Railway traction substation status monitoring method, system, equipment, and medium

Abstract

The present application relates to a method, system, equipment, and medium for monitoring the status of a railway traction substation. The monitoring method includes: obtaining real-time monitoring data of equipment in the railway traction substation; preprocessing the real-time monitoring data to obtain a preprocessed monitoring data set and performing protocol identification, and converting the heterogeneous data in the monitoring data set into structured data based on a preset protocol template library; extracting the time-frequency domain characteristic parameters of the equipment based on the structured data and constructing a multi-dimensional feature matrix; inputting the multi-dimensional feature matrix into a pre-trained hybrid diagnostic model to generate an equipment health score and a fault probability value; generating an early warning instruction based on the health score and the fault probability value in combination with a dynamic threshold algorithm; and generating a priority maintenance strategy based on the early warning instruction and the equipment maintenance resource constraints through a maintenance strategy optimization model. The present application realizes accurate perception and intelligent decision-making of equipment status in a multi-source heterogeneous data environment.

Description

Technical Field

[0001] The present application relates to the technical field of substation equipment monitoring, and in particular to a method, system, equipment, and medium for monitoring the status of a railway traction substation. Background Art

[0002] Railway traction substations, as a crucial component of the power system, are primarily responsible for providing a stable power supply for railway transportation. With the continuous expansion and modernization of the railway network, the equipment load on traction substations has gradually increased. The reliability and safety of the power system have become core elements for ensuring the normal operation of railway transportation. Substations contain a wide variety of equipment, including main transformers, circuit breakers, switchgear, GIS (gas-insulated switchgear), and distribution units. These devices bear heavy power loads and undergo frequent switching operations. Their long-term stable operation directly impacts the transmission and supply of electricity.

[0003] As the core equipment in a substation, the main transformer is responsible for the efficient conversion of electrical energy. Its stable operating state is crucial to ensuring power supply. Circuit breakers and GIS equipment, on the other hand, are responsible for protecting and controlling circuits, preventing serious damage to the power system from short circuits or faults. Given the complex and ever-changing operating environments of these devices, ensuring their safety and efficiency under various operating conditions has become a pressing technical challenge.

[0004] Currently, equipment monitoring in railway traction substations mostly relies on traditional local monitoring equipment. However, due to the wide variety of equipment and complex monitoring parameters, there is a certain degree of incompatibility in information exchange and processing. Furthermore, fault diagnosis often relies on simple threshold alarms, making it difficult to timely detect changes in equipment health, easily missing optimal maintenance opportunities and causing equipment failures. Therefore, how to achieve accurate perception and intelligent decision-making of traction substation equipment status in a multi-source heterogeneous data environment is an urgent problem that needs to be solved. Summary of the Invention

[0005] In order to achieve accurate perception and intelligent decision-making of equipment status in a multi-source heterogeneous data environment, the present application provides a railway traction substation status monitoring method, system, equipment, and medium.

[0006] In a first aspect, the present application provides a method for monitoring the status of a railway traction substation, which adopts the following technical solution:

[0007] A method for monitoring the status of a railway traction substation, the monitoring method comprising:

[0008] Obtain real-time monitoring data of equipment in railway traction substations;

[0009] Performing data preprocessing on the real-time monitoring data to obtain a preprocessed monitoring data set;

[0010] Performing protocol recognition on the preprocessed monitoring data set, and converting heterogeneous data in the monitoring data set into structured data according to a preset protocol template library;

[0011] Based on the structured data, extracting time-frequency domain feature parameters of the device and constructing a multi-dimensional feature matrix;

[0012] Inputting the multi-dimensional feature matrix into a pre-trained hybrid diagnostic model to generate a device health score and a failure probability value;

[0013] Generate an early warning instruction based on the health score and the failure probability value in combination with a dynamic threshold algorithm; wherein the dynamic threshold algorithm is adaptively adjusted based on the equipment operation data and environmental parameters in the historical database;

[0014] Based on the early warning instructions and equipment maintenance resource constraints, a priority maintenance strategy is generated through a maintenance strategy optimization model; wherein the maintenance strategy optimization model is solved with equipment health score, failure probability and equipment maintenance resources as constraints.

[0015] By adopting the above technical solution, multi-dimensional equipment data of railway traction substations is obtained. After protocol analysis to eliminate data silos, time-frequency domain features are extracted. Algorithmic models are used to implement equipment health scoring and fault probability diagnosis. Combined with the dynamic correction thresholds of environmental parameters, maintenance decisions are ultimately generated, forming a closed-loop logic of "data collection-feature analysis-dynamic warning-intelligent decision-making", thereby effectively improving the real-time performance of equipment status monitoring, the adaptability of dynamic thresholds, and the accuracy of maintenance decisions, realizing the transition from passive alarms to active predictions in the health management of equipment throughout its life cycle.

[0016] Optionally, the step of performing protocol identification on the preprocessed monitoring data set and converting heterogeneous data in the monitoring data set into structured data according to a preset protocol template library includes:

[0017] Build a protocol feature template library, analyze protocol message samples in historical monitoring data, and generate a feature transition probability matrix;

[0018] Receive the preprocessed monitoring data set and intercept the current data frame based on the sliding window;

[0019] Performing similarity matching on the byte transfer feature of the current data frame and the feature transfer probability matrix to determine the protocol type corresponding to the current data frame;

[0020] Performing hierarchical parsing on the current data frame according to the parsing rule corresponding to the protocol type to obtain a hierarchical parsing result;

[0021] The hierarchical analysis results are encapsulated into structured data with semantic tags; wherein the structured data includes device identification, measurement parameter name and standardized dimension value.

[0022] By adopting this technical solution, we achieve automatic recognition and standardized processing of heterogeneous protocol data. This not only adapts to environments where multiple protocols coexist, but also accurately extracts effective information from various data and converts it into structured data in a unified format. By storing the parsed data in a real-time database, the system provides timely and accurate data support for equipment health monitoring and maintenance decision-making.

[0023] Optionally, performing hierarchical parsing on the current data frame according to the parsing rule corresponding to the protocol type to obtain a hierarchical parsing result includes:

[0024] Based on the protocol type, a preset protocol rule library is searched and corresponding physical layer parsing parameters, data link layer parsing parameters and application layer parsing parameters are loaded;

[0025] Receiving a binary data stream of the current data frame;

[0026] Performing checksum calculation on the binary data stream according to the checksum algorithm type included in the physical layer parsing parameter, separating the payload field and generating a physical layer parsing result including a checksum status flag;

[0027] Based on the message structure template defined by the data link layer parsing parameters, performing a field interception operation on the payload field to generate a structured message object containing message metadata;

[0028] Converting the original register value in the payload field into an engineering value according to the register mapping table in the application layer parsing parameter to generate a standard format data object;

[0029] The physical layer parsing result, the structured message object and the standard format data object are metadata-associated to obtain a hierarchical parsing result.

[0030] By employing these technical solutions, we ensure the accuracy and consistency of data from reception to final engineering value conversion. By introducing protocol type queries and a corresponding parsing rule base, the system can flexibly handle multiple protocols, automatically select the correct parsing parameters, and effectively process and extract key information at each parsing level. Furthermore, through metadata association, the parsing results provide detailed traceability information, ensuring data integrity and traceability, and providing a reliable foundation for subsequent data storage, analysis, and decision-making.

[0031] Optionally, the step of extracting time-frequency domain feature parameters of the device based on the structured data and constructing a multi-dimensional feature matrix includes:

[0032] Performing equal-interval resampling processing on the structured data to generate an equal-interval time series data set;

[0033] Extracting time-domain statistical characteristic parameters from the equally spaced time series data set;

[0034] Performing frequency domain conversion processing on the equally spaced time series data set to calculate the main frequency component of the spectrum and the energy proportion of the preset frequency band;

[0035] Obtain device operation record data and analyze device-related characteristic parameters;

[0036] The time domain statistical characteristic parameters, the main frequency component of the spectrum, the preset frequency band energy ratio and the device-related characteristic parameters are aligned in a time window to construct a multi-dimensional characteristic matrix.

[0037] The above technical solution effectively improves the accuracy and comprehensiveness of railway traction substation condition monitoring. Equal-interval resampling ensures data temporal consistency. Time-domain feature extraction provides essential information such as equipment volatility and stability. Frequency-domain analysis reveals the equipment's frequency response and potential harmonic issues. Combined with equipment operation records, the actual operating status of the equipment is extracted, resulting in a comprehensive, multi-dimensional feature matrix.

[0038] Optionally, the method further includes a training step of the hybrid diagnostic model, the training step including:

[0039] Extract historical multi-dimensional feature matrix and historical fault event timestamps from the equipment monitoring database;

[0040] Performing abnormal data cleaning on the historical multidimensional feature matrix;

[0041] Performing time alignment processing on the cleaned historical multidimensional feature matrix and the historical fault event timestamps, and marking the equipment health level;

[0042] Constructing a labeled historical training dataset; the historical training dataset includes a historical multidimensional feature matrix, equipment health level labels, and historical fault event timestamps;

[0043] Based on the historical training data set, a health scoring model and a fault probability prediction model are trained respectively; wherein the health scoring model uses a gradient boosting decision tree algorithm to process the historical multidimensional feature matrix and outputs a device health score; the fault probability prediction model uses a temporal convolutional network to process the historical multidimensional feature matrix within a continuous time window and outputs a fault probability value;

[0044] The trained health score model and fault probability prediction model are jointly verified, and the model parameters are tuned according to the verification results.

[0045] By employing these technical solutions, we transform equipment monitoring data into effective health scores and fault prediction models. Combined with the XGBoost and STCNN algorithms, we ensure the accuracy of equipment status and fault prediction. This model enables the system to diagnose equipment health in real time and provide early warning of faults, significantly improving equipment management efficiency and reducing downtime and maintenance costs.

[0046] Optionally, the step of generating a warning instruction based on the health score and the failure probability value in combination with a dynamic threshold algorithm includes:

[0047] Obtaining the health score, failure probability value, and current environmental parameters of the device;

[0048] Extract the equipment operation data of the same type of equipment under similar environment parameters from the historical database according to the equipment type, and calculate the dynamic compensation factor based on the preset static threshold;

[0049] Based on the dynamic compensation factor, a nonlinear correction is performed on the preset static threshold to generate a dynamic warning threshold under the current environment; wherein the dynamic warning threshold includes a health score threshold and a failure probability threshold;

[0050] Calculating a health score degradation index based on a ratio of the health score to a health score threshold;

[0051] Calculating a failure probability deviation index based on a ratio of the failure probability value to the failure probability threshold;

[0052] The warning level is determined according to the product of the health score degradation index and the failure probability deviation index, and a graded warning instruction is generated.

[0053] By employing the above technical solution and combining it with a dynamic compensation factor to adjust static thresholds, a dynamic warning threshold adapted to current environmental conditions is generated. This process effectively improves the accuracy of equipment failure predictions, enabling timely issuance of warning instructions at different levels, helping maintenance personnel make informed decisions and avoid major equipment failures, thereby improving equipment operational safety and maintenance efficiency. Through dynamic correction, ratio calculation, and warning level classification, the system provides a flexible and real-time equipment monitoring mechanism, ensuring efficient and intelligent equipment management.

[0054] Optionally, the step of generating a priority maintenance strategy through a maintenance strategy optimization model based on the early warning instruction and the equipment maintenance resource constraint condition includes:

[0055] Obtaining a set of equipment warning instructions and equipment maintenance resource constraint data; wherein the warning instruction set includes equipment identification, health score, failure probability value, and warning level; the maintenance resource constraint data includes the number of available technicians, spare parts inventory, and maintenance time window;

[0056] Calculate the equipment maintenance urgency coefficient based on the health score, failure probability value and warning level;

[0057] Constructing an objective function including the maintenance urgency coefficient, and defining a resource consumption vector based on the equipment maintenance resource constraint data;

[0058] Establishing a mixed integer programming model, taking the objective function and the resource consumption vector as input parameters, and introducing the device power grid topology relationship to generate conflict resolution rules;

[0059] Solving the mixed integer programming model to obtain a set of candidate maintenance tasks;

[0060] The candidate maintenance tasks are verified for completeness of spare parts and coordinated with the power outage plan, and a priority maintenance strategy is generated based on the verification results.

[0061] By adopting the above technical solution, combined with a mixed integer programming model, resource constraints, and conflict resolution rules, a rational and efficient priority maintenance strategy can be generated for railway traction substation equipment. The system fully considers multiple factors, including equipment health, failure probability, resource constraints, and grid topology, ensuring that maintenance tasks are optimally scheduled within reasonable resource constraints and time windows. Furthermore, by verifying spare parts completeness and coordinating outage plans, maintenance task conflicts and resource shortages are avoided, ensuring power system stability and efficient equipment maintenance.

[0062] In a second aspect, the present application provides a railway traction substation status monitoring system, which adopts the following technical solutions:

[0063] A railway traction substation status monitoring system, the monitoring system comprising:

[0064] Data acquisition module, used to obtain real-time monitoring data of equipment in railway traction substations;

[0065] A data preprocessing module is used to perform data preprocessing on the real-time monitoring data to obtain a preprocessed monitoring data set;

[0066] A protocol identification and conversion module is used to perform protocol identification on the preprocessed monitoring data set and convert the heterogeneous data in the monitoring data set into structured data according to a preset protocol template library;

[0067] A feature matrix construction module is used to extract the time-frequency domain feature parameters of the device based on the structured data and construct a multi-dimensional feature matrix;

[0068] A fault diagnosis module, configured to input the multi-dimensional feature matrix into a pre-trained hybrid diagnosis model to generate a device health score and a fault probability value;

[0069] a fault warning module, configured to generate a warning instruction based on the health score and the fault probability value in combination with a dynamic threshold algorithm; wherein the dynamic threshold algorithm is adaptively adjusted based on the equipment operation data and environmental parameters in the historical database;

[0070] A maintenance strategy generation module is used to generate a priority maintenance strategy through a maintenance strategy optimization model based on the warning instructions and equipment maintenance resource constraints; wherein the maintenance strategy optimization model is solved with equipment health score, failure probability and equipment maintenance resources as constraints.

[0071] In a third aspect, the present application provides a computer device that adopts the following technical solution:

[0072] A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to the first aspect.

[0073] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0074] A computer-readable storage medium stores a computer program capable of being loaded by a processor and executing any one of the methods in the first aspect.

[0075] In summary, this application achieves at least one of the following beneficial technical effects: Through multi-source heterogeneous data fusion and intelligent analysis, it achieves full-dimensional state perception and accurate diagnosis of traction substation equipment. It also reduces data silos by standardizing protocol template libraries. A hybrid diagnostic model improves the accuracy of critical equipment fault identification. Combining a dynamic threshold algorithm with a resource constraint optimization model improves maintenance response speed, effectively avoiding sudden equipment failures and reducing the frequency of redundant maintenance. Furthermore, it leverages an equipment health scoring system to achieve optimal resource allocation throughout the entire lifecycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 This is a first flow chart of a method for monitoring the status of a railway traction substation according to one embodiment of the present application.

[0077] Figure 2 This is a second flow chart of the railway traction substation status monitoring method according to one of the embodiments of the present application.

[0078] Figure 3 This is a third flow chart of a railway traction substation status monitoring method according to one of the embodiments of the present application.

[0079] Figure 4 This is a fourth flow chart of a railway traction substation status monitoring method according to one embodiment of the present application.

[0080] Figure 5 This is a fifth flow chart of a railway traction substation status monitoring method according to one embodiment of the present application.

[0081] Figure 6 This is a sixth flow chart of a method for monitoring the status of a railway traction substation according to one embodiment of the present application.

[0082] Figure 7 This is the seventh flow chart of the railway traction substation status monitoring method according to one embodiment of the present application. DETAILED DESCRIPTION

[0083] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-7 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0084] The embodiment of the present application discloses a method for monitoring the status of a railway traction substation.

[0085] Reference Figure 1 A method for monitoring the status of a railway traction substation, the monitoring method comprising:

[0086] Step S101, obtaining real-time monitoring data of equipment in a railway traction substation;

[0087] Real-time monitoring data includes transformer core grounding current and dissolved gas concentration in oil, circuit breaker opening and closing time and speed, GIS (gas-insulated switchgear) SF6 gas density and contact temperature parameters, as well as total and resistive currents of lightning arresters, grounding grid down conductor on-resistance, dielectric loss tangent and capacitance change of capacitive equipment (such as bushings and transformers), grid bus voltage and current harmonics, auxiliary equipment status (such as power supply voltage and current and communication link status), and equipment status and statistical parameters (such as switch position signals, protection action alarms, operating hours, and over-limit count statistics). This comprehensive coverage of electrical, mechanical, chemical, and environmental parameters of core traction substation equipment ensures multi-dimensional and high-precision data collection.

[0088] In some embodiments of the present application, a multi-type sensor array is deployed to achieve digital sensing of equipment operating parameters. The core grounding current is collected using a zero-flux current transformer, which is based on the principle of electromagnetic induction and eliminates magnetic saturation effects through closed-loop feedback control. The concentration of dissolved gases in the oil is monitored online using gas chromatography-mass spectrometry (GC-MS). Permeable membrane separation technology is used to extract characteristic gases such as H2, CO, and C2H2 from the transformer oil, and the gas concentration is detected using an electrochemical sensor array. The circuit breaker opening and closing time is measured using a high-precision Hall displacement sensor combined with a travel encoder to measure the motion trajectory of the moving contact, combined with a Kalman filter algorithm to eliminate mechanical vibration noise. SF6 gas density is dynamically monitored using a vibrating wire pressure sensor (based on the principle of resonant frequency-pressure change), and contact temperature is measured non-contact using an infrared thermal imager (8-14μm band).

[0089] Furthermore, the arrester uses a dual-winding microcurrent sensor to achieve full current acquisition and extract the resistive component through a harmonic separation algorithm. The grounding grid on-resistance is injected with an anti-power frequency interference signal through a four-terminal variable frequency bridge and the impedance value is synchronously sampled. The dielectric loss of the capacitive device is captured by the phase measurement unit (PMU) and the phase difference of the capacitor end screen, and the capacitance change is compared in real time through a temperature drift compensation bridge. The bus harmonics use a 16-bit synchronous ADC with a windowed interpolation FFT algorithm to accurately extract characteristic harmonics. The power status is achieved by a Hall voltage sensor and a Rogowski coil to monitor dual power supply parameters, and the communication link is diagnosed through a dual mechanism of physical layer pulse detection and TCP / UDP heartbeat packets. The device status is acquired by a magnetic latching reed relay to collect the switch position signal, and the FPGA hardware counter counts the frequency and duration of limit violations in real time.

[0090] It is understandable that the digital mapping of equipment status is achieved through the above-mentioned multi-dimensional sensor network. Data collection covers the electrical, mechanical, and chemical parameters of the core equipment of the traction substation (main transformer, circuit breaker, GIS, switchgear, distribution device, etc.), providing high-precision raw data for subsequent analysis, solving the problem of single parameters and insufficient accuracy of traditional monitoring systems.

[0091] Step S102, performing data preprocessing on the real-time monitoring data to obtain a preprocessed monitoring data set;

[0092] Among them, in order to address the noise interference, sensor drift and environmental coupling effects in the original data, a multi-level processing mechanism is adopted to achieve data optimization, mainly including missing value compensation, outlier elimination and environmental compensation.

[0093] Specifically, missing value compensation uses an adaptive time series forecasting model (ARIMA) to predict missing period values ​​based on historical data trends. For example, when the oil chromatograph sensor loses data due to communication interruption, the H2 concentration series of the previous 6 hours is used to predict the data for the next 5 minutes, with a prediction error rate of less than 3%. Outlier detection can use the Isolation Forest unsupervised learning algorithm to construct a multidimensional feature space segmentation tree to identify data points that deviate from the main distribution, such as abnormal records where the circuit breaker tripping speed drops instantly to 0. Environmental compensation uses a multivariate regression model to establish a nonlinear mapping relationship between SF6 gas density and temperature and pressure, and fits the compensation formula using the least squares method:

[0094]

[0095] In the above formula, α=0.0035 / ℃ and β=0.12 are empirical coefficients used to ensure the comparability of density measurements under different working conditions.

[0096] Step S103, performing protocol identification on the pre-processed monitoring data set, and converting the heterogeneous data in the monitoring data set into structured data according to a preset protocol template library;

[0097] Among them, protocol types include but are not limited to IEC 61850, 101, 103, 104 and MODBUS standard protocols;

[0098] In one embodiment of this application, a protocol-adaptive mapping engine is constructed to address the issue of incompatible communication protocols across heterogeneous devices. Protocol identification employs a modified Hidden Markov Model (HMM) to train a matrix of characteristic byte transition probabilities for different protocols, enabling real-time classification and recognition of data frames.

[0099] For example, in the characteristic transition probability matrix of the IEC 104 protocol, the probability correlation between the start character 0x68 and the subsequent length bytes reaches 0.92, significantly different from the characteristic distribution of the MODBUS protocol. A layered mapping mechanism can be used in the data conversion phase: the physical layer implements bit stream synchronization and CRC checksum; the data link layer parses the message structure (such as the address code + function code + data field in MODBUS); and the application layer maps the values ​​to a unified data model (such as the CIM / E format), ultimately generating structured data objects with semantic tags.

[0100] Step S104: extracting the time-frequency domain feature parameters of the device based on the structured data and constructing a multi-dimensional feature matrix;

[0101] Specifically, representative feature parameters are extracted from structured data. These features can help identify the operating status of the device and predict faults. The feature extraction process is mainly divided into time domain feature extraction and frequency domain feature extraction. Combined with time and frequency domain analysis, it can comprehensively reflect the dynamic characteristics of the device. Time domain features typically include effective value, maximum value, minimum value, average value, etc., which directly reflect the state of the device at a specific moment. Frequency domain features, on the other hand, are extracted by performing Fourier transform on time series data to obtain spectral features such as main frequency and spectral amplitude. These frequency domain features can reveal periodic behavior or abnormal fluctuations within the device.

[0102] Furthermore, more targeted features can be extracted based on the type of equipment operation. For example, the time deviation rate and speed attenuation coefficient can be extracted from the opening and closing times and speeds of a circuit breaker. These features can help identify issues such as unstable operation and wear. Constructing a feature matrix from this multi-dimensional feature data facilitates subsequent health assessment and fault prediction.

[0103] Step S105: Input the multi-dimensional feature matrix into the pre-trained hybrid diagnosis model to generate the equipment health score and failure probability value;

[0104] A pre-built hybrid diagnostic model consists of a device health scoring model and a failure probability prediction model. The health scoring model assesses the device's health level based on various indicators (such as operating status, load, and temperature), while the failure probability prediction model calculates the probability of device failure based on factors such as the device's operating environment and historical failure data. By inputting a feature matrix, the model outputs a real-time health score and failure probability, providing a basis for device management and maintenance decisions.

[0105] In one embodiment of this application, the health scoring model can utilize the XGBoost ensemble learning algorithm, taking a feature matrix as input and outputting a health index score of 0-100 through weighted voting by multiple regression trees. The fault probability prediction model can utilize a spatiotemporal convolutional network (STCNN), fusing current features with historical state sequences (with a 72-hour sliding window) to output the probability of fault occurrence within the next 24 hours.

[0106] Step S106: Generate an early warning instruction based on the health score and the failure probability value in combination with a dynamic threshold algorithm;

[0107] Among them, the dynamic threshold algorithm is adaptively adjusted based on the equipment operation data and environmental parameters in the historical database;

[0108] Specifically, based on the device's health score and failure probability, a dynamic threshold algorithm is combined to automatically generate warning instructions. The core of the dynamic threshold algorithm is to adaptively adjust the warning threshold based on the device type, environmental parameters (such as temperature, humidity, load, etc.), and historical data. For example, under high load or extreme temperature conditions, the device's fault tolerance may decrease. In this case, the warning threshold needs to be appropriately adjusted to ensure that the system can issue timely warnings despite environmental changes, avoiding premature or delayed warnings.

[0109] Furthermore, the system determines whether a set real-time threshold has been exceeded based on the real-time health score and the probability of failure. If so, an early warning is triggered, prompting personnel to conduct inspections and maintenance. The introduction of dynamic thresholds allows the early warning system to more flexibly adapt to varying operating environments and equipment conditions, avoiding the potential for false positives or missed negatives associated with traditional static thresholds and improving the accuracy and reliability of equipment failure predictions.

[0110] Step S107: Based on the early warning instructions and the equipment maintenance resource constraints, a priority maintenance strategy is generated through a maintenance strategy optimization model;

[0111] Among them, the maintenance strategy optimization model is solved with equipment health score, failure probability and equipment maintenance resources as constraints.

[0112] Specifically, a maintenance strategy optimization model utilizes equipment health status and failure prediction information, combined with maintenance resource constraints (such as maintenance personnel, spare parts, and time), to generate an optimal maintenance strategy. This model prioritizes equipment maintenance by weighing equipment health scores, failure probabilities, and the availability of maintenance resources. Given limited resources, it ensures that equipment most in need of maintenance receives priority. Optimization models are typically solved using integer programming or other operations research algorithms to generate optimal maintenance plans, minimizing resource waste and improving maintenance efficiency.

[0113] In the above implementation, multi-dimensional equipment data of the railway traction substation is acquired, and after data silos are eliminated through protocol parsing, time-frequency domain features are extracted. An algorithm model is used to implement equipment health scoring and fault probability diagnosis. Combined with the dynamic correction threshold of environmental parameters, maintenance decisions are ultimately generated, forming a closed-loop logic of "data collection-feature analysis-dynamic early warning-intelligent decision-making", thereby effectively improving the real-time performance of equipment status monitoring, the adaptability of dynamic thresholds, and the accuracy of maintenance decisions, and realizing the transition from passive alarm to active prediction in the health management of equipment throughout its life cycle.

[0114] Reference Figure 2As an implementation of step S103, performing protocol identification on the pre-processed monitoring data set and converting heterogeneous data in the monitoring data set into structured data according to a preset protocol template library includes the following steps:

[0115] Step S201: construct a protocol feature template library, analyze protocol message samples in historical monitoring data, and generate a feature transition probability matrix;

[0116] By analyzing protocol message samples from historical monitoring data, a protocol feature template library is constructed. The protocol feature template library contains characteristic transition probability matrices for multiple communication protocols (such as IEC 61850, IEC 101, IEC 103, IEC 104, and MODBUS).

[0117] In one embodiment of the present application, a characteristic transition probability matrix is ​​constructed by statistically analyzing byte sequences in protocol message samples and calculating the transition probabilities between each pair of adjacent bytes, thereby generating a mathematical model that describes the flow patterns of protocol bytes. For example, the transition probability P(i, j) from byte i to byte j in a particular protocol represents the proportion of the number of times byte i is followed by byte j in the sample data to the total number of times byte i appears. In this way, a characteristic matrix can be created for each protocol, reflecting the order and relationship characteristics of bytes in different protocols.

[0118] For example, the IEC 61850 protocol may frequently use specific byte sequences, such as the start character 0x68, while the MODBUS protocol may use 0x01 as a common start byte. By counting these features, a reliable byte transition probability matrix can be generated for different protocols in the protocol template library.

[0119] Step S202, receiving the pre-processed monitoring data set, and intercepting the current data frame based on the sliding window;

[0120] The system receives a pre-processed monitoring data set and intercepts the current data frame using a sliding window technique. Sliding windows are a data processing technique used to incrementally process fixed-size blocks of data within a data stream. During this step, the data frame is divided according to the set window size for further analysis. This sliding window allows the system to process continuous data within the data stream and dynamically capture the latest data, ensuring that every piece of data in the data stream is processed promptly, thereby improving the timeliness and accuracy of the system's data processing.

[0121] Step S203, performing similarity matching between the byte transfer feature of the current data frame and the feature transfer probability matrix to determine the protocol type corresponding to the current data frame;

[0122] The system calculates the byte transition feature matrix of the current data frame and compares it to the feature transition probability matrix in the protocol feature template library. To identify the protocol, the byte transition feature matrix of the current data frame is obtained using a sliding window method and compared with the transition probability matrix for each protocol. The difference between each protocol matrix and the matrix of the current data frame is calculated to select the most similar protocol type. If the similarity value of the current data frame exceeds a set threshold, the data frame is classified as belonging to that protocol type; otherwise, it is marked as "unknown protocol."

[0123] Specifically, the similarity matching steps include:

[0124] Calculate the real-time byte transfer matrix W(i,j) for the current data frame; traverse all protocol feature matrices P_k(i,j) in the protocol feature template library; calculate the difference value S_k=Σ|W(i,j)-P_k(i,j)| between the real-time byte transfer matrix W(i,j) and each protocol matrix P_k(i,j); select the protocol type that minimizes the S_k value as the recognition result, and mark it as an unknown protocol when S_k exceeds the preset threshold.

[0125] Step S204, performing hierarchical parsing on the current data frame according to the parsing rules corresponding to the protocol type to obtain a hierarchical parsing result;

[0126] The layered parsing process involves protocol analysis at multiple levels, primarily the physical layer, data link layer, and application layer. First, the physical layer recovers the original bit stream based on the protocol type and communication medium, and verifies data integrity using checksum algorithms (such as CRC). The data link layer then splits the data packet into multiple fields, such as the address field, function code, and data payload field, according to the protocol specification. Finally, the application layer converts the original register values ​​or data into specific engineering physical quantities, such as temperature, current, and voltage, and performs standardization.

[0127] For example, for the MODBUS protocol, the physical layer may involve bit stream reassembly and CRC checking, the data link layer will extract the MODBUS address and function code fields, and the application layer will convert register data (such as temperature values) into standard engineering units, such as degrees Celsius.

[0128] It is understandable that through layered parsing, the system can gradually and deeply extract valid data from each layer, ensuring that data is correctly parsed from the physical layer to the application layer. This process ensures that protocol data can be accurately converted into data in a unified standard format, adapting to data fusion and processing in a multi-protocol environment.

[0129] Step S205: Encapsulate the hierarchical parsing results into structured data with semantic tags.

[0130] Among them, structured data includes equipment identification, measurement parameter name and standardized dimension value.

[0131] Specifically, the system converts the results of hierarchical parsing into structured data objects with semantic tags. The device identifier uniquely identifies the device, the measurement parameter name represents the physical meaning of the data, and the standardized dimension value is the engineering unit value after unified conversion.

[0132] It's understandable that structured data encapsulation can transform complex protocol data into a standardized format that's easy to understand and process, making subsequent data storage, querying, and analysis more efficient and systematic. By adding semantic tags to each data item, the system can clearly represent the data's meaning and facilitate subsequent data utilization.

[0133] The above implementation achieves automatic recognition and standardized processing of heterogeneous protocol data, not only adapting to environments where multiple protocols coexist, but also accurately extracting effective information from various data and converting it into structured data in a unified format. By storing the parsed data in a real-time database, the system can provide timely and accurate data support for equipment health monitoring and maintenance decision-making.

[0134] Reference Figure 3 As an implementation of step S204, performing hierarchical parsing on the current data frame according to the parsing rules corresponding to the protocol type to obtain the hierarchical parsing result includes:

[0135] Step S301: query the preset protocol rule library based on the protocol type and load the corresponding physical layer parsing parameters, data link layer parsing parameters and application layer parsing parameters;

[0136] The system queries a pre-configured protocol rule base based on the protocol type (e.g., MODBUS, IEC 61850, etc.). This rule base stores parsing rules for various protocols, including parameters related to physical layer parsing, data link layer parsing, and application layer parsing. Each protocol has a corresponding set of parsing parameters that guide how to parse each component of a data frame. The protocol type identifier (e.g., MODBUS, IEC 61850, etc.) serves as the query key, helping the system select the appropriate parsing parameter set.

[0137] Step S302, receiving a binary data stream of the current data frame;

[0138] Binary data streams are typically generated by power equipment or sensors, containing real-time monitoring data from the equipment. Transmitted in binary format, these streams include physical layer data, data link layer information, and application layer payloads. In subsequent steps, the binary data streams are parsed layer by layer according to protocol rules.

[0139] Step S303 , performing checksum calculation on the binary data stream according to the checksum algorithm type included in the physical layer parsing parameter, separating the payload field and generating a physical layer parsing result including a checksum status flag;

[0140] The system calculates a checksum on the received binary data stream based on the physical layer parsing parameters defined in the protocol. This checksum calculation is based on the protocol-specified checksum algorithm, such as CRC16_MODBUS or parity. The checksum algorithm verifies the integrity and accuracy of the data, ensuring that the data frame has not been corrupted during transmission. If the checksum fails, the frame may contain erroneous or corrupted data, and subsequent processing will be interrupted.

[0141] The system then separates the payload fields from the successfully verified data frames. These payload fields contain the actual monitoring data or control commands. Finally, the physical layer analysis results will include a verification status flag indicating the success or failure of the verification.

[0142] Step S304: Based on the message structure template defined by the data link layer parsing parameters, perform a field interception operation on the payload field to generate a structured message object containing message metadata;

[0143] The system intercepts and parses the payload field by field based on the message structure template defined by the data link layer parsing parameters. The data link layer parsing parameters contain the rules for how to allocate and organize data in the protocol, such as the start bit, address field, and function code field of the packet.

[0144] Furthermore, the system extracts relevant fields from the payload according to these rules and generates a structured message object. This structured message object contains not only the device's communication data, but also metadata related to this data, such as the data source, device identification, protocol version, and other information.

[0145] Step S305 , converting the original register value in the payload field into an engineering value according to the register mapping table in the application layer parsing parameter to generate a standard format data object;

[0146] The system converts the original register value contained in the payload field into an engineering value according to the register mapping relationship table in the application layer parsing parameter.

[0147] In this embodiment of the present application, the register mapping table contains the relationship between register addresses and measurement units and defines how to convert register values ​​into actual physical quantities (such as temperature and current). The system uses conversion formulas (such as linear transformations) to calculate the raw register values ​​and derive standardized engineering values. The converted data contains the actual measurement data of the device, such as temperature and voltage, converted into unified engineering units.

[0148] Step S306 , performing metadata association on the physical layer parsing result, the structured message object, and the standard format data object to obtain a hierarchical parsing result.

[0149] Specifically, the system associates metadata between the physical layer parsing results, structured message objects, and standard format data objects to generate a complete layered parsing result. Each parsing result includes associated metadata, such as verification status, protocol type, timestamp, device identifier, and data source. This metadata association allows the parsed data to be traced back to the original data source, ensuring data integrity and traceability.

[0150] The above implementation ensures data accuracy and consistency from reception to final engineering value conversion. By introducing protocol type queries and a corresponding parsing rule base, the system can flexibly handle multiple protocols, automatically select the correct parsing parameters, and effectively process and extract key information at each parsing level. Furthermore, through metadata association, the parsing results provide detailed traceability information, ensuring data integrity and traceability, and providing a reliable foundation for subsequent data storage, analysis, and decision-making.

[0151] Reference Figure 4 As an implementation of step S104, the steps of extracting the time-frequency domain feature parameters of the device based on the structured data and constructing a multi-dimensional feature matrix include:

[0152] Step S401, performing equal-interval resampling processing on the structured data to generate an equal-interval time series data set;

[0153] Specifically, in the condition monitoring of power equipment, sensors typically collect various operational data from the equipment. This data may be unevenly sampled, meaning the intervals between data timestamps are not constant, which can affect the accuracy of subsequent feature extraction and analysis. The goal of equal-interval resampling is to convert the original time series data into a dataset with uniform and equal time intervals, ensuring that all data points are located at fixed time points for systematic and continuous analysis. This processing method typically requires setting the sampling interval based on the temporal resolution of the data (for example, per second, per minute, or per hour).

[0154] In one embodiment of the present application, the process of equally spaced resampling can be implemented by an interpolation algorithm. Common interpolation methods include linear interpolation, cubic spline interpolation, etc. Among them, the cubic spline interpolation algorithm is widely used in the resampling process. It can generate a smooth curve between the interpolation points, adapt to the changing trend of the signal, thereby filling the missing values ​​in the original data and ensuring the smoothness of the data. Cubic spline interpolation can generate more accurate data than linear interpolation, especially when the signal changes greatly, and can effectively avoid the generation of interpolation errors.

[0155] Step S402, extracting time domain statistical feature parameters from the equally spaced time series data set;

[0156] Time-domain feature extraction can reveal fundamental signal properties in the time domain, such as amplitude variation and volatility. In power equipment monitoring, time-domain feature extraction helps us understand the stability of equipment operation and provides data for further fault diagnosis. Time-domain statistical feature parameters include mean, standard deviation, peak-to-peak value, crest factor, RMS value, and pulse factor.

[0157] It's understandable that extracting time-domain features provides preliminary signal characteristics for device health assessment, helping analyze signal stability and changing trends. In power equipment, time-domain features are often closely related to normal operation or potential failures. For example, signals with high standard deviations often indicate a risk of overload or damage.

[0158] Step S403, performing frequency domain conversion processing on the equally spaced time series data set to calculate the main frequency component of the spectrum and the energy proportion of the preset frequency band;

[0159] Frequency domain analysis is an important complement to time domain analysis. It reveals the frequency components of a signal by converting it into a frequency domain signal. In power equipment monitoring, frequency domain analysis can identify the equipment's vibration characteristics, frequency response, and other possible frequency anomalies.

[0160] In an embodiment of the present application, a fast Fourier transform (FFT) can be used to convert a time signal into a frequency domain signal and output spectrum amplitude distribution data. The core of the FFT is to mathematically decompose a continuous time signal into the sum of different frequency components. The amplitude of each frequency component can reveal the energy intensity of the signal at that frequency. For example, the vibration signal of a transformer may have a strong frequency component at 50Hz, while the energy in the high-frequency band is weak, which indicates that the operating frequency of the device is 50Hz, while the higher frequency part may be noise in the system or an abnormality of the device.

[0161] Specifically, spectrum amplitude distribution data reflects the energy distribution of a signal at different frequencies, allowing the determination of a device's operating frequency, vibration frequency, or harmonic frequency. Typically, the system pre-defines a characteristic frequency band based on the specific device type and calculates the energy percentage within that band to help identify the device's response characteristics in that specific frequency band and further analyze its operating status.

[0162] For example, for a power transformer, spectrum analysis may show a significant energy peak at 50 Hz, which is consistent with the operating frequency of the equipment, while energy peaks at 100 Hz and its higher multiples may indicate harmonic problems or mechanical vibration of the equipment.

[0163] Step S404, obtaining device operation record data and parsing device-related characteristic parameters;

[0164] Equipment operation records typically include key information such as the time it takes for a device to open and close, current fluctuations, and changes in gas concentration. By analyzing the equipment's operation records, we can extract relevant characteristic parameters, such as the deviation rate of opening and closing times and the rate of change in gas concentration.

[0165] Specifically, the opening and closing time deviation rate refers to the degree of deviation between the actual opening and closing time and the rated standard time. This parameter reflects the operational stability of the equipment. If the deviation rate is too large, it may indicate a problem with the switching operation or equipment aging. The gas concentration change rate is often used to monitor dissolved gases in transformers. It can reveal changes in the gas concentration in the transformer oil and thus indicate whether the equipment has internal faults. Drastic changes in gas concentration often indicate problems such as arcing, overheating, or short circuits within the equipment.

[0166] Step S405 , performing time window alignment processing on the time domain statistical characteristic parameters, the main frequency component of the spectrum, the energy proportion of the preset frequency band and the device-related characteristic parameters to construct a multi-dimensional characteristic matrix.

[0167] The system uses time window alignment technology to combine multiple feature parameters into a multidimensional feature matrix. Time window alignment aligns the various feature parameters (time domain features, frequency domain features, and device features) according to their timestamps, ensuring they match at the same point in time. The size and overlap of the time windows can be adjusted based on the device's operating cycle, sampling frequency, and analysis requirements. The aligned data can be used for multi-dimensional analysis, helping the system comprehensively assess the device's operating status.

[0168] The above implementation effectively improves the accuracy and comprehensiveness of railway traction substation condition monitoring. Equal-interval resampling ensures data temporal consistency; time-domain feature extraction provides essential information such as equipment volatility and stability; frequency-domain analysis reveals the equipment's frequency response and potential harmonic issues; and, combined with equipment operation records, extracts the equipment's actual operating status, ultimately constructing a comprehensive, multi-dimensional feature matrix.

[0169] Reference Figure 5 As a further embodiment of the monitoring method, a training step of the hybrid diagnostic model is also included, and the training step includes:

[0170] Step S501, extracting a historical multi-dimensional feature matrix and historical fault event timestamps from a device monitoring database;

[0171] The multidimensional feature matrix contains the equipment's operating status parameters at multiple time points. These parameters typically include various sensor data. The multidimensional feature matrix includes parameters such as traction transformer core current, oil gas concentration, and circuit breaker opening and closing times. These features provide rich information for subsequent health scoring and fault prediction.

[0172] Specifically, the core current of a traction transformer reflects the current load of the equipment and can be used to reveal the electrical load conditions during transformer operation. The gas concentration in the oil often indicates whether there are internal faults in the transformer; changes in the gas content are an early sign of equipment failure. The opening and closing time parameters of circuit breakers indicate the switching characteristics of the equipment; any abnormal switching behavior may lead to equipment failure.

[0173] Additionally, the fault event timestamp is a key piece of information used to record device failures, typically including the exact time the failure occurred. These timestamps serve as fault labels, helping to correlate historical device characteristics with actual failure events. This is fundamental to training efficient and reliable fault prediction models. Fault timestamps provide a time slice of device status information, helping the model understand changes in device status before and after a failure.

[0174] Step S502, cleaning abnormal data from the historical multidimensional feature matrix;

[0175] Among them, the steps of abnormal data cleaning include the following two aspects:

[0176] Eliminate invalid or outlier data: For example, when detecting extreme values, eliminate values ​​that far exceed the normal operating range of the equipment. For example, if the traction transformer core current is detected to be 10,000A, which clearly exceeds the maximum load range of the equipment, such data should be considered abnormal and eliminated.

[0177] Data removal during sensor failure periods: Sensors may fail for a period of time, resulting in distorted data. Typically, we use equipment maintenance logs to identify time periods with sensor failures and remove any data collected during these periods. This step is often used in conjunction with maintenance logs or abnormal alarm records to ensure that model training is not affected by this data.

[0178] Step S503: Time-align the cleaned historical multidimensional feature matrix with the historical fault event timestamps, and mark the equipment health level;

[0179] The goal of time alignment is to ensure that the historical data of each device is accurately aligned with the corresponding fault event label in the time dimension. The status characteristics of the device are often time-dependent, so it is necessary to match the feature data and the fault timestamp to ensure that they are analyzed within the same time window.

[0180] Specifically, based on the time of the equipment failure, the device's status characteristic data before the failure is extracted and the data at each time point is annotated with a health level. These health levels are typically assigned by human experts based on the equipment's long-term operating performance or failure history, and are categorized as I, II, or III. The health level reflects the equipment's operating status at the time of the failure and is often determined based on a comprehensive assessment of multiple monitoring parameters, such as temperature, load, and vibration.

[0181] Step S504, constructing a labeled historical training data set;

[0182] The historical training dataset includes historical multi-dimensional feature matrices, equipment health level annotations, and historical fault event timestamps;

[0183] After time alignment and labeling, the resulting labeled dataset becomes the foundation for model training. Each data sample includes multi-dimensional device features, failure timestamps, and device health levels, capturing the device's historical performance in different health states and the timeline of device failures. Combining these historical features with labels provides rich information for model learning.

[0184] Step S505: training a health score model and a fault probability prediction model based on the historical training data set;

[0185] The health score model uses a gradient boosting decision tree algorithm to process the historical multidimensional feature matrix and output the equipment health score. The fault probability prediction model uses a time series convolutional network to process the historical multidimensional feature matrix within a continuous time window and output the fault probability value.

[0186] As one implementation of the health scoring model, the XGBoost (Extreme Gradient Boosting) algorithm can be used for training. XGBoost is a powerful gradient boosting tree algorithm that improves prediction accuracy by integrating multiple decision trees. For the health scoring model, XGBoost gradually builds multiple regression trees to learn the relationship between device features and health levels.

[0187] During the training configuration process, the XGBoost algorithm allows different weights to be assigned to different features to reflect their importance in model training. In particular, in equipment monitoring, certain key parameters (such as hydrogen concentration and core current) are more valuable for fault prediction. Therefore, the model's sensitivity to these features can be increased by adjusting their weights. Furthermore, a cross-validation strategy is used to optimize model hyperparameters, such as tree depth (max_depth) and learning rate (learning_rate). Cross-validation effectively reduces model overfitting and ensures good generalization. Training is terminated when the mean absolute error of the validation set fails to improve within a preset threshold of consecutive iterations.

[0188] As an implementation of the fault probability prediction model, the fault probability prediction model is trained using a spatiotemporal convolutional neural network (STCNN). STCNN combines convolutional neural networks (CNN) and time series modeling, and can process the device's time series data and extract spatiotemporal features.

[0189] During the training configuration process, to accurately model the historical behavior of the device, the historical multidimensional feature matrix is ​​slidingly sliced ​​into 72-hour time windows to generate a temporally correlated input data sequence. A composite neural network architecture is constructed, consisting of a one-dimensional convolutional layer and a long short-term memory network layer. The one-dimensional convolutional layer has 64 filters and a kernel size of 3. Sample weights are applied to the feature matrix sequence before the fault event, with the weight value being five times that of normal samples. Furthermore, the cosine annealing algorithm is used to dynamically adjust the learning rate, typically starting training at a high learning rate and then gradually reducing it to the minimum value. This helps stabilize the model during training and avoids oscillations during gradient descent.

[0190] Understandably, in an imbalanced dataset, the ratio of positive to negative samples can be very different. Since the probability of a failure occurring is relatively low, a higher weight can be placed on positive samples (i.e., data before a failure occurs), thus alleviating the class imbalance problem.

[0191] Step S506 : jointly verify the trained health score model and fault probability prediction model, and optimize the model parameters according to the verification results.

[0192] Model validation is a crucial step in ensuring the model's effectiveness in practical applications. During the health score model validation process, the model's prediction accuracy is primarily assessed by calculating the MAE and R² values. Lower MAE and higher R² values ​​indicate that the model better captures the health status of the device. The failure probability prediction model is validated by plotting the ROC curve and calculating the AUC value to assess its classification performance. Setting thresholds to calculate precision and recall further evaluates the accuracy of the fault prediction model.

[0193] In this implementation, device monitoring data is converted into an effective health score and fault prediction model, combined with the XGBoost and STCNN algorithms to ensure the accuracy of device status and fault prediction. This model enables the system to diagnose device health in real time and provide early warning of faults, significantly improving device management efficiency and reducing downtime and maintenance costs.

[0194] Reference Figure 6 As an implementation of step S106, the step of generating a warning instruction based on the health score and the failure probability value in combination with the dynamic threshold algorithm includes:

[0195] Step S601: Obtain the health score, failure probability value, and current environmental parameters of the device;

[0196] The health score represents the current overall health of the device, taking into account multiple operating parameters (such as temperature, humidity, and load). The failure probability value indicates the likelihood of a device failure within a certain period of time in the future.

[0197] Furthermore, environmental parameters are crucial to equipment operation, especially for complex railway traction substations. Environmental parameters such as temperature, humidity, and load factor directly impact equipment performance. Temperature fluctuations can accelerate the aging of electrical and mechanical components, while humidity fluctuations can cause internal corrosion or electrical failures. Load factor determines the load-bearing status of equipment. Real-time collection of these environmental parameters provides essential input data for dynamic early warning.

[0198] Step S602 , extracting device operation data of the same type of devices under similar environments with current environmental parameters from a historical database based on the device type, and calculating a dynamic compensation factor based on a preset static threshold;

[0199] Equipment types typically include different types of power equipment, such as transformers, circuit breakers, and traction motors. Each device has distinct operating characteristics and failure modes. To make the early warning system more accurate, it is necessary to extract historical operating data from a database based on the equipment type and for equipment operating under similar environmental conditions. For example, historical operating data for equipment under similar conditions can be filtered based on parameters such as the current ambient temperature, humidity, and load factor. This data will help the system better understand the expected performance of the equipment in the current environment.

[0200] Furthermore, based on this historical data, the system can calculate a dynamic compensation factor to modify the original static thresholds (such as health score thresholds and failure probability thresholds) to adapt them to the current environmental conditions. Static thresholds are usually defined under standard conditions, but equipment performance may vary under different environmental conditions, so dynamic compensation factors are needed to adjust them nonlinearly.

[0201] It's understandable that the introduction of a dynamic compensation factor allows the system to automatically adjust the device's fault warning criteria based on the device's environment, improving both the accuracy and sensitivity of the warning. Without the dynamic compensation factor, static thresholds might prevent the system from providing accurate warnings in inappropriate environments.

[0202] Step S603, performing nonlinear correction on the preset static threshold based on the dynamic compensation factor to generate a dynamic warning threshold under the current environment;

[0203] Among them, the dynamic warning threshold includes the health score threshold and the failure probability threshold;

[0204] Specifically, the dynamic warning threshold is a nonlinear correction of the original static threshold using a dynamic compensation factor calculated based on historical data. The core of this process is to adjust the health score threshold and failure probability threshold based on environmental factors to adapt to the device's current operating environment. The dynamic compensation factor can appropriately adjust the static threshold upward or downward by reflecting the impact of environmental changes on device performance.

[0205] The nonlinear correction process involves nonlinearly adjusting the original static thresholds based on changes in factors such as temperature, humidity, and load. For example, temperature changes may have an exponential impact on the health score of a device, while load changes may produce a linear change in the probability of failure.

[0206] For example, if the dynamic compensation factor, calculated based on historical data, indicates a 5°C temperature increase, the health score threshold may increase by 10 points. Meanwhile, if the load exceeds 80% of the rated value, the failure probability threshold increases linearly in proportion. Ultimately, both the health score threshold and the failure probability threshold are adjusted based on the current environment.

[0207] Step S604: Calculate the health score degradation index based on the ratio of the health score to the health score threshold;

[0208] The health score degradation index is calculated by comparing the real-time health score to the dynamic health score threshold. This index reflects the degree of change in the device's current health status. The calculation formula is: Health score degradation = health score / health score threshold. When the health score degradation approaches 1, the device's health is good and has not yet reached the warning threshold. When the health score degradation approaches 0, the device's health has significantly deteriorated and may face the risk of failure.

[0209] Step S605: Calculate the failure probability deviation index based on the ratio of the failure probability value to the failure probability threshold;

[0210] The Failure Probability Deviation metric assesses the likelihood of equipment failure by calculating the ratio of the real-time failure probability value to the dynamic failure probability threshold. The calculation formula is: Failure Probability Deviation = Failure Probability Threshold / Failure Probability Value. A higher Failure Probability Deviation metric indicates a higher likelihood of equipment failure, and the system should prioritize maintenance for this equipment.

[0211] Step S606: determine the warning level according to the product of the health score degradation index and the failure probability deviation index, and generate a graded warning instruction.

[0212] By multiplying the health score degradation index and the failure probability deviation index, we can comprehensively assess the health and failure risk of the equipment, thereby generating an early warning level. The calculation formula is: Early Warning Level = Health Score Degradation × Failure Probability Deviation.

[0213] It is understandable that the range of warning levels is usually divided into multiple levels, such as "immediate maintenance", "planned maintenance" and "enhanced monitoring". The larger the value, the closer the equipment is to failure and the more urgent the treatment is.

[0214] In the above implementation, static thresholds are adjusted using dynamic compensation factors to generate dynamic warning thresholds tailored to current environmental conditions. This process effectively improves the accuracy of equipment failure predictions, enabling timely issuance of warning instructions at varying levels, helping maintenance personnel make informed decisions and avoid major equipment failures, thereby improving equipment operational safety and maintenance efficiency. Through dynamic correction, ratio calculation, and warning level grading, the system provides a flexible and real-time equipment monitoring mechanism, ensuring efficient and intelligent equipment management.

[0215] Reference Figure 7As an implementation method of step S107, the step of generating a priority maintenance strategy through a maintenance strategy optimization model based on the early warning instruction and the equipment maintenance resource constraint condition includes:

[0216] Step S701: Acquire equipment warning instruction set and equipment maintenance resource constraint data;

[0217] The early warning instruction set includes equipment identification, health score, failure probability value and early warning level, and maintenance resource constraint data includes the number of available technicians, spare parts inventory and maintenance time window;

[0218] Specifically, each device's health score represents its current health, the failure probability indicates the likelihood of future failures, and the warning level is assessed based on the health score and failure probability, reflecting the urgency of equipment maintenance. The number of available technicians refers to the total number of technicians currently available, the spare parts inventory refers to the number of various spare parts currently in stock, and the maintenance window is the time period or total time during which maintenance can be performed.

[0219] Step S702: Calculate the equipment maintenance urgency coefficient based on the health score, failure probability value, and warning level;

[0220] The health score and failure probability are normalized to unify the measurement scale and ensure model consistency. The warning level is then mapped to a weighted factor ranging from 1 to 3 based on the device's urgency, with 1 representing low priority and 3 representing high priority. A comprehensive urgency coefficient is generated by multiplying the normalized health score with the failure probability and then by the warning level weighting factor. For example, assuming a device's health score is 70, normalized to 0.7, the failure probability is 0.2, and the warning level is 2 (corresponding to a weighting factor of 2), the maintenance urgency coefficient is 0.7 × 0.2 × 2 = 0.28.

[0221] Step S703: constructing an objective function including a maintenance urgency coefficient, and defining a resource consumption vector based on equipment maintenance resource constraint data;

[0222] The maintenance urgency coefficient is used as part of the objective function to maximize the equipment's maintenance priority. Furthermore, the model also considers maintenance resource consumption, including technician work hours, spare parts consumption, and required maintenance time.

[0223] Specifically, the objective function is constructed based on the maintenance urgency coefficient, aiming to maximize the overall urgency coefficient, thereby ensuring that the most urgent equipment maintenance tasks are prioritized. The resource consumption vector describes the technician hours, spare parts quantity, and time required for each maintenance task. This vector is calculated based on the equipment's maintenance manual, historical maintenance records, and environmental complexity.

[0224] Step S704: Establish a mixed integer programming model, take the objective function and the resource consumption vector as input parameters, and introduce the device power grid topology relationship to generate conflict resolution rules;

[0225] The mixed integer programming (MIP) model is the core tool for solving optimization problems, aiming to find the optimal equipment maintenance strategy. The model includes the following elements:

[0226] Objective function: maximize the sum of maintenance urgency coefficients, that is, maximize the total urgency of equipment maintenance tasks with higher priority;

[0227] Constraints: Constrain the resource usage in the resource consumption vector to not exceed the available resources, that is, the technician hours, spare parts inventory and maintenance time window cannot exceed the actual resource limits;

[0228] Conflict resolution rules: Considering the interdependence of different devices on the power grid, especially those in power supply zones, the system needs to generate conflict resolution rules to prevent maintenance of devices in the same power supply zone from occurring simultaneously, thereby preventing any impact on power supply stability. Furthermore, a maintenance interval threshold for primary and backup devices is set, forcing maintenance on the backup device to wait 24 hours after primary device maintenance is complete before allowing maintenance on the backup device.

[0229] For example, device A and device B are located in the same power supply zone. If the maintenance of device A is carried out during a certain period, the maintenance of device B must be postponed to avoid power supply interruption. The system can calculate the optimal maintenance period arrangement through the MIP model.

[0230] Step S705, solving the mixed integer programming model to obtain a set of candidate maintenance tasks;

[0231] Specifically, after establishing a mixed integer programming model, an appropriate solution algorithm can be used to obtain the optimal set of maintenance tasks. When the number of devices is small, a branch-and-bound algorithm is used for precise solution to ensure the optimal solution. This algorithm gradually eliminates solutions that do not meet the constraints and ultimately finds the optimal solution. When the number of devices is large, a genetic algorithm can be used to generate a near-optimal solution. By simulating the process of natural selection, genetic algorithms use fitness evaluation, crossover, and mutation operations to quickly obtain a near-optimal solution.

[0232] Step S706: Verify the completeness of spare parts and the coordinated verification of the power outage plan for the candidate maintenance task, and generate a priority maintenance strategy based on the verification results.

[0233] After obtaining a set of candidate maintenance tasks, the system needs to perform spare parts completeness verification and outage plan coordination verification. Spare parts completeness verification is used to check whether the spare parts required for each maintenance task are sufficient in inventory. If the inventory of a spare part is insufficient, the tasks that depend on that spare part must be removed from the candidate maintenance task set. Outage plan coordination verification is used to ensure that the maintenance task is coordinated with the power system's outage plan. If there is a conflict between equipment maintenance and the outage plan, the system needs to adjust the maintenance window or postpone the maintenance task.

[0234] The above implementation combines a mixed-integer programming model, resource constraints, and conflict resolution rules to generate a rational and efficient priority maintenance strategy for railway traction substation equipment. This system fully considers multiple factors, including equipment health, failure probability, resource constraints, and grid topology, ensuring that maintenance tasks are optimally scheduled within reasonable resource constraints and time windows. Furthermore, by verifying spare parts completeness and coordinating outage plans, maintenance conflicts and resource shortages are avoided, ensuring power system stability and efficient equipment maintenance.

[0235] The embodiment of the present application also discloses a railway traction substation status monitoring system.

[0236] A railway traction substation status monitoring system, the monitoring system comprising:

[0237] Data acquisition module, used to obtain real-time monitoring data of equipment in railway traction substations;

[0238] The data preprocessing module is used to preprocess the real-time monitoring data to obtain a preprocessed monitoring data set;

[0239] The protocol identification and conversion module is used to perform protocol identification on the pre-processed monitoring data set and convert the heterogeneous data in the monitoring data set into structured data according to the preset protocol template library;

[0240] The feature matrix construction module is used to extract the time-frequency domain feature parameters of the device based on structured data and construct a multi-dimensional feature matrix;

[0241] The fault diagnosis module is used to input the multi-dimensional feature matrix into the pre-trained hybrid diagnosis model to generate the equipment health score and fault probability value;

[0242] A fault warning module generates warning instructions based on the health score and fault probability value, combined with a dynamic threshold algorithm. The dynamic threshold algorithm is adaptively adjusted based on the equipment operation data and environmental parameters in the historical database.

[0243] The maintenance strategy generation module is used to generate a priority maintenance strategy based on early warning instructions and equipment maintenance resource constraints through a maintenance strategy optimization model. The maintenance strategy optimization model is solved using equipment health score, failure probability, and equipment maintenance resources as constraints.

[0244] A railway traction substation status monitoring system in an embodiment of the present application can implement any of the above-mentioned monitoring methods, and the specific working processes of each module in the monitoring system can refer to the corresponding processes in the above-mentioned method embodiments.

[0245] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a module is merely a logical functional division, and in actual implementation, other division methods may be used, such as combining or integrating multiple modules into another system, or ignoring or not implementing certain features.

[0246] The embodiment of the present application also discloses a computer device.

[0247] The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for monitoring the status of a railway traction substation as described above is implemented.

[0248] The embodiment of the present application also discloses a computer-readable storage medium.

[0249] A computer-readable storage medium stores a computer program that can be loaded by a processor and executed by any one of the above-mentioned methods for monitoring the condition of a railway traction substation.

[0250] Among them, computer-readable storage media can be any tangible medium that contains or stores a program that can be used by or in combination with an instruction execution system, apparatus or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0251] It should be noted that, in the above embodiments, the description of each embodiment has different emphases. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0252] The above are all preferred embodiments of the present application and are not intended to limit the scope of protection of this application. Unless otherwise specified, any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features. In other words, unless otherwise specified, each feature is merely an example of a series of equivalent or similar features.

Claims

1. A method for monitoring the status of a railway traction substation, characterized in that: The monitoring method comprises: Obtain real-time monitoring data of equipment in railway traction substations; Performing data preprocessing on the real-time monitoring data to obtain a preprocessed monitoring data set; Performing protocol recognition on the preprocessed monitoring data set, and converting heterogeneous data in the monitoring data set into structured data according to a preset protocol template library; Based on the structured data, extracting time-frequency domain feature parameters of the device and constructing a multi-dimensional feature matrix; Inputting the multi-dimensional feature matrix into a pre-trained hybrid diagnostic model to generate a device health score and a failure probability value; Generate an early warning instruction based on the health score and the failure probability value in combination with a dynamic threshold algorithm; wherein the dynamic threshold algorithm is adaptively adjusted based on the equipment operation data and environmental parameters in the historical database; Based on the early warning instructions and equipment maintenance resource constraints, a priority maintenance strategy is generated through a maintenance strategy optimization model; wherein the maintenance strategy optimization model is solved with the equipment health score, failure probability, and equipment maintenance resources as constraints; the steps of performing protocol identification on the preprocessed monitoring data set and converting heterogeneous data in the monitoring data set into structured data according to a preset protocol template library include: Construct a protocol feature template library, analyze protocol message samples in historical monitoring data, and generate a feature transition probability matrix. The feature transition probability matrix is ​​a mathematical model that describes the flow patterns of protocol bytes by statistically analyzing the byte sequences in the protocol message samples and calculating the transition probability between each pair of adjacent bytes. Receive the preprocessed monitoring data set and intercept the current data frame based on the sliding window; Performing similarity matching on the byte transfer feature of the current data frame and the feature transfer probability matrix to determine the protocol type corresponding to the current data frame; Performing layered parsing on the current data frame according to the parsing rules corresponding to the protocol type to obtain a layered parsing result; wherein the layers include a physical layer, a data link layer, and an application layer; Encapsulating the hierarchical parsing result into structured data with semantic tags, wherein the structured data includes a device identifier, a measurement parameter name, and a standardized dimension value; performing hierarchical parsing on the current data frame according to the parsing rule corresponding to the protocol type to obtain the hierarchical parsing result comprises: Based on the protocol type, a preset protocol rule library is searched and corresponding physical layer parsing parameters, data link layer parsing parameters and application layer parsing parameters are loaded; Receiving a binary data stream of the current data frame; Performing checksum calculation on the binary data stream according to the checksum algorithm type included in the physical layer parsing parameter, separating the payload field and generating a physical layer parsing result including a checksum status flag; wherein the payload field includes actual monitoring data or control commands; Based on the message structure template defined by the data link layer parsing parameters, performing a field interception operation on the payload field to generate a structured message object containing message metadata; According to the register mapping relationship table in the application layer parsing parameter, the original register value in the payload field is converted into an engineering value to generate a standard format data object; The physical layer parsing result, the structured message object and the standard format data object are metadata-associated to obtain a hierarchical parsing result.

2. A railway traction substation status monitoring method according to claim 1, characterized in that: The steps of extracting the time-frequency domain characteristic parameters of the device based on the structured data and constructing a multi-dimensional characteristic matrix include: Performing equal-interval resampling processing on the structured data to generate an equal-interval time series data set; Extracting time-domain statistical characteristic parameters from the equally spaced time series data set; Performing frequency domain conversion processing on the equally spaced time series data set to calculate the main frequency component of the spectrum and the energy proportion of the preset frequency band; Obtain device operation record data and analyze device-related characteristic parameters; The time domain statistical characteristic parameters, the main frequency component of the spectrum, the preset frequency band energy ratio and the device-related characteristic parameters are aligned in a time window to construct a multi-dimensional characteristic matrix.

3. The method for monitoring the status of a railway traction substation according to claim 1, characterized in that: The method further includes a training step of the hybrid diagnostic model, the training step including: Extract historical multi-dimensional feature matrix and historical fault event timestamps from the equipment monitoring database; Performing abnormal data cleaning on the historical multidimensional feature matrix; Performing time alignment processing on the cleaned historical multidimensional feature matrix and the historical fault event timestamps, and marking the equipment health level; Constructing a labeled historical training dataset; the historical training dataset includes a historical multidimensional feature matrix, equipment health level labels, and historical fault event timestamps; Based on the historical training data set, a health scoring model and a fault probability prediction model are trained respectively; wherein the health scoring model uses a gradient boosting decision tree algorithm to process the historical multidimensional feature matrix and outputs a device health score; the fault probability prediction model uses a temporal convolutional network to process the historical multidimensional feature matrix within a continuous time window and outputs a fault probability value; The trained health score model and fault probability prediction model are jointly verified, and the model parameters are tuned according to the verification results.

4. A railway traction substation status monitoring method according to claim 1, characterized in that: The steps of generating a warning instruction based on the health score and the failure probability value in combination with a dynamic threshold algorithm include: Obtaining the health score, failure probability value, and current environmental parameters of the device; Extract the equipment operation data of the same type of equipment under similar environment parameters from the historical database according to the equipment type, and calculate the dynamic compensation factor based on the preset static threshold; Based on the dynamic compensation factor, a nonlinear correction is performed on the preset static threshold to generate a dynamic warning threshold under the current environment; wherein the dynamic warning threshold includes a health score threshold and a failure probability threshold; Calculating a health score degradation index based on a ratio of the health score to a health score threshold; Calculating a failure probability deviation index based on a ratio of the failure probability value to the failure probability threshold; The warning level is determined according to the product of the health score degradation index and the failure probability deviation index, and a graded warning instruction is generated.

5. A railway traction substation status monitoring method according to claim 4, characterized in that: Based on the warning instructions and the equipment maintenance resource constraints, the steps of generating a priority maintenance strategy through a maintenance strategy optimization model include: Obtaining a set of equipment warning instructions and equipment maintenance resource constraint data; wherein the warning instruction set includes equipment identification, health score, failure probability value, and warning level; the maintenance resource constraint data includes the number of available technicians, spare parts inventory, and maintenance time window; Calculate the equipment maintenance urgency coefficient based on the health score, failure probability value and warning level; Constructing an objective function including the maintenance urgency coefficient, and defining a resource consumption vector based on the equipment maintenance resource constraint data; Establishing a mixed integer programming model, taking the objective function and the resource consumption vector as input parameters, and introducing the device power grid topology relationship to generate conflict resolution rules; Solving the mixed integer programming model to obtain a set of candidate maintenance tasks; The candidate maintenance tasks are verified for completeness of spare parts and coordinated with the power outage plan, and a priority maintenance strategy is generated based on the verification results.

6. A railway traction substation status monitoring system, characterized in that: Used to execute the railway traction substation status monitoring method according to any one of claims 1 to 5, the monitoring system comprises: Data acquisition module, used to obtain real-time monitoring data of equipment in railway traction substations; A data preprocessing module is used to perform data preprocessing on the real-time monitoring data to obtain a preprocessed monitoring data set; A protocol identification and conversion module is used to perform protocol identification on the preprocessed monitoring data set and convert the heterogeneous data in the monitoring data set into structured data according to a preset protocol template library; A feature matrix construction module is used to extract the time-frequency domain feature parameters of the device based on the structured data and construct a multi-dimensional feature matrix; A fault diagnosis module, configured to input the multi-dimensional feature matrix into a pre-trained hybrid diagnosis model to generate a device health score and a fault probability value; a fault warning module, configured to generate a warning instruction based on the health score and the fault probability value in combination with a dynamic threshold algorithm; wherein the dynamic threshold algorithm is adaptively adjusted based on the equipment operation data and environmental parameters in the historical database; A maintenance strategy generation module is used to generate a priority maintenance strategy through a maintenance strategy optimization model based on the warning instructions and equipment maintenance resource constraints; wherein the maintenance strategy optimization model is solved with equipment health score, failure probability and equipment maintenance resources as constraints.

7. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 5 when executing the program.

8. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 5.

Images