$, the fuzzy rule can be expressed as:

$$\mathrm{If} x_1 \mathrm{is} A_1,x_2 \mathrm{is} A_2,\dots ,x_M \mathrm{is} A_M, \mathrm{then} y=f\left(x_1,x_2,\dots ,x_M\right).$$

(11)

$f\left(x\right)$: The fuzzy mapping function, used to calculate the output of the fuzzy rule.

2.

Embedding Rules into Enhancement Nodes

Enhancement nodes are key layers in FBLS, responsible for applying nonlinear transformations to the outputs of fuzzy subsystems. ILFR is embedded into enhancement nodes in the following manner: The output of fuzzy rules is combined with the membership degree of input variables to generate the output of enhancement nodes. The calculation formula for the output is:

$$o_{ik}={\sum }_{j=1}^K {\omega }_{ij}{\mu }_{ij}(x)$$

(12)

$\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} o_{ik}$ is the output of the $k$-th enhancement node; ${\omega }_{ij}$ is the weight of the fuzzy rule; and ${\mu }_{ij}(x)$ is the membership function of the input variable.

3.

Linguistic Rule Outputs

To achieve greater interpretability, the outputs of enhancement nodes are processed into natural language explanations. For instance, for the fuzzy rule “If $x_1$ fails and $x_3$ fails, then y is fault state 5”, the model may output an explanation such as: “All logical channels in this physical channel through Interlocking System I and Switch B are faulty, and the physical channel is shown as faulty”.

2.4. Redundant Path and Cross-Fault Analysis

2.4.1. Resolving the Redundant Path Problem

Redundancy paths refer to multiple nodes or channels in a network being interconnected through alternate or backup links. While such connections enhance the system’s fault tolerance to some extent, they also increase the difficulty of fault propagation path analysis. FBLS addresses this issue through the following method:

• Path selection using fuzzy rules

In the fuzzy subsystem layer of the FBLS model, fuzzy rules are defined to describe the membership relationships between nodes and paths. By calculating the fuzzy membership degree of each path, FBLS prioritizes paths with higher membership degrees for fault propagation analysis, thus reducing the interference caused by redundancy paths in the diagnostic results.

The formula is as follows:

$${\mu }_i(x)=e^{-c_i^2(x-p_i{)}^2}$$

(13)

where ${\mu }_i\left(x\right)$ is the membership degree of the path; and $c_i,p_i$ is the center and width of the fuzzy membership function.

By filtering paths with low membership degrees, the influence of redundant paths on propagation analyses is minimized.

2.

Feature Optimization in Enhancement Nodes

The enhancement node layer applies nonlinear feature mapping to adjust and enhance the outputs of fuzzy rules, capturing the differences between paths. Compared to traditional methods, FBLS can explicitly distinguish main paths from redundant paths, reducing the impact of redundant paths on fault propagation path analysis. The specific calculation formula is:

$$H=x_i(Z^T{W}_i+b_i)$$

(14)

where $H$ is the output of the enhancement node; and $Z^T$ is the transposed output matrix of the fuzzy subsystem.

2.4.2. Resolving the Cross-Fault Problem

Cross-faults occur when multiple fault points happen simultaneously and their propagation paths overlap, making fault source localization and propagation analysis more challenging. FBLS addresses cross-fault problems using the following methods:

• Parallel Modeling of Multiple Fuzzy Subsystems

FBLS establishes a separate set of fuzzy rules for each path in the fuzzy subsystem layer, enabling simultaneous analysis of multiple paths through parallel modeling. Each fuzzy subsystem operates independently, avoiding interference between overlapping paths. Assuming M paths are described by K fuzzy rules, the output of the rules is:

$$z_{ik}={\prod }_{i=1}^M {\mu }_i(x_{ik})$$

(15)

$\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} z_{ik}$ is the oiutput of the $k$-th fuzzy rule; and ${\mu }_i\left(x_{ik}\right)$ is the membership degree of the $i$-th path.

2.

Independent Diagnosis of Fault Intersection Points

For faults at intersection points, FBLS uses fuzzy rule membership functions to separate input features, ensuring that the independent contribution of each fault point is explicitly identified. By analyzing the output weight matrix of the enhancement nodes, FBLS accurately calculates the fault characteristics of the intersection points.

3. TIS System Network Monitoring Case Study

3.1. Network Fault Propagation in the TIS System

The network fault propagation in the TIS system involves interactions between multiple layers and components. The overall structure of the system can be divided into the interlocking platform, communication control layer, and electronic execution module layer (see Figure 3) [17]. The interlocking platform, as the core of the system, is responsible for logical operations and fault detection and is key to ensuring that the system operates normally and responds promptly. The communication control layer facilitates data transmission between modules and safety computers using CAN buses and IP-based network interfaces while ensuring communication security through an independent bus structure. The electronic execution module layer includes the specific execution components, which execute interlocking platform commands through connections with the communication control board.

Network faults primarily propagate through the communication control layer and may affect the equipment in the electronic execution module layer. Due to the high degree of integration between the communication layer and the execution modules, faults can spread quickly, potentially causing multiple modules to malfunction and affecting the system’s stability. Although the TIS system uses redundant design to improve fault tolerance, this redundancy may expand the fault display range. For instance, a single switch failure might manifest as multiple channel failures, thus concealing the actual issue and increasing the difficulty of fault localization.

3.2. TIS System Network Monitoring Interface

To better describe the network faults in the TIS system, based on graph theory principles, all physical devices in the communication control layer and their connection channels are visually displayed through physical connections, as shown in Figure 4. The system’s graphical interface fully corresponds to the actual devices and their connecting cables, achieving precise matching between physical objects and graphical representations. This allows users to quickly identify and address faulty devices [18].

In Figure 4 (an example of a network monitoring interface), nodes are represented as squares or rectangles, and channels are displayed as connecting lines. The status of nodes is differentiated by the color of their borders: green indicates the current primary control or normal state, yellow represents standby or pending switch, red signifies a fault, and gray indicates an unknown state. Similarly, channels use colors to indicate their health status: green for normal, orange for partial fault, red for fault, and gray for unknown. With this visual layout, operation and maintenance personnel can directly track alarm information on the interface, follow the connecting lines step by step to query associated nodes downstream or upstream, and quickly and accurately locate the fault source while distinguishing the false “multi-channel faults” caused by redundant designs.

• Node Device Design

After analyzing the 2-out-of-2 redundant structure of the communication control layer in the TIS system, the internal node devices of the system are identified, as shown in

Table 1. TIS System Network Monitoring Device Node.

Device	Node
TIS-A	A1
TIS-B	A2
Switch A	B1
Switch B	B2
Maintenance Terminal	C1
Operator Console A	C2
Operator Console B	C3
CTC-A	D1
CTC-B	D2
TSRS-A	E1
TSRS-B	E2
…	…

. These nodes include Operator Console A and Operator Console B machines, TIS-A and TIS-B, external devices such as CTC-A and CTC-B machines, as well as adjacent devices like CBI, TCC, and RBC hosts. These nodes typically possess the capability for primary–backup switching or mutual redundancy. When a primary control node encounters a fault, it can switch to a standby node to continue providing services.

2.

Node and Logical Channel Status Database

To accurately detect the propagation path of faults, it is necessary to obtain the status information of the physical channels between TIS nodes. However, the network monitoring data of the TIS system are relatively scattered, involving a large number of channels. The maintenance terminal can only collect the status information of logical channels, as shown in

Table 2. TIS System Network Logical Channel Information Tab.

Logical Channel	Code	Operating Status
TIS-A to Operator Console A (Switch A)	1	0–2
TIS-A to Operator Console B (Switch A)	2	0–2
TIS-A to Operator Console A (Switch B)	3	0–2
TIS-A to Operator Console B (Switch B)	4	0–2
TIS-B to Operator Console A (Switch A)	5	0–2
TIS-B to Operator Console B (Switch A)	6	0–2
TIS-B to Operator Console A (Switch B)	7	0–2
TIS-B to Operator Console B (Switch B)	8	0–2
Operator Console A to CTC-A	9	0–2
Operator Console A to CTC-B	10	0–2
Operator Console B to CTC-A	11	0–2
Operator Console B to CTC-B	12	0–2
Maintenance Terminal to TIS-A (Switch A)	13	0–2
Maintenance Terminal to TIS-B (Switch A)	14	0–2
Maintenance Terminal and TIS-A (Switch B)	15	0–2
Maintenance Terminal and TIS-B (Switch B)	16	0–2
Maintenance Terminal and Operator Console A (Switch A)	17	0–2
Maintenance Terminal and Operator Console B (Switch A)	18	0–2
Maintenance Terminal and Operator Console A (Switch B)	19	0–2
Maintenance Terminal and Operator Console B (Switch B)	20	0–2
…	…	…

, and cannot directly obtain status data from physical channels. Work statuses 0–2 represent unknown, normal, and faulty, respectively. Traditional methods often rely on expert databases to convert the logical channel status to the corresponding physical channel status, but since physical channels typically contain multiple logical channel status segments, this conversion is not entirely accurate. Moreover, the correspondence between physical and logical channels partially overlaps, making precise matching difficult. As a result, traditional methods face significant challenges in fault diagnosis.

To overcome this issue, this paper adopts a method based on the Fuzzy Broad Learning System (FBLS). The FBLS method directly uses the status information of the TIS logical channels as input data, without relying on the complex conversion process between physical and logical channels. This approach avoids mapping errors caused by the partial overlap between logical and physical channels in traditional methods, thereby improving diagnostic efficiency.

4. Experimental Design and Network Fault Analysis Based on FBLS

The overall framework of the FBLS-based TIS system network fault diagnosis model is shown in Figure 5, which is divided into four parts. In the offline phase, the historical operational data and network status information of the TIS system are first preprocessed and divided into training and testing sets. After normalization, the training set is input into the fuzzy system layer for automatic feature extraction. Then, a nonlinear transformation is used to construct the enhanced node layer, ultimately obtaining the optimized model parameters with minimal error, completing the offline training of the TIS network fault diagnosis model. In the online phase, the diagnosis model trained offline is used for real-time network status monitoring and fault diagnosis of the TIS system.

4.1. Experimental Design

To validate the effectiveness of FBLS in network fault diagnoses and propagation path analyses, this study designs a series of experiments based on the actual operational scenarios of the TIS system, evaluating FBLS’s performance in terms of fault feature extraction, propagation path analysis, diagnostic accuracy, and real-time performance.

• Data Preparation

The maintenance terminal of the TIS system can display the operational status of equipment in real time, replay historical information, and support fault information queries. By collecting alarm information from the network status [19], an alarm information table for the TIS network status is compiled. The faults indicated by these alarm messages are defined as different target labels, while the status information of the corresponding logical channels and nodes for each alarm is recorded to construct sample features, as shown in

Table 3. TIS Network Alarm Information and Fault Feature Labels.

Fault Labels	Alarm Information	Corresponding Logical Channel Indexes	Operating Status
Fault State 1	TIS-A to Switch A communication interrupted	1, 2, 13	0–2
Fault State 2	TIS-A to Switch B communication interrupted	3, 4, 15	0–2
Fault State 3	TIS-B to Switch A communication interrupted	5, 6, 14	0–2
Fault State 4	TIS-B to Switch B communication interrupted	7, 8, 16	0–2
Fault State 5	Operator Console A to Switch A communication interrupted	1, 5, 17	0–2
Fault State 6	Operator Console A to Switch B communication interrupted	3, 7, 19	0–2
Fault State 7	Operator Console B to Switch A communication interrupted	2, 6, 18	0–2
Fault State 8	Operator Console B to Switch B communication interrupted	4, 8, 20	0–2
Fault State 9	Maintenance Terminal to Switch A communication interrupted	13, 14, 17, 18	0–2
Fault State 10	Maintenance Terminal to Switch B communication interrupted	15, 16, 19, 20	0–2
Fault State 11	Operator Console A to CTC-A communication interrupted	9	0–2
Fault State 12	Operator Console A to CTC-B communication interrupted	10	0–2
Fault State 13	Operator Console B to CTC-A communication interrupted	11	0–2
Fault State 14	Operator Console B to CTC-B communication interrupted	12	0–2
Fault State 15	TIS-A to Secure Network A Switch communication interrupted	21, 22, 29, 30, 36, 38, 45, 46, 53, 54	0–2
Fault State 16	TIS-A to Secure Network B Switch communication interrupted	23, 24, 31, 32, 39, 40, 47, 48, 55, 56	0–2
Fault State 17	TIS-B to Secure Network A Switch communication interrupted	25, 26, 33, 34, 41, 42, 49, 50, 57, 58	0–2
Fault State 18	TIS-B to Secure Network B Switch communication interrupted	27, 28, 35, 36, 43, 44, 51, 52, 59, 60	0–2
…	…	…	…

. After data cleaning and normalization, the dataset is organized into an input matrix $X\in {\mathbb{R}}^{N\times M}$, where N represents the number of samples and M is the feature dimension for each sample.

Each row of the input matrix X represents an alarm sample, and each column corresponds to a feature of that sample, including the logical channel state and node state. Each alarm record is labeled with a different target, serving as the supervised learning objective of the model. The label value y represents the corresponding fault type, with a range of y ∈ {1,2,…,K}, where K is the total number of fault types. By combining the input matrix X with the label y, a complete training and testing dataset can be constructed, providing effective input data support for the FBLS model.

2.

Experimental Process

Data Preprocessing: The monitoring data are normalized to construct feature vectors for logical channels and node states, eliminating noise interference.

Model Training and Testing: The processed data are input into the FBLS model, with 80% of the samples used for training and 20% for testing. The model’s performance in terms of fault detection and propagation path prediction is evaluated.

Embedding Interpretable Linguistic Fuzzy Rules (ILFR): ILFR is embedded into the enhancement node layer to analyze its effect on improving the model’s interpretability.

Performance Evaluation: The performance of FBLS is compared with traditional methods (e.g., graph theory and fuzzy reasoning-based diagnostic methods) to evaluate the model’s diagnostic accuracy, real-time performance, and interpretability.

4.2. Optimizing Model Parameters

FBLS extracts fuzzy features of logical channel and node states through the fuzzy subsystem layer and uses enhancement nodes for the nonlinear optimization of these features, enabling precise detection of network faults.

Figure 6 illustrates the trends in training and testing accuracy as the parameter NumRule (number of rules) varies, with NumFuzz (number of fuzzy sets) differentiated by color. It can be observed that when NumRule = 100 and NumFuzz = 50, the training accuracy reaches a maximum of 98%, and the testing accuracy achieves 97%. As the number of rules increases, both the training and testing accuracy improve; however, an excessively high number of rules may lead to increased model complexity with diminishing returns.

From the 3D perspective shown in Figure 7, it can be observed that when the parameter combinations fall within an optimal range, the model exhibits stable and high-accuracy performance. This demonstrates that parameter optimization has a significant impact on the model’s performance.

4.3. Experimental Results

• Model Performance Evaluation

To verify the effectiveness and advantages of FBLS in terms of accuracy and interpretability, we compared FBLS with several representative diagnostic algorithms, including: SVM (Support Vector Machine), a commonly used classifier in traditional machine learning; BLS (Broad Learning System), a broad learning model without a fuzzy layer; BP (Backpropagation Neural Network), a classical BP neural network; RFA (Random Forest Algorithm), an ensemble model based on decision trees; and FELM (Fuzzy Extreme Learning Machine), a fuzzy neural network model.

Table 4. Performance Comparison of Different Models.

Algorithm	RMSE	MAPE	Training Accuracy	Testing Accuracy
SVM	13.9039	8.22%	98.48%	86.59%
BLS	8.4395	3.24%	98.78%	95.12%
BP	8.4741	3.68%	99.24%	95.12%
RFA	6.0863	1.62%	98.48%	97.56%
FELM	6.5848	2.99%	99.24%	95.73%
FBLS	5.5497	1.19%	98.32%	98.17%

summarizes the RMSE (Root Mean Squared Error), MAPE (Mean Absolute Percentage Error), training accuracy, and testing accuracy of each algorithm on the same dataset. The results demonstrate that FBLS exhibits significant advantages across all metrics:

(1) Minimal Numerical Deviation

The RMSE of the FBLS model is only 5.5497, significantly reducing the error compared to other methods. Additionally, the MAPE value is as low as 1.19%, the smallest among all methods, indicating an extremely low deviation from the true fault labels.

(2) High Accuracy and Strong Generalization Ability

FBLS achieves a training accuracy of 98.32% and a testing accuracy of 98.17%, significantly outperforming methods such as SVM (86.59%), BLS (95.12%), BP (95.12%), and FELM (95.73%). Even when compared with the relatively well-performing RFA (97.56%), FBLS still has room for further improvement.

Currently, for network fault diagnoses in the TIS system in China, methods based on graph theory and fuzzy inference are predominantly used. To better demonstrate the superior performance of FBLS, scatter plots of the actual labels and predicted labels under both the FBLS model and the fuzzy inference model have been generated, as shown in Figure 8. These plots provide an intuitive comparison of the distribution of predicted labels from the fuzzy inference model and the FBLS model in relation to actual labels.

In terms of diagnostic accuracy, under a noisy environment, the FBLS model achieves an average diagnostic accuracy of 95.2% on the test set, which is significantly higher than that of traditional graph theory and fuzzy inference-based methods, which range between 85% and 88%.

2.

Real-Time Response Evaluation

To quantify the real-time performance of FBLS, the experiment defines three key time metrics:

• Fault detection time ($T_d$): The time required for the model to detect a logical channel fault trigger.
• Diagnosis time ($T_{\mathrm{diag}}$): The time required for the model to infer the fault type and propagation path.
• Fault display time ($T_{\mathrm{display}}$): The time taken for the detected fault to appear on the monitoring interface.

For the optimized method proposed in this paper, the fault detection and display times remain relatively constant, while FBLS primarily optimizes the fault diagnosis time ($T_{\mathrm{diag}}$).

(1) Collecting Fault Logical Channel Information

In the TIS (Train Control System), network faults are typically reflected through logical channels. Changes in logical channel states (such as faults, timeouts, packet loss, etc.) help detect anomalies in the system. The first step in real-time response evaluation is to collect logical channel information across the system, and then to monitor and record their state changes.

• Data Collection: When a fault occurs during system operation, the network monitoring system captures the corresponding fault information. This typically includes the status of different nodes and the state of signal transmission links.
• Data Preparation: Once fault information is collected, it is formatted into input data (feature vectors) that the FBLS algorithm can use. These input data include features related to the fault.

According to the technical specifications of the integrated train control and interlocking system, the minimum determination time for communication fault detection should be greater than 3000 ms. The calculation for the minimum communication fault determination time is given by:

$$T_d=(500+t_1)\times 3+(500+t_2)\times 3$$

(16)

$\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} T_d$ is the minimum determination time for communication faults; $t_1$ is the frame transmission time; and $t_2$ is the communication request frame transmission time.

(2) Fault Type Identification Using the FBLS Model

The core task of the FBLS algorithm is to quickly determine fault types based on the collected fault feature data and predict fault propagation paths. By combining fuzzy logic and broad learning, FBLS can provide accurate classifications and inferences when dealing with complex fault patterns.

• Feature Input: The collected fault feature data are fed into the FBLS model. During the training phase, FBLS processes these inputs through its fuzzy subsystems, maps them onto different fuzzy rules, and extracts useful fault information.
• Inference Process: The enhancement node layer of FBLS performs nonlinear mapping, integrating outputs from multiple fuzzy subsystems. Based on these calculations, the FBLS model determines the specific fault type and predicts the affected system areas (fault propagation paths).

In this experiment, a total of 140 fault datasets were selected for testing, with a training time of 24.9560 s and a testing time of 1.1181 s.

These results indicate that the FBLS model can complete training in a relatively short time and perform fault diagnosis efficiently, significantly improving diagnostic efficiency.

To further quantify the real-time response performance of FBLS, we calculate the average diagnosis time ($T_{\mathrm{diag}}$) using the following formula:

$$T_{\mathrm{diag}}={\displaystyle \frac{T_{\mathrm{train}}+T_{\mathrm{test}}}{N}}$$

(17)

${\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} T}_{\mathrm{diag}}$ is the average diagnosis time; $T_{\mathrm{train}}$ is the training time; $T_{\mathrm{test}}$ is the testing time; and $N$ is the number of fault datasets.

For the experimental data, the FBLS model’s average diagnosis time per fault dataset is only 0.1836 s, demonstrating exceptionally high diagnostic efficiency and confirming that FBLS meets real-time response requirements in practical applications.

(3) Display of Diagnostic Results via the Network Monitoring Interface

After completing fault type identification and propagation path prediction, the FBLS algorithm displays the diagnostic results in real-time through the network monitoring interface for operation and maintenance personnel, with a display time ($T_{\mathrm{display}}$) of 48 ms. Through the graphical interface, operators can intuitively understand the specific fault conditions and quickly locate the fault source.

For example, consider a network cable fault between TIS-B and Switch A. This fault causes all logical channels from TIS-B to Switch A to be completely interrupted. However, since Operator Display Machine A, Operator Display Machine B, and the Maintenance Machine can still communicate normally with TIS-A via Switch A, the fault is limited to a communication disruption between TIS-B and Switch A.

After diagnosis using the FBLS model, the system accurately displays the fault propagation path and the affected nodes on the interface, visually marking the fault location, as illustrated in Figure 9. This assists maintenance personnel in quickly pinpointing and resolving the issue.

3.

Interpretability Analysis

In addition to its advantages in classification accuracy and structure, another key strength of FBLS is its interpretability in terms of fault analyses and explanations. Among the six models mentioned above, only FELM and FBLS integrate ILFR, enabling model interpretability. In contrast, the other four models solely perform classification tasks and cannot intuitively display the specific learning mechanisms or present the acquired knowledge in an understandable manner. This lack of interpretability is one of the major challenges in many existing machine learning-based fault diagnosis methods [10]. However, FELM does not expand the enhancement nodes into a set of EEUs. Compared to FBLS, it can only ensure model performance by setting a large number of nodes, which results in the system containing an excessive number of rules.

By embedding Interpretable Linguistic Fuzzy Rules (ILFR), the experiment fully validated the significant advantages of the FBLS model in terms of interpretability. ILFR not only transforms complex fuzzy logic operations into natural language rules but also intuitively demonstrates the relationship between input features and fault diagnosis results. The fuzzy rules generated by ILFR cover the network fault scenarios involved in the TIS maintenance terminal, as shown in

Table 5. TIS Network Status Alarm Information Table.

Alarm Device	Alarm Sub-Device	Alarm Description	Remarks
Maintenance Console	Connected to TIS-A System Channel	Communication Interruption/Left Network Communication Interruption/Right Network Communication Interruption
Maintenance Console	Connected to TIS-B System Channel	Communication Interruption/Left Network Communication Interruption/Right Network Communication Interruption
Operator Console	Operator Console A/B Machine	Communication Interruption with TIS-A/B System Channel
TIS A/B System	XX Left Network Logical State Abnormality		XX devices include TSRS, RBC, TIS, TCC, CBI, etc.
TIS A/B System	XX Right Network Logical State Abnormality		XX devices include TSRS, RBC, TIS, TCC, CBI, etc.
TIS A/B System	XX-A System Left Network Channel Interruption
TIS A/B System	XX-B System Left Network Channel Interruption
TIS A/B System	XX-A System Right Network Channel Interruption
TIS A/B System	XX-B System Right Network Channel Interruption
TIS A/B System	XX Protocol Version Verification Error
TIS A/B System	XX Data Version Verification Error
Operator Console	Operator Console A/B Machine	Communication Interruption with CTC A/B Machine Channel
Operator Console	Operator Console A/B Machine	CTC Protocol Version Verification Error
TIS A/B System	Communication Board of Frequency Shift Cabinet (1–10) CANA CPU1/CANA CPU2/CANA Communication Fault

, providing clear logical reasoning for maintenance personnel.

In this experiment, each feature is mapped to five fuzzy sets, representing Low, Slight Low, Medium, Slight High, and High.

Table 6. Rule description of fault classification for equipment.

Fuzzified Features
Conditions	X1	X2	X3	X4	X5	X6	X7	X8
	2345	2345	12,345	2345	5	12,345	2345	12,345
	X9	X10	X11	X12	X13	X14	X15	X16
	DC	DC	DC	DC	12,345	45	45	45
Conclusions	Y1: (x_1: Slight Low, x_2: Slight Low, x_13: Low) → Fault in State 1
	Y2: (x_3: Low, x_4: Slight Low, x_15: Slight High) → Fault in State 2
	Y3: (x_5: High, x_6: Low, x_14: Slight High) → Fault in State 3
	Y4: (x_7: Slight Low, x_8: Low, x_16: Slight High) → Fault in State 4

presents the first fuzzy rule, detailing both its conditions and conclusion. The condition section transparently indicates whether each feature is activated and, if so, the corresponding fuzzy mapping. Specifically, “DC” denotes that a feature is not activated, while the numbers 1–5 correspond to different neuro-semantic fuzzy categories. For example, if X₁ is marked as “DC”, it means X₁ is not activated in the first rule; if X₁ is assigned “2345”, it signifies that X₁ is activated and falls under one or more fuzzy conditions, i.e., Slight Low, Medium, Slight High, or High. The conclusion section consists of a four-dimensional linear output, where the condition section explicitly defines the prerequisite conditions for feature activation, and the output vector determines the specific fault type based on which conclusion is met.

In the network fault diagnosis of the TIS system, the network cable fault between TIS-B and Switch A serves as an example to analyze the impact range of the fault and the corresponding rule generation process. Specifically, when the network cable fails, logical channels 5, 6, and 14 simultaneously experience faults. Based on these conditions, ILFR (Interpretable Linguistic Fuzzy Rules) can generate a clear rule to determine the potential fault between TIS-B and Switch A and present this rule in natural language, making it easier for operation and maintenance personnel to understand and act upon. For instance, the generated rule can be expressed as “If logical channels 5, 6, and 14 experience communication interruptions, then there is a fault in the network cable between TIS-B and Switch A”. A visual representation of this rule is shown in Figure 10.

5. Conclusions

This paper proposed a method based on the Fuzzy Broad Learning System (FBLS) to improve the efficiency and accuracy of network fault diagnoses in railway signaling systems. By integrating Interpretable Linguistic Fuzzy Rules (ILFR) with FBLS, the proposed approach enables effective analyses of fault propagation paths in the Train Control and Interlocking Integrated System (TIS) of high-speed railways. Experimental results demonstrated that the proposed method achieves high diagnostic accuracy and strong interpretability, providing effective technical support for the safe operation of railway signaling systems.

In railway networks, real-time fault diagnoses and predictions are critical. Future research should focus on integrating FBLS with real-time data stream processing technologies to develop real-time fault diagnosis and prediction systems. By dynamically predicting fault propagation paths, it will be possible to provide early warnings for potential faults, thereby reducing the impact of faults on system operations and enhancing the safety and reliability of railway systems.