Pantograph and catenary state monitoring system and method

Abstract

The application discloses a pantograph-catenary state monitoring system and method, and belongs to the technical field of pantograph-catenary monitoring, and comprises the following steps: training a monitoring model to identify a safety event; obtaining pantograph-catenary state monitoring information; determining a first temperature sequence of a target monitoring area in a first time period; judging a current train operation condition based on a train state monitoring model; determining a first abnormality judgment parameter; predicting a second temperature sequence of the target monitoring area in a second time period; determining a second abnormality judgment parameter; determining a comprehensive abnormality judgment value; determining abnormality early warning information of at least one time point; adjusting monitoring model parameters based on the current train operation condition; determining the occurrence probability of the safety event through the monitoring model; and determining safety alarm information of at least one time point. The application improves the processing speed and efficiency of responding to safety events, enhances the safety performance and disaster prevention capability, and achieves the technical effect of improving the pantograph-catenary state monitoring precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pantograph-catenary monitoring, in particular to a pantograph-catenary state monitoring system and method. BACKGROUND

[0002] The work camping car is a special train for providing rest and life for railway work personnel during construction, which mainly provides stable power supply for workers through the overhead line system power supply device on the work camping car. The work camping car overhead line system power supply device is a device installed on the work camping car, which takes power from the overhead line through the pantograph to provide power for the camping car.

[0003] The existing work camping car power supply system lacks online monitoring and automatic drop bow interlocking of the pantograph-catenary, and cannot identify hidden dangers such as electric arc and overheating in advance. Once the abnormality continues, there is a lack of quick drop bow protection, which is easy to cause the overhead line to burn and break, and manual handling is required, which brings high-voltage electric shock risk. SUMMARY

[0004] To solve the above problems, the first aspect of the present application provides a pantograph-catenary state monitoring method, comprising the following steps: training a monitoring model to identify safety events;

[0005] Obtaining pantograph-catenary state monitoring information;

[0006] Based on the pantograph-catenary state monitoring information, determining a first temperature sequence of the target monitoring area in a first time period;

[0007] Based on the train state monitoring model, the current running condition of the train is judged;

[0008] Based on the current running condition of the train, a first abnormality judgment parameter is determined through the corresponding physical model;

[0009] Based on the first temperature sequence of the target monitoring area in the first time period, a second temperature sequence of the target monitoring area in a second time period is predicted through a temperature prediction model;

[0010] Based on the second temperature sequence of the target monitoring area in the second time period and the first temperature sequence of the target monitoring area in the second time period, a second abnormality judgment parameter is determined;

[0011] Based on the first abnormality judgment parameter and the second abnormality judgment parameter, a comprehensive abnormality judgment value is determined;

[0012] Based on the comprehensive abnormality judgment value and the abnormality threshold set, abnormal early warning information of at least one time point is determined;

[0013] Based on the current running condition of the train, the monitoring model parameters are adjusted;

[0014] Based on the pantograph-catenary state monitoring information, the occurrence probability of the safety event is determined through the monitoring model.

[0015] Based on the occurrence probability of the safety event and the dynamic probability threshold set, the safety alarm information of at least one time point is determined.

[0016] Preferably, the training monitoring model identifying the safety event comprises:

[0017] Defining the knowledge graph entity and the relationship between the entities;

[0018] Constructing the training graph based on the safety event set;

[0019] Training the monitoring model to extract the global features based on the training graph;

[0020] Training the monitoring model to extract the key features based on the training graph subgraph.

[0021] Preferably, the determination of the first temperature sequence of the target monitoring area in the first time period comprises:

[0022] Determining the first temperature value of the target monitoring area at at least one time point in the first time period;

[0023] The calculation method of the first temperature value is:

[0024] ;

[0025] Wherein, is the first temperature value at the time point, is the weight of the temperature data in the temperature matrix, is the temperature value at the time point in the temperature matrix, is the temperature value at the time point in the temperature matrix, is the number of temperature values in the temperature matrix; Based on the first temperature value at at least one time point in the first time period, the first temperature sequence of the target monitoring area in the first time period is determined. Preferably, the calculation method of the first abnormality judgment parameter is:

[0026]

[0027] ;

[0028] ;

[0029] Wherein, is the first abnormality judgment parameter, is the dynamic temperature feature deviation degree, is the theoretical temperature difference, is the actual temperature difference, is the adjustment factor; ​​

[0030] is a static temperature characteristic deviation, is an instantaneous temperature change rate, is a temperature rise speed threshold, is a decay function;

[0031] The decay function is calculated as follows:

[0032] ;

[0033] wherein, is a base of a natural logarithm, is a duration of a current scene, is a decay time constant.

[0034] Preferably, the predicting, by the temperature prediction model, a second temperature sequence of the target monitoring area in the second time period comprises:

[0035] obtaining a plurality of temperature prediction models;

[0036] for each of the plurality of temperature prediction models, determining a fitting degree between the first temperature sequence and the temperature prediction model;

[0037] based on the fitting degrees between the first temperature sequence and each of the temperature prediction models, determining a target temperature prediction model from the plurality of temperature prediction models;

[0038] based on the target temperature prediction model, determining a second temperature sequence of the target monitoring area in the second time period.

[0039] Preferably, the second abnormality judgment parameter is calculated as follows:

[0040] ;

[0041] wherein, is a temperature error degree at a moment, is a real-time temperature at the moment, is a predicted temperature at the moment, is a limit temperature, is an environmental temperature. Preferably, the determining, based on the first abnormality judgment parameter and the second abnormality judgment parameter, a comprehensive abnormality judgment value comprises: determining a fusion abnormality judgment value based on the first abnormality judgment parameter, the second abnormality judgment parameter, and the fitting degree;

[0042]

[0043]

[0044] ​​​The calculation method of the fusion anomaly judgment value is:

[0045] ;

[0046] Wherein, is the fusion anomaly judgment value, is the fitting degree, is the environmental influence factor;

[0047] Determine the comprehensive anomaly judgment value based on the fusion anomaly judgment value, the first anomaly judgment parameter, the second anomaly judgment parameter and the fitting degree;

[0048] The calculation method of the comprehensive anomaly judgment value is:

[0049] ;

[0050] Wherein, is the final anomaly judgment value, is the Sigmoid function;

[0051] ;

[0052] Wherein, is the base of natural logarithm, is the fitting degree.

[0053] Preferably, the adjusting monitoring model parameters based on the current running condition of the train comprises:

[0054] Determine the local subgraph feature weight based on the current running condition of the train;

[0055] The method for determining the local subgraph feature weight based on the current running condition of the train is:

[0056] ;

[0057] Wherein, is the local subgraph feature weight, is the basic subgraph feature weight, is the base of natural logarithm, is the attenuation coefficient, temperature rise amplitude.

[0058] Preferably, the calculation method of the occurrence probability of the safety event is:

[0059] ;

[0060] Wherein, is the occurrence probability of the safety event, is the nonlinear activation function, is the weight matrix of the global feature, global features, graph pooling features from the complete dataset, weight matrix of local features, local features, subgraph max pooling features.

[0061] The dynamic probability threshold set is calculated in the following manner:

[0062] ;

[0063] wherein, is the adjusted probability threshold, is the basic probability threshold, is the temperature rise amplitude, is the adjustment coefficient.

[0064] The second aspect of the present application provides a pantograph-catenary state monitoring system for executing the pantograph-catenary state monitoring method of any one of the above-mentioned schemes, comprising:

[0065] a training module for training the monitoring model;

[0066] an acquisition module for acquiring the pantograph-catenary state monitoring information;

[0067] a first determination module for determining a first temperature sequence of the target monitoring area in a first time period;

[0068] a research and judgment module for researching and judging the current operation condition of the train;

[0069] a first abnormality judgment parameter calculation module for determining a first abnormality judgment parameter;

[0070] a prediction module for predicting a second temperature sequence of the target monitoring area in a second time period;

[0071] a second abnormality judgment parameter calculation module for determining a second abnormality judgment parameter;

[0072] a comprehensive calculation module for determining a comprehensive abnormality judgment value according to the first abnormality judgment parameter and the second abnormality judgment parameter;

[0073] a warning module for determining abnormality warning information;

[0074] an adjustment module for adjusting the monitoring model parameters;

[0075] a second determination module for determining the occurrence probability of a safety event;

[0076] an alarm module for determining safety alarm information.

[0077] By adopting the above technical solutions, the present application mainly has the following technical effects:

[0078] By analyzing the target image through the trained monitoring model, the probability of occurrence of a safety event in the target image is obtained, and after determining the alarm information based on a probability threshold, the alarm information is sent to the relevant monitoring device of the train and / or the emergency handling department is activated, automatic pantograph lowering is realized, thereby greatly improving the processing speed and efficiency of responding to safety events, enhancing safety performance and disaster prevention ability, and avoiding further escalation of safety events; at the same time, during train starting, stopping and braking, by reducing the local subgraph feature weight and dynamically improving the probability threshold, the possibility of false positives is reduced, and the technical effect of improving the accuracy of pantograph-catenary state monitoring is realized. BRIEF DESCRIPTION OF DRAWINGS

[0079] Figure 1 is a flowchart of the pantograph-catenary state monitoring method of the present application;

[0080] Figure 2 is a circuit diagram of the image acquisition submodule of the acquisition module;

[0081] Figure 3 is a circuit diagram of the temperature alarm submodule in the alarm module;

[0082] Figure 4 is a partial enlarged view (one) in Figure 3

[0083] Figure 5 is a partial enlarged view (two) in Figure 3

[0084] Figure 6 is a partial enlarged view (three) in Figure 3 DETAILED DESCRIPTION

[0085] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some examples or embodiments of the present application, and for those skilled in the art, without creative labor, the present application can also be applied to other similar scenarios according to these drawings. Unless it is obvious from the language environment or otherwise stated, the same reference numbers in the drawings represent the same structures or operations.

[0086] It should be understood that the "module" used herein is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

[0087] ​​​Flowcharts in the present specification are used to illustrate the operations performed by the system according to the embodiments of the present specification. It should be understood that the preceding or subsequent operations are not necessarily performed in sequence. Instead, the steps can be processed in reverse order or simultaneously. Meanwhile, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0088] It should be noted that in the present application, in the calculation process,

[0089] Please refer to Figure 1 The first aspect of the present application provides a pantograph-catenary state monitoring method, comprising the following steps:

[0090] S1, training a monitoring model to identify safety events;

[0091] In some embodiments, the safety event refers to an event that can affect the safe operation of the train.

[0092] In some embodiments, the monitoring model can be any kind of deep learning model that can be used to process graph structure data.

[0093] In some embodiments, the monitoring model can be a graph convolutional network (GCN) model. In some embodiments, the graph convolutional network (GCN) is suitable for capturing and utilizing the topological structure information in graph data, such as social networks, molecular structures, transportation networks, etc. GCN learns the representation of nodes by performing convolution operations on graphs, so as to capture the dependency relationship between nodes.

[0094] In some embodiments, the training of the monitoring model to identify safety events comprises:

[0095] S101, defining knowledge graph entities and relationships between entities;

[0096] In some embodiments, the entity refers to the basic unit in the knowledge graph, and the relationship refers to the link connecting different entities, used to represent various interactions and contacts between entities.

[0097] In some embodiments, the entity can include a wire, a spark, and an environmental object. In some embodiments, the wire refers to a catenary wire for carrying current, the spark refers to light and heat generated due to abnormal discharge phenomenon, and the environmental object can include a bird nest, a plastic bag, and the like.

[0098] In some embodiments, the relationship between the entities can include electrical connection, representing the current conduction path between adjacent wires; spark attachment, representing the generation of sparks on the wire; and environmental contact, representing the spatial interaction of foreign matter with the wire.

[0099] S102. Construct a training graph based on security event sets;

[0100] In some embodiments, the set of safety events may include: images of adjacent wires breaking, images of sparks attached to wires, and images of foreign objects (bird nests, plastic bags) hanging on wires.

[0101] In some embodiments, the training graph refers to graph structure training data, which includes nodes, variables, node features, and labels.

[0102] In some embodiments, the YOLO algorithm can be used to identify entities in a security event set and the relationships between them.

[0103] For example, the YOLO algorithm can be used to identify images with sparks attached to a wire. Then, two nodes can be defined to represent the wire and the spark respectively, and an edge can be defined between the wire and the spark to represent that the spark is attached to the wire, thus constructing a training graph of sparks attached to the wire.

[0104] By constructing a training knowledge graph, the various elements of image content and their relationships are represented in graph form. Nodes in the graph can represent objects or events in the video, while edges represent the relationships between them. This transforms complex image content into a structured graphical representation, which not only helps in understanding the interactions between various elements in a scene but also enables the use of graph algorithms for advanced analysis tasks.

[0105] S103. Train the monitoring model based on the training graph to extract global features;

[0106] In some embodiments, the global features refer to the features of all nodes in the training graph and their interrelationships.

[0107] In some embodiments, the monitoring model extracts global features in the following way:

[0108] ,

[0109] in, Indicates the first The node feature matrix of the layer; Represents a nonlinear activation function; Represents the standardized adjacency matrix; Indicates the first The node feature matrix of the layer; Represents the learnable weight matrix; This indicates the bias term.

[0110] S104. Extract key features from the monitoring model trained on the training graph subgraph;

[0111] In some embodiments, the training graph subgraph is a local structure of a security event in the training graph. For example, the security event may be a spark attached to a wire, and the local structure may include a wire node and a local structure with at least one spark node attached.

[0112] In some embodiments, the key features refer to the characteristics of nodes associated with security events and their interrelationships.

[0113] In some embodiments, the monitoring model extracts key features in the following way:

[0114] ,

[0115] in, For nodes In the The feature vector of the layer, It is a non-linear activation function. For nodes For nodes Attention weights For nodes exist The feature vector of the layer, For nodes The set of neighbors.

[0116] S2. Obtain pantograph-catenary condition monitoring information;

[0117] In some embodiments, the pantograph-catenary condition monitoring information refers to key information used to assess and monitor the interaction between the pantograph and the overhead contact line. In some embodiments, the pantograph-catenary condition monitoring information includes image data and temperature data.

[0118] In some embodiments, image data of the pantograph can be acquired by equipping a visible light camera on the roof of the vehicle. In some embodiments, a visible light camera is a device that uses electromagnetic waves in the visible light band to capture and record images. It can receive visible light from the subject through a lens, then convert these light signals into electrical signals, and generate a visualized image after processing.

[0119] In some embodiments, temperature data of the pantograph can be acquired by equipping a dual-spectrum thermal imaging infrared camera on the roof of the vehicle. In some embodiments, the dual-spectrum thermal imaging infrared camera is an advanced monitoring device that combines infrared thermal imaging and visible light imaging technologies. It can simultaneously provide temperature data and visible light images of the target object. Its principle is to use a lens to collect the infrared radiation emitted by the target object, and after filtering, light splitting, and detector signal conversion, a thermal imaging image is finally formed.

[0120] S3. Based on the pantograph-catenary condition monitoring information, determine the first temperature sequence of the target monitoring area within the first time period;

[0121] In some embodiments, the target monitoring area includes the contact strip where the pantograph carbon slide plate and the contact wire are mechanically and electrically coupled.

[0122] In some embodiments, the first time period can be a historical time period. For example, the first time period can be a historical 3-second period.

[0123] In some embodiments, determining the first temperature sequence of the target monitoring area within a first time period includes:

[0124] S301. Determine the first temperature value of the target monitoring area at least at one moment within the first time period;

[0125] In some embodiments, after continuously imaging the pantograph-catenary using a dual-spectrum thermal imaging infrared camera, a temperature matrix corresponding to each pixel in the target contact area can be obtained. The effective pixels of the temperature matrix can be extracted according to the actual projection mask of the 'carbon sliding plate-contact line', and weighted and summed according to the pixel area ratio or radiation intensity normalization weight. The calculation result is used as the first temperature value at that moment in the first time period.

[0126] In some embodiments, the first temperature value is calculated as follows:

[0127]

[0128] in, for The first temperature value at that moment, The first element in the temperature matrix The weight of each temperature data point In the temperature matrix Time of the first A temperature value, This represents the number of temperature values ​​in the temperature matrix.

[0129] In some embodiments, to facilitate subsequent calculations, a weighted summation method can be used to determine the first temperature value by weight normalization.

[0130] S302. Based on the first temperature value at least one moment within the first time period, determine the first temperature sequence of the target monitoring area within the first time period.

[0131] In some embodiments, the first temperature sequence is a sequence obtained by arranging the first temperature values ​​at at least one moment in chronological order. The chronological order can be either ascending or descending.

[0132] S4, determining the current running condition of the train based on the train state monitoring model;

[0133] In some embodiments, the train state monitoring model can be any model for determining the running state of the train according to the train running information.

[0134] In some embodiments, the train state monitoring model can be a state machine model based on speed and acceleration double parameters, the input of which can be the real-time speed v (km / h) sampled by the on-board speed sensor at a period of 100 ms and the differential calculated acceleration a (m / s2), and then the running state of the train is determined according to the real-time speed and acceleration through a determination condition.

[0135] Exemplary determination conditions can include: acceleration state (start): v≥0 and a≥+0.08 m / s2; stable state: |a|<0.08 m / s2 and v≥2 km / h; deceleration state (braking): a≤-0.08 m / s2; stop state: v<2 km / h and lasting for more than 3 s.

[0136] In some embodiments, the train state detection model is:

[0137] ;

[0138] wherein, is the current running condition of the train, is the train speed, is the acceleration;

[0139] S5, determining the first abnormality judgment parameter based on the corresponding physical model according to the current running condition of the train;

[0140] In some embodiments, the first abnormality judgment parameter can be a real-time deviation index for describing the temperature characteristics of the target monitoring area.

[0141] In some embodiments, the first abnormality parameter can include: a theoretical-actual temperature characteristic deviation degree.

[0142] In some embodiments, when the train starts, the relative speed between the slide plate and the contact line rises rapidly from zero, the friction generates heat sharply, resulting in a sharp rise in temperature; when the train runs at a constant speed, the contact resistance generates heat relatively smoothly, the slide plate and the contact line maintain stable sliding friction, the heat generation and heat dissipation gradually balance, the temperature reaches dynamic equilibrium, and the change range is small; when the train is in the parking / braking stage, the emergency braking may cause the instantaneous pressure between the slide plate and the contact line to increase, and the local temperature rise caused by the instantaneous extrusion of the slide plate and the contact line. Further, when an abnormal event such as short circuit occurs in the contact line, the temperature will also change significantly. Therefore, according to the energy coupling characteristics of the train operating conditions, the current operating conditions of the train can be divided into traction acceleration, steady state, braking deceleration and other research and judgment scenes, and corresponding physical models can be established respectively, which can effectively reduce the possibility of false positives and false negatives.

[0143] In some embodiments, the calculation method of the first abnormality judgment parameter is:

[0144] ;

[0145] Wherein, is the first abnormality judgment parameter, is the dynamic temperature characteristic deviation, is the theoretical temperature difference, is the actual temperature difference, is the adjustment factor;

[0146] is the static temperature characteristic deviation, is the instantaneous temperature change rate, is the temperature rise speed threshold, is the attenuation function;

[0147] In some embodiments, the actual temperature difference can be calculated by the temperature rise amplitude, in some embodiments, the actual temperature difference can be calculated from the time when the train enters the current acceleration / deceleration scene as the starting point and automatically clears at the end of the operating condition, in some embodiments, the actual temperature difference can be calculated from the first temperature sequence of the target monitoring area in the first time period.

[0148] In some embodiments, the calculation method of the theoretical temperature difference can be calculated by first calculating the total heat generation, wherein the total heat generation is equal to the sum of the friction work and the Joule work, then converting the total heat generation into the temperature rise of the carbon slide plate, and then multiplying the empirical correction coefficient of the speed / acceleration on the heat dissipation. Calculate.

[0149] In some embodiments, The adjustment factor can be set by historical data and database.

[0150] In some embodiments, the calculation method of the theoretical temperature difference is:

[0151] ;

[0152] wherein, is the theoretical temperature difference, is the friction coefficient, is the contact pressure, is the displacement amount under the current running condition of the train, is the duration of the current running condition of the train, is the current size, is the pantograph-catenary contact resistance, is the specific heat capacity of the carbon slide plate, is the effective mass of the carbon slide plate, is the acceleration correction coefficient, is the speed correction coefficient;

[0153] In some embodiments, the displacement amount under the current running condition of the train and the duration of the current running condition of the train are both calculated from the time when the train enters the current acceleration / deceleration scenario as the starting point, and are automatically cleared at the end of the running condition.

[0154] In some embodiments, the friction coefficient , the contact pressure and the pantograph-catenary contact resistance can be obtained by design standards, and the current size can be obtained by design standards or measured by a current sensor.

[0155] In some embodiments, the effective mass of the carbon slide plate can be obtained by pre-setting, and is usually in the range of 0.3-0.4 kg. In some embodiments, if the slide plate model, cross-sectional size or material is replaced, the effective mass of the carbon slide plate can be updated by re-filling.

[0156] In some embodiments, the acceleration correction coefficient and the speed correction coefficient can be set based on historical data according to the acceleration interval or the speed interval.

[0157] In some embodiments, the instantaneous temperature change rate can be defined as the instantaneous change rate of temperature with time, i.e. the derivative of temperature with respect to time. In some embodiments, the instantaneous temperature change rate can be calculated from the first temperature sequence of the target monitoring area in the first time period, and is calculated by the quotient of the temperature rise amplitude and the temperature rise time.

[0158] In some embodiments, the temperature rise speed threshold The temperature rise rate threshold value can be set by historical data, a database, or the like. The temperature rise rate threshold value can be set according to the current value of the current. The temperature rise rate threshold value can be set according to the current value of the current.

[0159] It should be noted that, for the convenience of subsequent calculation, the temperature feature deviation degree can be normalized to ensure that the temperature feature deviation degree .

[0160] In some more preferred embodiments, the attenuation function is calculated as follows:

[0161] ;

[0162] wherein, e is the base of the natural logarithm, t is the duration of the current scenario, and τ is the decay time constant.

[0163] In some embodiments, by setting the attenuation function , in the initial stage of parking, the sudden decrease in wind speed after the train stops can be suppressed, and the negative temperature drift caused by the exponential decrease of the carbon slide heat dissipation capacity with time is misjudged as an abnormality, thereby reducing the false alarm rate. On the other hand, after parking, early abnormal temperature rise can be captured in time to ensure that the risk of missing the potential poor contact or micro-arc fault approaches zero, thereby improving the detection sensitivity.

[0164] S6, based on the first temperature sequence of the target monitoring area in the first time period, predicting a second temperature sequence of the target monitoring area in a second time period through a temperature prediction model;

[0165] In some embodiments, the temperature prediction model is a model for describing the temperature variation trend of the contact line abutting region of the pantograph slide plate.

[0166] In some embodiments, the second time period can be a future time period. For example, the second time period can be a time period of 10S in the future. In some embodiments, the second time period can be a time period after the first time period.

[0167] The second temperature sequence is a sequence obtained by arranging the second temperature values of at least one time in the second time period in chronological order. For example, the second temperature sequence can include second temperature value 1, second temperature value 2, second temperature value N, and the like in a future time period. In some embodiments, the second temperature sequence includes temperature values of at least one future time. In some embodiments, the second temperature sequence can be obtained by processing the first geological parameter sequence through mathematical fitting, artificial intelligence, or the like.

[0168] In some embodiments, the predicting, by the temperature prediction model, the second temperature sequence of the target monitoring area in the second time period comprises:

[0169] S601, obtaining a plurality of temperature prediction models;

[0170] The temperature prediction model can be a model reflecting the change of temperature value over time. For example, the temperature prediction model can include a linear change model, a nonlinear change model (such as a power-law change model, an exponential change model, a logarithmic change model), etc. In some embodiments, the temperature prediction model can be obtained through a network, or obtained by calling from a storage device, a database, etc.

[0171] S602, for each of the plurality of temperature prediction models, determining a fitting degree between the first temperature sequence and the temperature prediction model;

[0172] The fitting degree can represent the fitting matching degree between the first temperature sequence and the temperature prediction model. For example, the higher the fitting degree of the regression curve of the temperature change and the observation value of the first temperature sequence, the higher the fitting degree can be.

[0173] S603, determining a target temperature prediction model from the plurality of temperature prediction models based on the fitting degrees between the first temperature sequence and each temperature prediction model.

[0174] In some embodiments, the temperature prediction model with the highest fitting degree can be selected from the plurality of temperature prediction models as the target temperature prediction model, and then the target temperature prediction model is used as the temperature prediction model of the target monitoring area.

[0175] S604, determining the second temperature sequence of the target monitoring area in the second time period based on the target temperature prediction model.

[0176] S7, determining a second anomaly judgment parameter based on the second temperature sequence of the target monitoring area in the second time period and the first temperature sequence of the target monitoring area in the second time period;

[0177] In some embodiments, the first temperature sequence in the second time period is a sequence obtained by arranging the first temperature values at at least one time point in time sequence. The time sequence can be time in order or time in reverse order.

[0178] In some embodiments, the first temperature value in the second time period can be obtained by using a dual-spectrum thermal imaging infrared camera to continuously image the bow net.

[0179] In some embodiments, the second anomaly judgment parameter can be a real-time deviation index for describing the temperature characteristics of the target monitoring area. In some embodiments, the second anomaly parameter includes: a predicted-actual temperature characteristic deviation degree.

[0180] In some embodiments, the second abnormality judgment parameter is calculated in the following manner:

[0181] ;

[0182] wherein, is a temperature error degree at the moment, is a real-time temperature at the moment, is a predicted temperature at the moment, is a limit temperature, is an ambient temperature; In some embodiments, the real-time temperature at the moment can be obtained according to the first temperature sequence in the second time period, the predicted temperature at the moment can be obtained according to the second temperature sequence in the second time period, and the limit temperature can be set in advance.

[0183] In some embodiments, the first abnormality judgment parameter and the second abnormality judgment parameter can be dynamically weighted and fused to calculate the comprehensive abnormality judgment value in real time. In some embodiments, in order to facilitate subsequent calculation, the first abnormality judgment parameter and the second abnormality judgment parameter can be normalized before dynamic weighting and fusion.

[0184] In some embodiments, the determination of the comprehensive abnormality judgment value based on the first abnormality judgment parameter and the second abnormality judgment parameter comprises:

[0185] S801, determining a fusion abnormality judgment value based on the first abnormality judgment parameter, the second abnormality judgment parameter and the fitting degree;

[0186] In some embodiments, the fusion abnormality judgment value is calculated in the following manner:

[0187]

[0188] In some embodiments, the fusion abnormality judgment value is calculated in the following manner:

[0189]

[0190] ;

[0191] wherein, is the fusion abnormality judgment value, is the fitting degree, is an environmental influence factor;

[0192] In some embodiments, the fitting degree can be obtained in step S603, and in some embodiments, in order to facilitate subsequent calculation, the fitting degree can be normalized to ensure that the fitting degree​​​​ .

[0193] In some embodiments, the environmental influence factor may be set according to the weather, for example, in bad environments such as rain and fog, the system can automatically obtain a larger K value, increase the prediction compensation term, and prevent the model from missing or false reporting due to sensor degradation.

[0194] S802, determine a comprehensive abnormality judgment value based on the fused abnormality judgment value, the first abnormality judgment parameter, the second abnormality judgment parameter, and the fitting degree;

[0195] In some embodiments, the comprehensive abnormality judgment value is calculated in the following manner:

[0196] ;

[0197] wherein, is the final abnormality judgment value, is a Sigmoid function;

[0198] ;

[0199] wherein, is the base of the natural logarithm, is the fitting degree;

[0200] In some embodiments, when , it indicates that the prediction model is consistent with the measured height, and thus the false positive rate is reduced by trusting the fused model ; when , the criterion can be smoothly transitioned to suppress switching oscillation; when , the product term ( ) is dominant to improve the sensitivity to transient faults.

[0201] In some embodiments, in the process of determining the comprehensive abnormality judgment value in step S8, first, through weighted fusion, the environmental sensitivity coefficient dynamically compensates for the sensor signal-to-noise ratio attenuation caused by rain, fog, and dust accumulation, and eliminates observation bias; then, through the Sigmoid continuous weight, a nonlinear decision boundary is formed in the transition zone, which synchronously improves the transient abnormality likelihood ratio and suppresses single-source noise. The above design can maintain the basic detection sensitivity in bad weather, and on the other hand, it can achieve fast response in transient fault scenarios, achieving the synergistic gain of significantly reducing the false positive rate and increasing the early warning lead time, and the whole process does not require manual switching, taking into account environmental robustness and fault sensitivity.

[0202] S9, determine abnormality warning information at at least one time point based on the comprehensive abnormality judgment value and the abnormality threshold set;

[0203] In some embodiments, the set of dynamic anomaly thresholds refers to a set of thresholds for determining whether the target monitoring area is abnormal.

[0204] In some embodiments, the abnormality warning information can be determined by comparing the abnormality determination value with the set of dynamic anomaly thresholds. When one or more data exceeds the threshold in the set of dynamic anomaly thresholds, the abnormality warning information is generated.

[0205] In some embodiments, the corresponding warning level and processing method can be set according to the threshold size in the set of dynamic anomaly thresholds. As an example, the abnormality warning information can be:

[0206]

[0207] In some embodiments, after determining the abnormality warning information, the abnormality warning information can be sent to the related monitoring device of the train. Potential safety hazards such as short circuit and fire caused by overheating of the contact line can be found in time, so that corresponding preventive measures can be taken to avoid accidents and improve the safety of the system. On the other hand, the warning information is sent in advance to provide enough time for maintenance personnel to take measures such as adjusting operating parameters and performing equipment maintenance to prevent the expansion of temperature anomalies.

[0208] S10, adjusting the monitoring model parameters based on the current operating conditions of the train;

[0209] In some embodiments, the adjusting the monitoring model parameters based on the current operating conditions of the train includes:

[0210] S1001, determining the local subgraph feature weight based on the current operating conditions of the train;

[0211] In some embodiments, the way of determining the local subgraph feature weight based on the current operating conditions of the train is:

[0212] ;

[0213] wherein, is the local subgraph feature weight, is the base subgraph feature weight, is the base of the natural logarithm, is the decay coefficient, is the temperature rise amplitude.

[0214] In some embodiments, as the actual temperature rise amplitude increases, the value of the exponential term decreases, resulting in a decrease in the value of the local subgraph feature weight, thereby reducing the local subgraph feature weight.

[0215] In some embodiments, the base subgraph feature weight and the decay coefficient can be set through historical data, a database, and the like.

[0216] In some embodiments, the decay coefficient can be set with a unit to maintain dimensional consistency in the calculation process. In some embodiments, the unit of the decay coefficient can be .

[0217] S11, based on the pantograph-catenary state monitoring information, determining the occurrence probability of the safety event through a monitoring model;

[0218] In some embodiments, the image data obtained by the visible light camera can be analyzed by the trained monitoring model to obtain the possibility of the occurrence of the safety event in the target image.

[0219] In some embodiments, the calculation method of the occurrence probability of the safety event is as follows:

[0220] ;

[0221] wherein, is the occurrence probability of the safety event, is a nonlinear activation function, is a weight matrix of the global feature, is the global feature, a graph pooling feature from a complete data set, is a weight matrix of the local feature, is the local feature, a subgraph maximum pooling feature.

[0222] Through analysis of the target image by the trained detection model, safety hazards can be effectively identified, thereby helping the emergency handling department to take early warning and prevention measures to prevent the occurrence of safety events, such as automatically lowering the pantograph.

[0223] In some embodiments, through the synergistic effect of the global feature and the local feature, more accurate and robust classification can be achieved in complex scenarios (such as video event detection), thereby improving the accuracy of pantograph-catenary state monitoring.

[0224] S12, based on the occurrence probability of the safety event and the dynamic probability threshold set, determining safety alarm information at at least one time point;

[0225] In some embodiments, the dynamic probability threshold set refers to a threshold set for determining whether a safety event exists.

[0226] In some embodiments, the dynamic probability threshold set can be set based on the current operating condition of the train.

[0227] In some embodiments, the calculation method of the dynamic probability threshold set is as follows:

[0228] ;

[0229] wherein, is the adjusted probability threshold, is the basic probability threshold, is the temperature rise amplitude, is the adjustment coefficient.

[0230] In some embodiments, the basic probability threshold, the adjustment coefficient can be set by historical data, database.

[0231] In some embodiments, the adjustment coefficient is set in units to maintain dimensional consistency in the calculation process, and in some embodiments, the adjustment coefficient has a unit of .

[0232] In some embodiments, the safety alarm information can be confirmed by comparing the occurrence probability and the dynamic probability threshold set. For example, the dynamic probability threshold set is generated according to the current running condition of the train, and the occurrence probability and the dynamic probability threshold set are compared, and when one or more data exceeds the threshold in the dynamic probability threshold set, the safety alarm information is generated.

[0233] In some embodiments, after determining the safety alarm information, the safety alarm information can be sent to the related monitoring device of the train, and / or the emergency handling department is activated, the bow is automatically lowered, so that the processing speed and efficiency of responding to safety events are greatly improved, the safety performance and disaster prevention ability are enhanced, and further escalation of safety events is avoided.

[0234] In some embodiments, during the starting, stopping and braking of the train, due to the changes of current, contact pressure, contact resistance and other factors, the contact between the contact wire and the slide plate is prone to be poor, thereby generating sparks and other abnormal conditions. By reducing the local subgraph feature weight and dynamically improving the probability threshold during the starting, stopping and braking of the train, the possibility of false positives is reduced, and the effect of further improving the bow net state monitoring accuracy is realized.

[0235] The second aspect of the present application provides a bow net state monitoring system for executing the aforementioned bow net state monitoring method, comprising:

[0236] a training module for training a monitoring model;

[0237] an acquisition module for acquiring bow net state monitoring information;

[0238] a first determination module for determining a first temperature sequence of the target monitoring area in a first time period;

[0239] a judgment module configured to judge a current operation condition of the train;

[0240] a first abnormality judgment parameter calculation module configured to determine a first abnormality judgment parameter;

[0241] a prediction module configured to predict a second temperature sequence of the target monitoring area in a second time period;

[0242] a second abnormality judgment parameter calculation module configured to determine a second abnormality judgment parameter;

[0243] a comprehensive calculation module configured to determine a comprehensive abnormality judgment value according to the first abnormality judgment parameter and the second abnormality judgment parameter;

[0244] a warning module configured to determine abnormality warning information;

[0245] an adjustment module configured to adjust a monitoring model parameter;

[0246] a second determination module configured to determine a probability of occurrence of a safety event;

[0247] an alarm module configured to determine safety alarm information.

[0248] In some embodiments, Figure 2 is a circuit diagram of the image acquisition submodule of the acquisition module. Wherein, VEDIO represents video input, 0.1u, 510pF represent capacitor size respectively, 10K, 680K represent resistance size, LM1881 is a chip, COMP VIDIN represents composite video input, RST is a reset pin, GND represents ground wire, VCC represents power supply, CPMP SYNC O represents composite synchronization output, VERT SYNC O represents field synchronization output, ODD / EVEN O table odd / even field indication, VERT SYNC O represents field synchronization output, BACK PORCH O is a back shoulder reference level output.

[0249] In some embodiments, Figure 3 is a circuit diagram of the temperature alarm submodule in the alarm module, Figures 4-6 is Figure 3 is an enlarged view of each part.

[0250] Wherein, Figure 4In the figure, 8051 represents a chip, P1.0-1.7 represents a general I / O port P1, RST represents a reset input, P3.0 represents RXD serial port receiving, P3.1 represents RXD serial port sending, P3.2 represents INT0 external interrupt 0, P3.3 represents INT1 external interrupt 1, P3.4 represents T0 timer 0 external counting input, P3.5 represents T1 timer 1 external counting input, P3.6 represents WR external RAM write gating, P3.7 represents RD external RAM read gating, XTAL2 / XTAL1 represents a crystal oscillator, GND represents a ground wire, P2.0-2.7 represents a general I / O port P2, PSEN represents an external program memory gating, ALE / PROG EA=1 represents starting from an internal ROM; EA=0 forces an external ROM, P0.0-0.7 represents a general I / O port P0, and VCC represents power supply.

[0251] Figure 5 In the figure, LCD 1602 represents a liquid crystal display module, GND represents a ground wire, VCC represents a power supply, VO represents a contrast adjustment voltage, RS represents an instruction register / data register, RW represents reading / writing, E represents enabling, D0-D7 represent an 8-bit parallel data port, A represents a backlight anode, K represents a backlight cathode, RP1 represents a row resistance, and RESP represents a reset button.

[0252] Figure 6 In the figure, VCC represents a power supply, R3 represents a base limiting resistor, Q1 (PNP) represents a PNP triode, Res2 represents a base pull-down resistor, LS1 (speaker) represents a loudspeaker / buzzer, GND represents a ground wire,

[0253] The third aspect of the present application provides a computer readable storage medium, which stores computer instructions, when the computer reads the computer instructions in the storage medium, the computer executes the catenary state monitoring method as described above.

[0254] The above has described the basic concept, and it is obvious that the above detailed disclosure is only used as an example and does not constitute a limitation on the present specification. Although it is not explicitly stated here, those skilled in the art can make various modifications, improvements and corrections to the present specification. Such modifications, improvements and corrections are suggested in the present specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of the present specification.

[0255] Finally, it should be understood that the embodiments described in the specification are only used to illustrate the principles of the embodiments of the specification. Other variations can also belong to the scope of the specification. Therefore, as an example but not limitation, alternative configurations of the embodiments of the specification can be considered consistent with the teachings of the specification. Accordingly, the embodiments of the specification are not limited to the embodiments explicitly introduced and described in the specification.

Claims

1. A pantograph state monitoring method, characterized by, The method comprises the following steps: training a monitoring model to identify a safety event; obtaining pantograph-catenary state monitoring information; determining a first temperature sequence of a target monitoring area in a first time period based on the pantograph-catenary state monitoring information; judging a current running condition of a train based on a train state monitoring model; determining a first abnormality judgment parameter based on the corresponding physical model according to the current running condition of the train, the first abnormality judgment parameter comprising a theoretical-actual temperature characteristic deviation degree; predicting a second temperature sequence of the target monitoring area in a second time period based on the first temperature sequence of the target monitoring area in the first time period through a temperature prediction model; determining a second abnormality judgment parameter based on the second temperature sequence of the target monitoring area in the second time period and the first temperature sequence of the target monitoring area in the second time period, the second abnormality parameter comprising a predicted-actual temperature characteristic deviation degree; determining a comprehensive abnormality judgment value based on the first abnormality judgment parameter and the second abnormality judgment parameter; determining abnormal early warning information of at least one time point based on the comprehensive abnormality judgment value and an abnormal threshold set; adjusting monitoring model parameters based on the current running condition of the train; determining a probability of occurrence of a safety event based on the pantograph-catenary state monitoring information through the monitoring model; determining safety warning information of at least one time point based on the probability of occurrence of the safety event and a dynamic probability threshold set. The training of the monitoring model to identify a safety event comprises: defining knowledge graph entities and relationships between the entities; constructing a training graph based on a safety event set; training the monitoring model to extract global features based on the training graph; training the monitoring model to extract key features based on a subgraph of the training graph.

2. A pantograph line condition monitoring method according to claim 1, characterized in that, The determination of the first temperature sequence of the target monitoring area in the first time period comprises: determining a first temperature value of the target monitoring area at at least one time point in the first time period; the calculation method of the first temperature value is: ； wherein, is a first temperature value at a time instant, is a weight of the temperature data in the temperature matrix, is a temperature value at a time instant, is a temperature value at a time instant, is a number of temperature values in the temperature matrix; determining the first temperature sequence of the target monitoring area in the first time period based on the first temperature value at at least one time point in the first time period.

3. A pantograph line condition monitoring method according to claim 2, characterised in that, The calculation method of the first abnormality judgment parameter is: ； wherein is a first abnormality determination parameter, is a dynamic temperature characteristic deviation degree, is a theoretical temperature difference, is an actual temperature difference, is an adjustment factor; is a static temperature characteristic deviation, is an instantaneous temperature change rate, is a temperature rise speed threshold value, is a decay function; The attenuation function The calculation method is as follows: ； wherein, is the base of the natural logarithm, is the duration of the current scene, is the decay time constant.

4. A pantograph line condition monitoring method according to claim 3, characterised in that, The prediction of the second temperature sequence of the target monitoring area in the second time period through the temperature prediction model comprises: obtaining a plurality of temperature prediction models; for each of the plurality of temperature prediction models, determining a fitting degree between the first temperature sequence and the temperature prediction model; determining a target temperature prediction model from the plurality of temperature prediction models based on the fitting degrees between the first temperature sequence and each temperature prediction model; determining the second temperature sequence of the target monitoring area in the second time period based on the target temperature prediction model.

5. A pantograph line condition monitoring method according to claim 4, characterised in that, The calculation method of the second abnormality judgment parameter is: ； wherein, is the temperature error degree at the moment, is the real-time temperature at the moment, is the predicted temperature at the moment, is the limit temperature, is the ambient temperature.

6. A pantograph line condition monitoring method according to claim 5, characterised in that, The determination of the comprehensive abnormality judgment value based on the first abnormality judgment parameter and the second abnormality judgment parameter comprises: determining a fusion abnormality judgment value based on the first abnormality judgment parameter, the second abnormality judgment parameter, and the fitting degree; The calculation method of the fusion abnormality judgment value is: ； wherein, is a fusion abnormality determination value, is a fitting degree, is an environmental influence factor; determining the comprehensive abnormality judgment value based on the fusion abnormality judgment value, the first abnormality judgment parameter, the second abnormality judgment parameter, and the fitting degree; The calculation method of the comprehensive abnormality judgment value is: ； wherein, is the final abnormality determination value, is a sigmoid function; ； wherein, is the base of the natural logarithm, is the goodness of fit.

7. The pantograph line condition monitoring method of claim 1, wherein, The adjustment of the monitoring model parameters based on the current running condition of the train comprises: Determine the local subgraph feature weight based on the current running condition of the train; The manner of determining the local subgraph feature weight based on the current running condition of the train is: ； wherein, is a local subgraph feature weight, is a base subgraph feature weight, is a base of the natural logarithm, is a decay coefficient, temperature rise amplitude.

8. A pantograph line condition monitoring method according to claim 7, characterised in that, The manner of calculating the occurrence probability of the safety event is: ； wherein, is a probability of occurrence of a safety event, is a non-linear activation function, is a weight matrix of global features, is a global feature, a graph pooling feature from a complete dataset, is a weight matrix of local features, is a local feature, a subgraph max pooling feature; The manner of calculating the dynamic probability threshold set is: ； wherein, is the adjusted probability threshold, is the base probability threshold, is the temperature rise amplitude, is the adjustment coefficient.

9. A pantograph condition monitoring system, characterized by A pantograph-catenary state monitoring method according to any one of claims 1-8, comprising: A training module for training a monitoring model; An acquisition module for acquiring pantograph-catenary state monitoring information; A first determination module for determining a first temperature sequence of the target monitoring area in a first time period; A research and judgment module for researching and judging the current running condition of the train; A first abnormality judgment parameter calculation module for determining a first abnormality judgment parameter; A prediction module for predicting a second temperature sequence of the target monitoring area in a second time period; A second abnormality judgment parameter calculation module for determining a second abnormality judgment parameter; A comprehensive calculation module for determining a comprehensive abnormality judgment value according to the first abnormality judgment parameter and the second abnormality judgment parameter; An early warning module for determining abnormality early warning information; An adjustment module for adjusting the monitoring model parameters; A second determination module for determining the occurrence probability of the safety event; An alarm module for determining safety alarm information; The training monitoring model comprises: Defining knowledge graph entities and the relationships between the entities; Constructing a training graph based on a safety event set; Training the monitoring model to extract global features based on the training graph; Training the monitoring model to extract key features based on the training graph subgraph.

Images