Electromechanical equipment fault monitoring and early warning method

Abstract

The invention discloses an electromechanical equipment fault monitoring and early warning method, which comprises the following steps of: acquiring strong current data and weak current data of subway electromechanical equipment, and respectively carrying out preprocessing and time synchronization on the strong current data and the weak current data to obtain a strong current data sequence and a weak current data sequence; the strong current data sequence and the weak current data sequence are subjected to attention cross fusion and then input into a fault diagnosis model, and a fault type and the confidence coefficient thereof are obtained; inputting the fault type into a pre-constructed fault propagation knowledge graph, and querying to obtain a potential fault propagation path; and inputting the strong current data sequence, the weak current data sequence, the fault type, the confidence coefficient of the fault type and the potential fault propagation path into a Bayesian network, dynamically calculating a fault propagation probability and generating an early warning level.

Description

Technical Field

[0001] The present invention relates to the technical field of electromechanical equipment fault monitoring, and particularly to a method for monitoring and warning of electromechanical equipment faults. Background Art

[0002] Subway electromechanical equipment is an important infrastructure to ensure the safe operation of the subway. Subway electromechanical equipment is divided into a strong electricity part and a weak electricity part. The strong electricity part (such as 1500V DC traction power supply, 380V power distribution, etc.) undertakes high-power loads and needs to meet the high-energy consumption requirements of train traction, environmental control ventilation, escalators, etc.; the weak electricity part (signal control, communication transmission, BAS environmental monitoring, etc.) covers the equipment interaction network throughout the line and has the characteristics of low voltage, wide coverage, and high sensitivity.

[0003] Traditional manual inspection of electromechanical equipment is basically in a passive state of receiving abnormalities. The existing technology obtains image status information through image recognition of images taken of these electromechanical equipment, that is, specific alarm indicator lights on the subway electromechanical equipment light up or the dial pointers are abnormal to discover abnormalities. In this case, it is very difficult to predict potential fault hazards in advance. Once an alarm occurs, it means that the situation is relatively serious and corresponding measures must be taken within a short time. If there is no immediate response, it is very likely to affect the safety of subway operation.

[0004] Furthermore, the existing inspection methods do not distinguish the characteristics of the strong and weak electricity systems, mix the strong electricity part (such as power supply equipment) and the weak electricity part (such as signal control equipment) with the same detection logic, and ignore the characteristics of time-series data in strong electricity monitoring and discrete data characteristics in weak electricity monitoring, resulting in a high false alarm and missed alarm rate. In addition, there is a deep coupling between the strong electricity and weak electricity in subway electromechanical equipment (such as traction power supply fluctuations triggering signal system error codes). The existing technology lacks the ability to model the cross-system fault propagation path and cannot warn of derivative faults (such as voltage dips causing communication interruptions).

[0005] Designing a method for monitoring and warning of electromechanical equipment faults to address the above problems in the existing technology is the purpose of the research of the present invention. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to propose a method for monitoring and warning of electromechanical equipment faults, which can solve the above problems.

[0007] The present invention provides a method for monitoring and warning of electromechanical equipment faults, including: Collecting the strong electricity data and weak electricity data of subway electromechanical equipment, respectively preprocessing and time-synchronizing the strong electricity data and weak electricity data to obtain a strong electricity data sequence and a weak electricity data sequence; Performing attention cross-fusion on the strong electricity data sequence and the weak electricity data sequence and then inputting them into a fault diagnosis model to obtain the fault type and its confidence level; Input the fault type into the pre-built fault propagation knowledge graph to query the potential fault propagation paths. Input the strong power data sequence, weak power data sequence, fault type and its confidence level, and the potential fault propagation paths into the Bayesian network to dynamically calculate the fault propagation probability and generate a warning level.

[0008] Furthermore, the acquisition of the strong power data and weak power data of the subway electromechanical equipment, and the preprocessing and time synchronization of the strong power data and weak power data respectively to obtain the strong power data sequence and the weak power data sequence include: Collect the voltage data and current data of the subway electromechanical equipment, filter the voltage data and current data to obtain the strong power data sequence; Collect the communication status data and control instruction data of the subway electromechanical equipment through network logs, perform error checking and cleaning processing on the communication status data, and record the time stamps of the control instruction data to obtain the weak power data sequence.

[0009] Furthermore, the communication status data includes: packet loss rate, latency, bit error rate; The control instruction data includes: equipment operation start / stop signal, parameter adjustment signal.

[0010] Furthermore, the input of the strong power data sequence and the weak power data sequence into the fault diagnosis model after attention cross-fusion to obtain the fault type and its confidence level includes: Align the strong power data sequence and the weak power data sequence according to the time stamp, and perform normalization processing on the strong power data sequence and the weak power data sequence; Fuse the normalized strong power data sequence and weak power data sequence through the multi-head attention mechanism to obtain a fused data sequence; Input the fused data sequence into the fault classification model, aggregate the fused data sequence through the fully connected layer of the fault classification model, and use the Softmax activation function to output the fault type and its confidence level.

[0011] Furthermore, the fusion of the normalized strong power data sequence and weak power data sequence through the multi-head attention mechanism to obtain a fused data sequence includes: Extract the high-dimensional feature representations of the strong power data sequence and the weak power data sequence respectively through a convolutional neural network to obtain the strong power high-dimensional feature representation and the weak power high-dimensional feature representation ; Convert the strong power high-dimensional feature representation and the weak power high-dimensional feature representation into a strong power query matrix Q, a weak power index matrix K, and a weak power value matrix V respectively. The calculation formulas are as follows: , , , Among them, is the weight of the strong - electricity query matrix Q, is the weight of the weak - electricity index matrix K, is the weight of the weak - electricity value matrix V; The attention weight is calculated through the query matrix Q, the index matrix K, and the value matrix V. The calculation formula is as follows: , Among them, is the scaling factor, and softmax is the activation function to ensure weight normalization; Through the attention weight, the strong - electricity high - dimensional feature representation and the weak - electricity high - dimensional feature representation are fused by the attention vector to obtain the fused data sequence.

[0012] Furthermore, the fault propagation knowledge graph is constructed through the following steps: Collect the historical strong - electricity data of subway electromechanical equipment and their fault types, weak - electricity data and their fault types, and expert knowledge and experience data; Define the subway electromechanical equipment as equipment nodes, the corresponding strong - electricity data and weak - electricity data of the subway electromechanical equipment as state nodes, and the corresponding fault types of the subway electromechanical equipment as fault nodes; Obtain the causal relationship and dependency relationship between equipment nodes, state nodes, and fault nodes through expert knowledge and experience data, and use the causal relationship and dependency relationship between equipment nodes, state nodes, and fault nodes as edge relationships; Set the propagation weight for the edge relationship according to expert knowledge and experience to construct the fault propagation knowledge graph.

[0013] Furthermore, inputting the fault type into the pre - constructed fault propagation knowledge graph to query the potential fault propagation path includes: Use the fault type as the query starting point of the knowledge graph, and use the depth - first search algorithm to start from the query starting point to find the associated propagation path; Screen the associated propagation paths with propagation weights greater than the weight threshold as potential fault propagation paths.

[0014] Furthermore, the Bayesian network is obtained through the following steps: Extract the fault nodes as equipment state nodes through the potential propagation path, and use the intermediate state nodes obtained from the potential propagation path as path propagation nodes; Based on the propagation path weight of the potential fault propagation path, define the causal relationship edge from the parent node to the child node, initialize the conditional probability table, and construct the Bayesian network.

[0015] Further, the step of inputting the strong power data sequence, weak power data sequence, fault type and its confidence level, and potential fault propagation path into the Bayesian network to dynamically calculate the fault propagation probability and generate a warning level includes: Set the corresponding nodes of the strong power data sequence and weak power data sequence as Boolean observed evidences, and input the fault confidence level as the prior probability of the parent node into the Bayesian network; Traverse the paths in the Bayesian network, calculate the fault propagation probability of the parent node to the child node level by level, and generate a warning level according to the fault propagation probability.

[0016] Further, the step of generating a warning level according to the fault propagation probability includes: If the fault propagation probability ≤ 0.5, it is defined that the fault influence range is limited, and a low-level warning is generated; If 0.5 < the fault propagation probability ≤ 0.8, it is defined that the fault may affect some devices, and a medium-level warning is generated; If the fault propagation probability > 0.8, it is defined that the fault may trigger a system-level fault, and a high-level warning is generated.

[0017] Advantages of the present invention: First, by collecting the strong power data and weak power data of subway electromechanical equipment, and through processing such as filtering, verification, and cleaning, the quality of the original data is significantly improved, and the interference of noise and error codes is reduced. Appropriate preprocessing means are adopted for different signals to provide the optimal input for subsequent multi-source fusion, feature extraction, and analysis.

[0018] Second, by using a convolutional neural network to extract high-order features of strong and weak power respectively, the spatial and internal structure of the signal is fully utilized. The multi-head attention mechanism realizes the dynamic association of heterogeneous feature points, automatically captures key fault symptoms and diagnostic clues of multi-factor coupling, greatly improves the fault recognition accuracy and generalization ability of the model. The attention distribution weight reflects the importance of strong power and weak power data to the diagnostic conclusion during a fault, which helps to improve interpretability. The fault type and accurate confidence level are output through the fault classification model to achieve soft decision-making and uncertainty quantification, which is convenient for subsequent probability modeling and hierarchical warning.

[0019] Third, by converting the expert experience and historical events accumulated offline into a machine-readable knowledge graph, a causal relationship model of compound, hierarchical, and cross-device is realized. The depth-first search and weight screening mechanism can quickly discover potential fault propagation paths that may have a wide impact / cross systems, fill the shortcoming that traditional systems cannot capture complex chain risks, realize the screening of possible fault propagation paths, and effectively control the reasoning complexity and false alarm and missed alarm.

[0020] Fourth, a knowledge graph is used to guide the automatic modeling of Bayesian networks to solve the problems of scalable and interpretable modeling of fault propagation chains. The conditional probability table (CPT) realizes risk quantification under multi-factor coupling, allows dynamic input of real-time confidence, and converts qualitative knowledge into quantitative analysis. The probability is calculated layer by layer and the "low / medium / high" warnings are output hierarchically to realize dynamic risk stratification prompts across strong and weak power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0022] Figure 1 It is the flowchart of the method in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] For the convenience of those skilled in the art, the structure of the present invention will be further described in detail below in combination with the accompanying drawings. It should be understood that in the steps mentioned in this embodiment, unless specifically stated otherwise, the order can be adjusted according to actual needs, and even can be executed simultaneously or partially simultaneously.

[0024] As Figure 1 shown, an embodiment of the present invention provides a method for monitoring and warning of mechanical and electrical equipment failures, including: S1 Collect the strong power data and weak power data of subway mechanical and electrical equipment, preprocess and synchronize the time of the strong power data and weak power data respectively to obtain a strong power data sequence and a weak power data sequence; S101 Collect the voltage data and current data of subway mechanical and electrical equipment, filter the voltage data and current data to obtain a strong power data sequence; S102 Collect the communication status data and control instruction data of subway mechanical and electrical equipment through network logs, perform error checking and cleaning processing on the communication status data, and record the time stamps of the control instruction data to obtain a weak power data sequence.

[0025] Furthermore, the communication status data includes: packet loss rate, delay, bit error rate; The control instruction data includes: equipment operation start / stop signal, parameter adjustment signal.

[0026] In this step, subway electromechanical equipment usually operates coordinately by a strong power system (driving, power supply) and a weak power system (control, communication). Abnormalities in the strong power may affect the weak power. For example, voltage fluctuations may cause the failure of control instruction execution. Abnormalities in the weak power may trigger strong power faults. For example, the loss of control signals may lead to motor overload and even tripping.

[0027] The voltage and current of subway electromechanical equipment are the core energy parts of equipment operation and are usually used to directly monitor whether the equipment is in normal operation. Filtering is a key step to remove noise and outliers in the data to ensure data accuracy.

[0028] The control instruction data (equipment operation start / stop signals, parameter adjustment signals) and communication status data (packet loss rate, delay, bit error rate) of subway electromechanical equipment reflect the control logic of the equipment, the signal transmission quality, and whether the operation instructions are normal. If the communication signal is abnormal (such as an increase in the packet loss rate) or the control instruction cannot be correctly executed, it may cause problems in the operation of the strong power equipment.

[0029] The communication link is subject to various interferences during data transmission, resulting in errors in communication data. For example, due to network congestion or link failures, some data packets may be lost. Through verification means (such as CRC verification), errors in the data transmission process can be quickly detected to prevent incorrect data from entering the subsequent analysis process. Eliminate invalid information to avoid misleading the analysis results.

[0030] S2 inputs the strongly and weakly powered data sequences after attention cross-fusion into a fault diagnosis model to obtain the fault type and its confidence level; S201 aligns the strongly and weakly powered data sequences according to timestamps and normalizes the strongly and weakly powered data sequences; In this step, the strongly and weakly powered data sequences are normalized to the same range (such as [0,1]) for subsequent fusion.

[0031] S202 fuses the normalized strongly and weakly powered data sequences through a multi-head attention mechanism to obtain a fused data sequence; S2021 respectively extracts high-dimensional feature representations of the strongly and weakly powered data sequences through a convolutional neural network to obtain a strongly powered high-dimensional feature representation and a weakly powered high-dimensional feature representation ; S2022 converts the strongly powered high-dimensional feature representation and the weakly powered high-dimensional feature representation into a strongly powered query matrix Q, a weakly powered index matrix K, and a weakly powered value matrix V respectively. The calculation formulas are as follows: , , , Among them, is the weight of the strong - electricity query matrix Q, is the weight of the weak - electricity index matrix K, is the weight of the weak - electricity value matrix V; In this step, the query matrix Q usually represents the focus or query requirements for specific target data. The strong - electricity data sequence is closely related to the direct operating state of the electromechanical equipment. The changes in it are often the direct manifestation of the faults of the electromechanical equipment. The goal of the strong - and - weak - electricity collaborative diagnosis is to find the root cause or propagation path of the fault through the interactive analysis of the strong - and - weak - electricity data. When the strong - electricity data shows anomalies (such as excessive current fluctuations), it is necessary to find out which weak - electricity signals (indexes) and eigenvalues can explain this anomaly. Therefore, the strong - electricity data is used as the query matrix Q to query the weak - electricity data. The weak - electricity data itself does not directly represent the equipment operating state, but is a partial reflection of the communication and control links. If the weak - electricity is used as the query first, the obtained analysis results may only describe the communication state itself and cannot reflect the actual impact on the strong - electricity state.

[0032] The index K of the weak - electricity data (such as the time series of communication status, control instructions) can be regarded as an identifier describing the health of equipment control and communication. It is used to find the weak - electricity information most relevant to the focus in the strong - electricity. For example, when an anomaly appears in the strong - electricity at a certain time point, the weak - electricity data may show the loss of control signals or the decline of communication quality at the corresponding time point, thus indicating the source of the problem. The value V in the weak - electricity data is a possible explanation for the strong - electricity state. For example, the packet loss rate can indicate whether the communication is interrupted, and the control instruction data can indicate whether there are incorrect instructions sent to the equipment.

[0033] S2023 calculates the attention weights through the query matrix Q, the index matrix K, and the value matrix V. The calculation formula is as follows: , Among them, is the scaling factor, and softmax is the activation function to ensure the normalization of the weights; S2024 performs attention - vector fusion on the strong - electricity high - dimensional feature representation and the weak - electricity high - dimensional feature representation to obtain the fused data sequence.

[0034] In this step, the attention mechanism can dynamically assign weights to different parts of the high-voltage data and low-voltage data, capturing the important correlation relationships between them. For example: If there is an abnormal fluctuation in the current in the high-voltage sequence while there is no change in the control instructions in the low-voltage sequence, it may be a physical fault of the high-voltage equipment itself. If the communication status data (such as packet loss rate) in the low-voltage sequence abnormally increases, it may be that the low-voltage control system has caused a fault in the high-voltage equipment.

[0035] S203 inputs the fused data sequence into the fault classification model, aggregates the fused data sequence through the fully connected layer of the fault classification model, and uses the Softmax activation function to output the fault type and its confidence level.

[0036] In this step, the fault classification model can be constructed using a neural network model. Historical high-voltage data sequences of subway electromechanical equipment and low-voltage data sequences corresponding to the time stamps are collected as sample inputs, and the manually labeled fault types and their confidence levels are used as sample outputs. This fault classification model is trained, and the category with the highest probability is selected as the fault type, and this probability is output as the confidence level.

[0037] S3 inputs the fault type into the pre-constructed fault propagation knowledge graph to query and obtain the potential fault propagation paths; Specifically, the fault propagation knowledge graph is constructed through the following steps: Collect historical high-voltage data of subway electromechanical equipment and their fault types, low-voltage data and their fault types, and expert knowledge and experience data; Define the subway electromechanical equipment as equipment nodes, the corresponding high-voltage data and low-voltage data of the subway electromechanical equipment as state nodes, and the corresponding fault types of the subway electromechanical equipment as fault nodes; Obtain the causal relationships and dependency relationships between equipment nodes, state nodes, and fault nodes through expert knowledge and experience data, and use the causal relationships and dependency relationships between equipment nodes, state nodes, and fault nodes as edge relationships; Set propagation weights for the edge relationships according to expert knowledge and experience to construct the fault propagation knowledge graph.

[0038] In this step, the subway electromechanical equipment operates in coordination by the strong power system (such as drive and power supply) and the weak power system (such as control and communication). There are complex interactions between the strong power system and the weak power system. Strong power faults (such as voltage fluctuations or current anomalies) will affect weak power functions (such as signal transmission or control logic errors). Weak power anomalies (such as signal packet loss or control command loss) may further trigger strong power system faults (such as motor overload or equipment tripping). The complexity between the strong and weak powers is high, and traditional single-domain fault modeling methods cannot effectively capture the correlation between them. In addition, electromechanical equipment faults may experience multi-hop propagation in multiple subsystems. For example: current fluctuation → control signal error → motor overload → tripping.

[0039] Current single-factor analysis methods cannot effectively infer potential multi-step propagation paths and fault root causes. Traditional fault diagnosis methods, such as rule-based expert systems or statistical models, are often static and cannot dynamically adjust predictions according to changes in equipment status, and perform poorly on unknown faults or unseen correlation relationships.

[0040] Map strong power (such as current and voltage) data to state nodes, and also model relevant weak power data (control instruction data (equipment operation start-stop signal, parameter adjustment signal), communication status data (packet loss rate, delay, bit error rate)) as state nodes, and connect these nodes through causal relationships. For example: current fluctuation → voltage instability → control signal failure → motor tripping. Using the graph structure of the knowledge graph, the multi-step relationship from strong power problems to weak power problems can be represented, and the correlation points can be inferred. Map electromechanical equipment to equipment nodes, and establish functional dependencies between equipment and other nodes, such as controller B → motor A. The knowledge graph combines domain expert knowledge to construct mapping relationships and propagation weights. At the same time, newly mined causal relationships (such as new equipment, new faults) from historical data can be added in batches to provide basic data for subsequent queries of potential fault propagation paths.

[0041] S301 uses the fault type as the query starting point of the knowledge graph, and starts from the query starting point using the depth-first search algorithm to find the associated propagation path; S302 screens the associated propagation paths with propagation weights greater than the weight threshold as potential fault propagation paths.

[0042] In this step, Depth-First Search (DFS) is an algorithm used to traverse or search graph or tree data structures. It starts from a starting point (or root node), explores a branch as far as possible until it reaches the end point or there are no more unvisited nodes, and then backtracks to continue searching other branches. In the fault propagation knowledge graph, a certain node (such as a device or state in the knowledge graph) is often connected to multiple nodes through causal relationships, dependencies, etc. Depth-First Search can start from a certain node and find all possible paths related to it.

[0043] Suppose the knowledge graph contains the following propagation relationships: motor overcurrent → rectifier module failure (propagation probability 0.8), rectifier module failure → communication signal interruption (propagation probability 0.7). When "motor overcurrent" is input, the query results may be: Path 1: motor overcurrent → rectifier module failure (probability 0.8), Path 2: motor overcurrent → rectifier module failure → communication signal interruption (probability 0.56). Relevant propagation paths are obtained through the search algorithm, and then invalid paths with low propagation possibilities (low-weight paths) are removed through the weight threshold to focus on the key propagation chains. Necessary paths are retained to improve the calculation efficiency of the subsequent Bayesian network.

[0044] S4 inputs the strong electrical data sequence, weak electrical data sequence, fault type and its confidence level, and the potential fault propagation path into the Bayesian network, dynamically calculates the fault propagation probability, and generates a warning level.

[0045] Specifically, the Bayesian network is obtained through the following steps: Extract fault nodes as device state nodes through potential propagation paths, and obtain intermediate state nodes as path propagation nodes through potential propagation paths; Based on the propagation path weights of potential fault propagation paths, define the causal relationship edges from the parent node to the child node, initialize the conditional probability table, and construct the Bayesian network.

[0046] In this step, the fault propagation of subway electromechanical equipment involves multi-level dependencies. Traditional rule engines or static fault trees are difficult to quantify dynamic propagation risks. Fault propagation is affected by multiple factors such as the environment, device state, and historical data, and there is a lot of uncertainty. In addition, there are various types of data signals in subway electromechanical equipment. The multi-level causal relationship can be intuitively represented through the directed acyclic graph (DAG) structure of the Bayesian network. The Bayesian network maps all data into nodes and uniformly models the dependencies through the conditional probability table (CPT). The Bayesian network quantifies uncertainty through probability distributions and converts qualitative knowledge into quantitative probabilities.

[0047] The device status node represents a specific fault (such as "motor overcurrent") and is the starting point and core of reasoning. The path propagation node represents an intermediate state (such as "rectifier module fault"), quantifies the probability attenuation in the propagation process, and can decompose complex propagation paths into computable probability chains. Convert the propagation weight from expert experience or historical statistics (such as the weight of "motor overcurrent → rectifier module fault" being 0.8) into a conditional probability P(rectifier module fault | motor overcurrent) = 0.8 to make the logical relationship quantifiable. Initialize the parent node probability using the fault confidence level output by S2 (such as P(motor overcurrent) = 0.85) to ensure that the starting point of reasoning is consistent with the actual diagnosis result.

[0048] S401 sets the corresponding nodes of the strong electricity data sequence and the weak electricity data sequence as Boolean observation evidences, and inputs the fault confidence level as the prior probability of the parent node into the Bayesian network; S402 traverses the paths in the Bayesian network, calculates the fault propagation probability of the parent node to the child node level by level, and generates a warning level according to the fault propagation probability.

[0049] S4021 If the fault propagation probability ≤ 0.5, it is defined that the fault influence range is limited, and a low-level warning is generated; S4022 If 0.5 < the fault propagation probability ≤ 0.8, it is defined that the fault may affect some devices, and a medium-level warning is generated; S4023 If the fault propagation probability > 0.8, it is defined that the fault may trigger a system-level fault, and a high-level warning is generated.

[0050] In this step, existing rule-based or static fault tree technologies often only focus on the fault chain within a single system and are difficult to model the "cross-system" fault propagation path. For example, after a voltage dip occurs in the electrical system, it may trigger an interruption in the communication system link, but static analysis methods usually cannot perceive and warn of such derivative faults. Using a Bayesian network, infer the occurrence probability of its possible downstream faults (child nodes) based on the parent node (usually the known and directly observed direct fault). Quantify the risk of a fault propagating from one node to other nodes, not just "whether there is" but "what is the probability of propagation". Achieve real-time warning of dynamic and multi-level fault associations in complex systems, not just relying on static logic or manual experience. Divide the propagation probability into three levels and give warnings of different severity levels to the management system or operation and maintenance personnel.

[0051] For example, when a voltage dip occurs in the subway traction power supply system, the traditional operation and maintenance monitoring system can only issue warnings for power supply-related faults and cannot insight into the further impact of electrical faults on the communication system (such as the signal network) - for example, the voltage dip causes abnormal communication power supply, ultimately leading to communication link jitter or even interruption. Existing rules and fault trees often cannot model and timely warn of such "cross-system" derivative risks.

[0052] Through step S3, "voltage dip → abnormal communication power supply → communication interruption" is screened out as a potential propagation path. Through step S4, Bayesian network modeling is carried out for node setting, including voltage dip (device status node), abnormal communication power supply (intermediate propagation node), and communication interruption (device status node). CPT initialization is performed: P(abnormal communication power supply|voltage dip)=0.75, P(communication interruption|abnormal communication power supply)=0.8. The parent node "voltage dip" is input with a prior probability of 0.82 (obtained in step S2). Dynamic propagation probability calculation: P(abnormal communication power supply) = 0.82 × 0.75 = 0.615, P(communication interruption) = 0.615 × 0.8 = 0.492.

[0053] Early warning classification: The propagation probability of abnormal communication power supply is 0.615 (0.5 < 0.615 ≤ 0.8), generating a medium-level early warning. The propagation probability of communication interruption is 0.492 (≤0.5), generating a low-level early warning.

[0054] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0055] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, as well as the combination of flows and / or blocks in the flowchart and / or block diagram. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0056] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0057] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable apparatus to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable apparatus provide steps for realizing the functions specified in one process or a plurality of processes and / or one block or a plurality of blocks. Figure 1 one process or a plurality of processes and / or Figure 1 steps for realizing the functions specified in one block or a plurality of blocks.

[0058] It should be noted that, in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of other elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In a unit claim listing several means, several of these means can be embodied by one and the same item of hardware. The use of the words first, second, and third, etc. does not denote any order. These words can be interpreted as names.

[0059] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0060] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

[0061] In the present invention, unless otherwise clearly defined and limited, the terms "mounted", "connected", "coupled", "fixed", etc. shall be construed in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral body; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, and it can be the internal communication of two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0062] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic descriptions of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

Claims

1. A method for monitoring and early warning of electromechanical equipment faults, characterized in that: include: Collect the strong and weak current data of subway electromechanical equipment, pre-process and time-synchronize the strong and weak current data, and obtain the strong and weak current data sequences; The strong current data sequence and the weak current data sequence are cross-fused and input into the fault diagnosis model to obtain the fault type and its confidence level; Input the fault type into the pre-built fault propagation knowledge graph and query to obtain the potential fault propagation path; The strong current data sequence, weak current data sequence, fault type and its confidence, and potential fault propagation path are input into the Bayesian network to dynamically calculate the fault propagation probability and generate the warning level.

2. A method for monitoring and early warning of electromechanical equipment faults according to claim 1, characterized in that: The strong current data and weak current data of the subway electromechanical equipment are collected, and the strong current data and the weak current data are preprocessed and time synchronized respectively to obtain the strong current data sequence and the weak current data sequence, including: Collect voltage and current data of subway electromechanical equipment, filter the voltage and current data, and obtain a strong current data sequence; The communication status data and control instruction data of subway electromechanical equipment are collected through network logs, the communication status data is error checked and cleaned, and the control instruction data is timestamped to obtain the weak current data sequence.

3. A method for monitoring and early warning of electromechanical equipment faults according to claim 2, characterized in that: The communication status data includes: packet loss rate, delay, bit error rate; The control instruction data includes: equipment operation start and stop signals and parameter adjustment signals.

4. The method for monitoring and early warning of electromechanical equipment faults according to claim 1, characterized in that: The strong current data sequence and the weak current data sequence are cross-fused and input into the fault diagnosis model to obtain the fault type and its confidence level, including: Align the strong current data sequence and the weak current data sequence according to the timestamp, and normalize the strong current data sequence and the weak current data sequence; The normalized strong current data sequence and weak current data sequence are fused through the multi-head attention mechanism to obtain a fused data sequence; The fused data sequence is input into the fault classification model, the fused data sequence is aggregated through the fully connected layer of the fault classification model, and the Softmax activation function is used to output the fault type and its confidence.

5. A method for monitoring and early warning of electromechanical equipment faults according to claim 4, characterized in that: The normalized strong current data sequence and weak current data sequence are fused by the multi-head attention mechanism to obtain a fused data sequence including: The high-dimensional feature representation of strong current data sequence and weak current data sequence is extracted through convolutional neural network respectively, and the high-dimensional feature representation of strong current is obtained. and weak high-dimensional feature representation ; Representing strong electric high-dimensional features and weak high-dimensional feature representation They are respectively converted into strong current query matrix Q, weak current index matrix K, and weak current value matrix V. The calculation formulas are as follows: , , , in, is the weight of the strong power query matrix Q, is the weight of the weak current index matrix K, is the weight of the weak electric value matrix V; The attention weight is calculated by query matrix Q, index matrix K, and value matrix V. The calculation formula is as follows: ， in, is the scaling factor, and softmax is the activation function to ensure weight normalization; The strong high-dimensional features are represented by attention weights and weak high-dimensional feature representation Perform attention vector fusion to obtain a fused data sequence.

6. The electromechanical equipment fault monitoring and early warning method according to claim 1 is characterized in that: The fault propagation knowledge graph is constructed through the following steps: Collect historical high-voltage data and fault types of subway electromechanical equipment, low-voltage data and fault types, and expert knowledge and experience data; The metro electromechanical equipment is defined as the equipment node, the strong and weak current data corresponding to the metro electromechanical equipment is defined as the state node, and the fault type corresponding to the metro electromechanical equipment is defined as the fault node; Obtain the causal relationship and dependency relationship between device nodes, status nodes, and fault nodes through expert knowledge and experience data, and use the causal relationship and dependency relationship between device nodes, status nodes, and fault nodes as edge relationships; Set propagation weights for edge relationships based on expert knowledge and experience, and build a fault propagation knowledge graph.

7. A method for monitoring and early warning of electromechanical equipment faults according to claim 6, characterized in that: The fault type is input into the pre-built fault propagation knowledge graph, and the potential fault propagation paths obtained by query include: Take the fault type as the query starting point of the knowledge graph, and use the depth-first search algorithm to find the associated propagation path from the query starting point; The associated propagation paths whose propagation weights are greater than a weight threshold are screened as potential fault propagation paths.

8. A method for monitoring and early warning of electromechanical equipment faults according to claim 7, characterized in that: The Bayesian network is obtained by the following steps: Extract the fault node as the device state node through the potential propagation path, and obtain the intermediate state node of the potential propagation path as the path propagation node; Based on the propagation path weights of the potential fault propagation paths, the causal relationship edges from the parent node to the child node are defined, the conditional probability table is initialized, and the Bayesian network is constructed.

9. A method for monitoring and early warning of electromechanical equipment faults according to claim 8, characterized in that: The method of inputting the strong current data sequence, the weak current data sequence, the fault type and its confidence, and the potential fault propagation path into the Bayesian network, dynamically calculating the fault propagation probability and generating the warning level includes: The nodes corresponding to the strong current data sequence and the weak current data sequence are set as Boolean observation evidence, and the fault confidence is input into the Bayesian network as the prior probability of the parent node; Traverse the paths in the Bayesian network, calculate the fault propagation probability of the parent node to the child node level by level, and generate the warning level according to the fault propagation probability.

10. A method for monitoring and early warning of electromechanical equipment faults according to claim 9, characterized in that: Generating the warning level according to the fault propagation probability includes: If the fault propagation probability is ≤0.5, it is defined as a limited fault impact range, and a low-level warning is generated; If 0.5<fault propagation probability≤0.8, it is defined as the fault may affect some equipment and an intermediate warning is generated; If the fault propagation probability is > 0.8, it is defined as a fault that may cause a system-level failure and generate an advanced warning.

Images