Dynamic cooperative control method for regenerative energy of traction power supply system fused with power prediction

Abstract

The invention provides a dynamic cooperative control method for regenerative energy of a traction power supply system fused with power prediction, and relates to the technical field of traction power supply systems, and the method comprises the steps: obtaining a plurality of train position vectors of a plurality of running trains; executing dual-thread asynchronous matching to obtain a plurality of line environment characteristic parameters and a plurality of multi-source sensor dynamic fusion weights; multiple pieces of train operation state data are collected; constructing a plurality of spatio-temporal joint feature vectors; working condition migration probability prediction is carried out, and a plurality of ultra-short-term net power demands are output; power supply interval topology affiliation prediction is carried out, and a plurality of power supply interval topology codes are positioned; and according to the space-time coupling relationship, executing the dynamic closed-loop cooperative control of the regenerated energy of the far-end traction power supply system. The technical problem that a traction power supply system in the prior art depends on a simple load distribution rule and cannot respond to severe load fluctuation in time, so that power distribution is uneven, and the stability of a power grid is affected is solved.

Description

Technical Field

[0001] This invention relates to the field of traction power supply system technology, and in particular to a dynamic collaborative control method for regenerative energy in traction power supply systems based on fusion power prediction. Background Technology

[0002] In urban rail transit systems, the core function of the traction power supply system is to provide power to operating trains, ensuring smooth traction and braking. When a train is running, the traction substation transmits electrical energy to the overhead contact line through a bidirectional converter, which then supplies power to the train. When the train brakes, regenerative energy is generated due to inertia, and this braking energy is usually fed back to the power grid through a bidirectional converter. However, the current power grid in urban rail transit systems is not very responsive to this regenerative energy feedback. The main reason is that the fed-back regenerative energy contains significant harmonics, which may cause voltage fluctuations and power imbalances in the power grid, thus affecting the stability and power quality of the grid.

[0003] Furthermore, when the power grid load is too high or the electrical energy supplied by the rail transit system cannot be fully absorbed, the feedback regenerative energy may not only overload the power grid but also prevent the train's braking system from activating in time, creating potential safety hazards. Specifically, under traction conditions, if the load is high and the output of nearby substations is insufficient, the contact wire voltage will drop, leading to reduced train power operation. Under braking conditions, if the absorption capacity of nearby substations is insufficient, braking energy will be fed back to the upstream power grid. Due to the excessive instantaneous power, the grid often cannot absorb it in time, causing the train's protection system to activate and preventing further braking. This series of problems not only affects the stability of the system but also wastes potential regenerative energy.

[0004] It should be noted that the information disclosed in this background section is intended only to enhance the understanding of the overall background of the present invention, and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] In response to the above-mentioned defects or improvement needs of existing technologies, this invention provides a dynamic collaborative control method for regenerative energy in traction power supply systems that integrates power prediction. This method solves the technical problem that existing traction power supply systems rely on simple load distribution rules, which cannot respond in a timely manner to drastic load fluctuations, resulting in uneven power distribution and thus affecting the stability of the power grid.

[0006] The specific technical solution is as follows:

[0007] According to a first aspect of the present invention, a method for dynamic coordinated control of regenerative energy in a traction power supply system based on fused power prediction is provided, the method comprising:

[0008] Dynamic train pose correction is performed based on incremental sensing data from the trackside end to obtain multiple train position vectors for multiple operating trains. Dual-thread asynchronous matching is then performed based on these multiple train position vectors to obtain multiple line environment feature parameters and dynamic fusion weights from multiple multi-source sensors. Multiple train operating status data are collected through multiple onboard heterogeneous sensing units pre-deployed on the multiple operating trains, including speed vectors, current direction indicators, and IMU inertial measurement outputs. The multiple train operating status data are fused and stitched together based on the dynamic fusion weights from the multiple multi-source sensors to construct multiple spatiotemporal joint feature vectors. Using the multiple line environment feature parameters as correction factors, operating condition migration probability prediction is performed based on the multiple spatiotemporal joint feature vectors to output multiple ultra-short-term net power demands. Power supply section topology assignment prediction is performed based on the multiple train operating status data and multiple train position vectors to locate multiple power supply section topology codes. Based on the spatiotemporal coupling relationship between the multiple power supply section topology codes and multiple ultra-short-term net power demands, dynamic closed-loop collaborative control of regenerative energy in the remote traction power supply system is executed.

[0009] In one implementation, dynamic correction of train pose is performed based on incremental sensing data from the trackside end to obtain multiple train position vectors for multiple operating trains, including:

[0010] An RFID positioning subarray is constructed by deploying RFID readers / writers along the trackside posts of the bidirectional track at a predetermined deployment density. A catenary tension monitoring subarray is constructed by deploying fiber optic strain sensors along the bidirectional track positioning and installation node array based on predetermined key mechanical node rules. An optical ranging subarray is constructed by uniformly deploying heterogeneous optical sensors along the unidirectional track based on track spatial attributes. A heterogeneous sensing spatiotemporal synchronization protocol is used to drive the RFID positioning subarray, catenary tension monitoring subarray, and optical ranging subarray in parallel, collecting RFID position beacon signals, catenary tension changes, and optical ranging data. Multiple initial positioning coordinates of multiple running trains are analyzed based on the RFID position beacon signals. The catenary tension changes and optical ranging data are fused to perform dynamic pose-assisted correction of the multiple initial positioning coordinates, resulting in multiple train position vectors.

[0011] In one implementation, dual-thread asynchronous matching is performed based on the plurality of train position vectors to obtain multiple line environment feature parameters and multiple multi-source sensor dynamic fusion weights, including:

[0012] A pre-built dual-threaded processing architecture is provided, comprising a geographic information retrieval engine and a scene adaptation engine. The geographic information retrieval engine projects multiple train position vectors onto an electronic map and retrieves multiple line environment feature parameters, including real-time gradient, curve radius, and line impedance parameters. The scene adaptation engine, after matching multiple operating scene types based on the multiple train position vectors, interacts with a pre-set scene weight configuration table to extract the dynamic fusion weights of multiple multi-source sensors. The geographic information retrieval engine and the scene adaptation engine execute in parallel and asynchronously.

[0013] In one implementation, the catenary tension change and optical ranging data are fused to perform dynamic pose-assisted correction of the multiple initial positioning coordinates, resulting in the multiple train position vectors, including:

[0014] Based on the spatial topology mapping relationship of the RFID positioning subarray, the catenary tension monitoring subarray, and the optical ranging subarray, multiple tension gradient distributions and multiple optical reference distance deviations are segmented from the catenary tension change and optical ranging data using the multiple initial positioning coordinates. Multiple lateral offset vectors are generated by reverse calculation based on the suspension point topology relationship of the multiple tension gradient distributions in the catenary tension monitoring subarray. Multiple longitudinal compensation vectors are generated based on the spatial geometric relationship of the multiple optical reference distance deviations in the electronic map. Dynamic pose coupling correction of the multiple initial positioning coordinates is performed using the multiple lateral offset vectors and multiple longitudinal compensation vectors to obtain the multiple train position vectors.

[0015] In one implementation, the plurality of line environmental characteristic parameters are used as correction factors, and operating condition migration probability is predicted based on the plurality of spatiotemporal joint feature vectors to output multiple ultra-short-term net power demands. Prior to this, the following steps are included:

[0016] A discrete operating condition state space is constructed based on multi-dimensional train operating condition states. Under operating condition transition constraints, the multi-dimensional train operating condition states are combined and enumerated to obtain Y sample operating condition transition vectors. Y sets of sample joint feature vectors of the Y sample operating condition transition vectors are retrieved locally. Y baseline transition probabilities are calculated based on the recurrence frequency of the Y sets of sample joint feature vectors. The Y sets of sample joint feature vectors are used as training data to construct a Y operating condition transition probability prediction model based on an LSTM model. The Y baseline transition probabilities are used as the initial probability distribution, and the Y operating condition transition probability prediction model is used as a dynamic corrector to fill the probability matrix of the discrete operating condition state space, thus completing the construction of the dynamic probability evolution model.

[0017] In one implementation, the plurality of line environmental characteristic parameters are used as correction factors, and operating condition migration probability is predicted based on the plurality of spatiotemporal joint feature vectors to output multiple ultra-short-term net power demands, including:

[0018] The system interactively obtains the first real-time operating condition status of the first operating train; based on the first real-time operating condition status, after setting state transition constraints in the dynamic probability evolution model, it loads the first spatiotemporal joint feature vector into the dynamic probability evolution model to perform operating condition transition probability prediction, obtaining multiple time-series predicted feature vectors under multiple predicted transition operating condition probabilities; it smoothly concatenates the first spatiotemporal joint feature vector and multiple time-series predicted feature vectors to obtain multiple multimodal spatiotemporal feature vectors; it performs power compensation calculation based on the multiple line environment feature parameters and multiple multimodal spatiotemporal feature vectors to obtain multiple initial net power demands; after serializing the multiple initial net power demands based on the operating condition risk attributes, it performs risk probability weighted sorting correction based on the multiple predicted transition operating condition probabilities to select the first ultra-short-term net power demand.

[0019] In one implementation, power supply section topology attribution prediction is performed based on the multiple train operation status data and multiple train position vectors, and multiple power supply section topology codes are located, including:

[0020] The electronic map is segmented based on the power supply interval boundaries to obtain distributed power supply grid units, where each power supply grid unit is bound to a unique power supply interval code. Multiple IMU inertial measurement outputs from the multiple train operation status data are parsed to obtain multiple initial acceleration vectors. Multiple IMU inertial prediction outputs are extracted from the multiple time-series prediction feature vectors to obtain multiple incremental acceleration vectors. Multiple velocity vectors and multiple current direction identifiers are extracted from the multiple train operation status data. After calculating the basic displacement distance based on the multiple velocity vectors, multiple initial acceleration vectors, and multiple incremental acceleration vectors, the operating condition constraints are corrected by combining the multiple current direction identifiers to output multiple ultra-short-term displacement distances. Multiple updated position coordinates are predicted based on the multiple ultra-short-term displacement distances and multiple train position vectors. The multiple updated position coordinates are mapped to the distributed power supply grid units to locate the multiple power supply interval topology codes.

[0021] In one implementation, based on the spatiotemporal coupling relationship between the multiple power supply section topology codes and multiple ultra-short-term net power demands, dynamic closed-loop coordinated control of regenerative energy in the remote traction power supply system is performed, including:

[0022] Based on the topology coding of the multiple power supply sections, the multiple ultra-short-term net power demands are aggregated to obtain P sets of ultra-short-term net power demands corresponding to P train power supply sections in the distributed power grid unit; the P sets of ultra-short-term net power demands are summed to obtain P ultra-short-term predicted total loads; the working mode of the traction substations in the P train power supply sections is adjusted and the load ratio is allocated according to the P ultra-short-term predicted total loads.

[0023] In one implementation, adjusting the operating mode and allocating the load ratio of traction substations in the P train power supply sections based on the P ultra-short-term predicted total loads includes:

[0024] Based on the first group of ultra-short-term net power demand, the first updated position coordinates of the first group of operating trains are retrieved in reverse order; the first set of double-end spatial distance values ​​between the double-end traction substations in the power supply section of the first train and the first set of updated position coordinates are calculated; if the first ultra-short-term predicted total load is negative, the double-end traction substations are switched to inverter mode, and the inverter load of the first group of operating trains is pre-intervention in a proportional allocation based on the first set of double-end spatial distance values; if the first ultra-short-term predicted total load is positive, the double-end traction substations are switched to rectification mode, and the rectification load of the first group of operating trains is pre-intervention in a proportional allocation based on the first set of double-end spatial distance values.

[0025] In one implementation, it further includes:

[0026] If the first ultra-short-term predicted total load is positive and exceeds the maximum rectification capacity of the dual-end traction substation, the rectification function of the adjacent power supply section traction substation is activated to provide power support; if the first ultra-short-term predicted total load is negative and exceeds the absorption capacity of the dual-end traction substation, the inverter function of the adjacent power supply section traction substation is activated to absorb braking energy in coordination.

[0027] Beneficial effects of the embodiments of the present invention:

[0028] Dynamic correction of train posture using incremental sensor data from the trackside end enables accurate determination of the real-time positions of multiple trains. This process ensures precise train tracking during operation, providing a reliable data foundation for subsequent load allocation and energy regulation. Dual-thread asynchronous matching based on train position vectors efficiently and accurately extracts track environment features and dynamically fuses weights from electronic maps and multi-source sensors. By processing large amounts of data synchronously, it can quickly respond to changing operating environments, adjust control strategies in a timely manner, and improve the system's responsiveness and adaptability. Collecting train operating status data from multiple onboard sensors provides detailed operating parameters for each train. The comprehensiveness and high precision of this information provide accurate input for subsequent dynamic control. Dynamic fusion of multi-source sensor data generates a spatiotemporal joint feature vector, which more accurately reflects the train's operating status and its interaction with the environment, further enhancing the perception of train operating conditions. Based on multi-source fused data, combined with track loop... Using environmental characteristics as a correction factor, the system predicts the ultra-short-term net power demand of trains through a probabilistic prediction model of operating condition migration. Based on the current status and future operating conditions of the trains, it can accurately calculate the upcoming power demand, promptly capture the changing trends of the power grid load, and provide data support for subsequent power allocation. Through train operating status and location data, it accurately predicts the power supply section and its topology for each train, effectively supporting refined management of power demand. This allows the power supply system to dynamically adjust power supply sections based on real-time data, ensuring load balance in each power supply section. By collecting real-time information on train status, location, and power demand, combined with the topology coding of power supply sections, it enables coordinated energy scheduling between different power supply sections. It can dynamically adjust the power supply mode based on the net power demand and power supply capacity of each area, ensuring the effective absorption and distribution of renewable energy. By optimizing power flow and dynamically adjusting, the power grid can improve system energy efficiency and operational reliability while ensuring stable train operation.

[0029] Of course, implementing any product or method of the present invention does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This invention provides a schematic flowchart of the dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction.

[0032] Figure 2 The diagram illustrates the process of dynamic train posture correction using the dynamic collaborative control method for regenerative energy in traction power supply systems based on fusion power prediction provided by this invention. Detailed Implementation

[0033] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0034] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0035] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0036] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0037] The present invention provides a dynamic collaborative control method for regenerative energy in traction power supply systems based on integrated power prediction. This method addresses the technical problem that existing traction power supply systems rely on simple load distribution rules, which cannot respond promptly to drastic load fluctuations, leading to uneven power distribution and consequently affecting the stability of the power grid.

[0038] Example 1: See Figure 1 The flowchart of the dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction provided in this embodiment of the invention includes:

[0039] Y100: Based on incremental sensing data from the trackside end, perform dynamic correction of train posture to obtain multiple train position vectors for multiple operating trains.

[0040] Trackside incremental sensing data refers to data collected by sensors arranged along the track, such as RFID, fiber optic grating sensors, and overhead contact line tension sensors. These sensors monitor the relative position of the train and the track in real time, or other relevant data, such as changes in the tension of the overhead contact line. Using the data provided by these incremental sensors, the train's position is dynamically corrected to obtain accurate position information. The resulting train position vector includes the train's longitudinal and lateral positions, as well as its tilt or rotation angle.

[0041] Y200: Performs dual-thread asynchronous matching based on the multiple train position vectors to obtain multiple line environment feature parameters and multiple multi-source sensor dynamic fusion weights.

[0042] Dual-thread asynchronous matching means that the processing involves two parallel threads, each handling different tasks. One thread acquires and processes the train's track position data, while the other processes related environmental features such as gradient and curve radius. These two threads run asynchronously in parallel, improving computational efficiency. Specifically, based on the train's position vector, the specific track segment where the train is located is first determined. This can be based on coordinates, using an electronic map or track database. The track position is then matched with corresponding track environmental feature parameters, such as real-time gradient, curve radius, and track impedance. Simultaneously, another thread calculates and updates the dynamic fusion weights of the sensors using a matching model based on the train's dynamic state data, such as speed and acceleration. These weights indicate the reliability of different sensors or their impact on the final prediction result.

[0043] Y300: Collects multiple train operation status data through multiple on-board heterogeneous sensing units pre-deployed on multiple operating trains, wherein the train operation status data includes speed vector, current direction identifier and IMU inertial measurement output.

[0044] Onboard heterogeneous sensing units refer to various sensors installed on the train, such as speed sensors, IMU (Inertial Measurement Unit) sensors, accelerometers, and current sensors. These sensors can provide multi-dimensional information such as the train's speed, acceleration, traction current direction, and tilt angle. The velocity vector represents the train's speed information in vector form, including both magnitude and direction. The current direction indicator shows the direction of the train's traction current, helping to determine whether the train is in a traction, braking, or stationary state. The IMU inertial measurement output provides information about the train's acceleration, angular velocity, and direction changes, used to calculate the train's dynamic state.

[0045] Y400: Based on the dynamic fusion weights of the multiple multi-source sensors, the multiple train operation status data are fused and spliced ​​to construct multiple spatiotemporal joint feature vectors.

[0046] The data from each sensor has different importance during the fusion process, and the weight of each sensor is adjusted based on the actual operating environment and dynamic state. For example, the speed sensor has a larger weight when traveling at high speeds, while the acceleration data from the IMU is more important when traveling at low speeds. By weighted fusion of data collected from multiple sensors, a spatiotemporal joint feature vector containing temporal and spatial information is generated. For example, a spatiotemporal joint feature vector contains information such as the train's current position, speed, acceleration, and traction current direction. These feature vectors provide important data input for subsequent prediction of operating condition migration probabilities.

[0047] Y500: Using the multiple line environment characteristic parameters as correction factors, and based on the multiple spatiotemporal joint feature vectors, perform operating condition migration probability prediction and output multiple ultra-short-term net power demands.

[0048] Line environmental characteristics, such as gradient and curve radius, affect train operating conditions. Based on these characteristics, the spatiotemporal joint feature vector is corrected to ensure that the prediction results are more consistent with the actual operating conditions. The prediction model is used to predict the probability of operating condition migration based on these feature vectors and correction factors, and finally outputs the ultra-short-term net power demand. Net power demand refers to the balance between traction power and braking energy.

[0049] Y600: Based on the multiple train operation status data and multiple train position vectors, predict the topology affiliation of the power supply section and locate the topology codes of multiple power supply sections.

[0050] A power supply section consists of multiple substations or power supply equipment. Trains rely on different power sources at different locations. By predicting the train's operating status and location, it is possible to infer when and where the train will enter or leave a power supply section. Through electronic maps or track databases, each power supply section is assigned a unique topology code. Based on the train's location and status, it is determined which power supply section the train belongs to, thereby achieving intelligent management and scheduling of power supply.

[0051] Y700: Based on the spatiotemporal coupling relationship between the topology codes of the multiple power supply sections and the multiple ultra-short-term net power demands, execute dynamic closed-loop collaborative control of regenerative energy of the remote traction power supply system.

[0052] The ultra-short-term net power demand varies dynamically across different power supply sections and train locations. By combining spatiotemporal factors, such as train speed, acceleration, and gradient, the power demand in different power supply sections can be predicted more accurately. Dynamic closed-loop coordinated control, based on real-time power demand prediction, dynamically adjusts the operating mode and load distribution of remote converters. If the load in a certain power supply section is too high, the load in other power supply sections will be automatically adjusted to ensure that regenerative energy is reasonably absorbed and maintain system balance. Specifically, based on spatiotemporal coupling, the power demand in different power supply sections is predicted. If the net power demand in a certain power supply section is too high, the remote converter switches to inverter mode to absorb more regenerative energy; if the demand is too low, it switches to rectification mode to ensure stable power supply. By coordinating the operating modes and power output ratios of different power supply sections, the energy balance of the entire system is ensured, avoiding excessive fluctuations in grid load, thereby improving system efficiency and stability. By combining advanced artificial intelligence technology and utilizing real-time data analysis and predictive models, dynamic optimization and intelligent scheduling can be achieved, thereby further improving the energy efficiency and stability of the traction power supply system, ensuring the rational allocation and absorption of renewable energy, and ultimately achieving the balance of grid load and the sustainable operation of the system.

[0053] In one implementation, dynamic train pose correction is performed based on incremental sensing data from the trackside end, resulting in multiple train position vectors for multiple operating trains, including:

[0054] Y110: Deploy RFID readers / writers along the trackside pillars of the bidirectional track based on a preset deployment density to construct an RFID positioning subarray; Y120: Based on preset key mechanical node rules, traverse the bidirectional track positioning and installation node array to deploy fiber optic strain sensors and construct a catenary tension monitoring subarray; Y130: Based on track spatial attributes, uniformly deploy heterogeneous optical sensors along the unidirectional track to construct an optical ranging subarray; Y140: Use a heterogeneous sensing spatiotemporal synchronization protocol to drive the RFID positioning subarray, catenary tension monitoring subarray, and optical ranging subarray in parallel, collecting RFID position beacon signals, catenary tension changes, and optical ranging data; Y150: Analyze multiple initial positioning coordinates of multiple running trains based on the RFID position beacon signals; Y160: Fuse the catenary tension changes and optical ranging data to perform dynamic pose-assisted correction of the multiple initial positioning coordinates to obtain the multiple train position vectors.

[0055] RFID is a mature wireless identification technology suitable for automated positioning and real-time monitoring. According to preset density requirements, RFID readers are installed on both sides of the bidirectional track (trackside). These RFID readers are arranged at specific intervals to form a dense RFID positioning sub-array, ensuring continuous and stable monitoring of the train's position information during operation. RFID tags are installed on the train. Whenever the train passes an RFID reader installed on a trackside post, the RFID reader automatically reads the tag information and calculates the train's real-time position. The RFID sensors transmit the train's position data to the monitoring system in real time via wireless communication, providing the basic data for subsequent dynamic positioning and track correction.

[0056] Based on the structural characteristics and mechanical models of railways, a series of key nodes are selected on the track. These nodes are typically track support points, catenary suspension points, important curves, or locations with steep gradients. Fiber optic strain sensors are installed at these nodes. These sensors can accurately measure changes in the forces acting on the track and catenary, especially changes in tension and strain. When a train passes, the sensors detect minute deformations in the track or catenary, thus reflecting the train's dynamic load. By deploying these sensors into a catenary tension monitoring subarray, the tension of the catenary and the strain of the track can be monitored in real time, allowing for the timely detection of potential equipment failures or track problems.

[0057] Based on factors such as the spatial layout of the track, track type, and train speed, the location and deployment density of optical sensors along the unidirectional track are analyzed and determined. These sensors will be evenly distributed along the unidirectional track to ensure that the measurement coverage is sufficiently wide. Heterogeneous optical sensors, such as lidar and optical sensors, are deployed. These sensors calculate the precise distance between the train and the track by emitting light beams and measuring the reflected signals. Multiple optical sensors work together to form an optical ranging subarray to acquire the distance data between the train and the track in real time.

[0058] To ensure that different types of sensors can work collaboratively and that the collected data is synchronized in time and space, a spatiotemporal synchronization protocol is employed. This protocol ensures that each sensor can acquire data within the same time window, and that its data is accurately aligned within the spatiotemporal framework. Supported by the synchronization protocol, these sensor arrays acquire data in parallel. Different sensor arrays simultaneously record data using their respective measurement methods, such as wireless RFID reading, strain detection of catenary tension changes, and beam reflection from optical ranging. RFID position beacon signals are used to obtain the train's current position and track information; catenary tension changes are used to monitor changes in the force on the catenary, determining the train's traction status and the stability of the track's power supply; and optical ranging data is used to accurately measure the distance between the train and the track or other equipment.

[0059] RFID sensors communicate wirelessly with RFID tags on the train. When the train passes a track location equipped with an RFID reader, it automatically reads the RFID tag information to obtain the train's location information. The RFID tag contains unique identification information, and the RFID reader calculates the train's location based on this information. By analyzing the signals provided by the RFID sensor, the initial positioning coordinates of the train are calculated. These position coordinates represent the spatial position of the train on the track at a certain moment, and can be represented by a map, coordinate system, or other reference system.

[0060] The information provided by the catenary tension sensor helps determine the train's traction status and load condition, while the optical distance sensor provides distance information between the train and the track. Combining these two technologies allows for the calculation of the train's dynamic state and precise position on a specific track section. After acquiring the initial position information, by fusing the catenary tension change and optical distance data, mathematical models such as Kalman filtering and least squares methods are used to dynamically correct the train's initial positioning coordinates. This corrects deviations caused by measurement errors and sensor accuracy limitations, resulting in more accurate train position information. The corrected position information forms the train position vector, which contains the train's spatial coordinates and state parameters, such as speed and acceleration, at a specific moment on the track.

[0061] In one implementation, dual-thread asynchronous matching is performed based on the multiple train position vectors to obtain multiple line environment feature parameters and multiple multi-source sensor dynamic fusion weights, including:

[0062] Y210: A pre-built dual-thread processing architecture is constructed, comprising a geographic information retrieval engine and a scene adaptation engine; Y220: The geographic information retrieval engine projects the multiple train position vectors onto an electronic map and retrieves the multiple line environment feature parameters, including real-time gradient, curve radius, and line impedance parameters; Y230: After matching multiple operating scene types based on the multiple train position vectors, the scene adaptation engine extracts the dynamic fusion weights of the multiple multi-source sensors by interacting with a preset scene weight configuration table; Y240: The geographic information retrieval engine and the scene adaptation engine are executed in parallel and asynchronously.

[0063] The geographic information retrieval engine is responsible for extracting track environment feature parameters related to the train's current location from electronic maps or geographic information systems. Specifically, based on the train's current location, the geographic information retrieval engine obtains real-time key information such as gradient, curve radius, and track impedance from the map. This information forms the basis for judging the train's operating status. The scene adaptation engine is responsible for identifying and matching the current train's operating scene based on the train's current location vector and a preset scene weight configuration table. Different operating scenes have different control requirements, so the scene adaptation engine can dynamically adjust the fusion weights of sensor data to adapt to the current operating state. The geographic information retrieval engine and the scene adaptation engine run in two independent threads, ensuring that their respective processing tasks do not interfere with each other, thereby achieving parallel processing and improving efficiency.

[0064] The geographic information retrieval engine projects multiple train position vectors onto a pre-built electronic map. This projection process maps the train's spatial location to the geographic coordinate system on the map, ensuring accurate positioning. Based on the train's location, the engine further retrieves environmental feature parameters from the surrounding area. Among these parameters, the real-time gradient describes the gradient of the current track segment, crucial for traction and braking control, especially during ascents and descents; the curve radius of the track is a key factor in determining the train's turning radius, significantly impacting traction power and braking distance; and the track impedance parameter reflects the track's electrical characteristics, affecting traction current and grid load.

[0065] The scene adaptation engine matches the train's position vector with known operating scene types, such as uphill, downhill, curves, and straight sections. Each scene presents different operating modes, power demands, and traction characteristics, requiring different processing methods. After determining the train's operating scene, the scene adaptation engine extracts sensor data fusion weights appropriate for that scene based on a pre-defined scene weight configuration table. This table defines the importance and weight of each sensor data point in different scenes; for example, on slopes, catenary tension data has a higher weight, while on curves, optical ranging data is more important. The weights are dynamically adjusted based on actual operating conditions and environmental data to reflect the train's needs and response characteristics in different environments in real time.

[0066] In one implementation, the catenary tension change and optical ranging data are fused to perform dynamic pose-assisted correction of the multiple initial positioning coordinates, resulting in the multiple train position vectors, including:

[0067] Y161: Based on the spatial topology mapping relationship of the RFID positioning subarray, the catenary tension monitoring subarray, and the optical ranging subarray, multiple tension gradient distributions and multiple optical reference distance deviations are segmented from the catenary tension change and optical ranging data using the multiple initial positioning coordinates; Y162: Multiple lateral offset vectors are generated by back-calculating the suspension point topology relationship of the multiple tension gradient distributions in the catenary tension monitoring subarray; Y163: Multiple longitudinal compensation vectors are generated based on the spatial geometric relationship of the multiple optical reference distance deviations in the electronic map; Y164: Dynamic pose coupling correction of the multiple initial positioning coordinates is performed using the multiple lateral offset vectors and multiple longitudinal compensation vectors to obtain the multiple train position vectors.

[0068] Using RFID positioning technology, the approximate location and initial coordinates of a train can be determined, which can be considered a coarse positioning. Contact wire tension sensors can detect changes in tension within the contact wire; these changes are closely related to the contact state between the train and the contact wire, as well as the train's operating status, thus helping to infer changes in the train's relative position. Optical distance sensors provide precise distance information by measuring changes in the distance between the train and a reference point, further enhancing the accuracy of the position information. Mapping the contact wire tension changes and optical distance measurement data to the initial positioning coordinates, and through spatial topological mapping, multiple tension gradient distributions and optical reference distance deviations are segmented. These data represent subtle positional changes during train operation, facilitating precise corrections based on the original positioning.

[0069] The tension gradient measured by the catenary tension monitoring subarray reflects the relative changes between the train and the power supply catenary as the train runs on the track. These changes cause alterations in the catenary tension distribution. The catenary suspension point topology refers to the spatial positions of the catenary supports and anchor points. By inversely calculating the relationship between the distances between suspension points and tension changes, the train's lateral offset can be calculated. Based on the tension gradient and the topological relationship of the suspension points, multiple lateral offset vectors are calculated, representing the train's precise left-right offset on the track. This helps correct the train's lateral position.

[0070] The reference distance deviation provided by the optical ranging sensor refers to the difference between the actual measured distance and the expected reference distance. This difference represents the train's precise longitudinal displacement, i.e., the forward and backward direction. The spatial geometry in the electronic map defines the track's shape, gradient, curves, and other features. Based on this, the train's longitudinal offset can be determined according to the optical ranging reference deviation. By combining the optical reference distance deviation with the spatial geometry of the electronic map, multiple longitudinal compensation vectors are calculated, representing the precise correction amount of the train in the longitudinal direction.

[0071] By combining lateral offset vectors and longitudinal compensation vectors, the train's position in space is corrected. Lateral offset and longitudinal compensation provide precise positional correction information for the train on the track. Using the lateral and longitudinal corrections, combined with the train's initial positioning coordinates, the train's pose (position and orientation) is dynamically corrected. This process makes the train's positional information more accurate, reflecting its true position on the track. Through the dynamically corrected pose, multiple train position vectors are ultimately obtained, providing real-time, precise positioning of the train on the track.

[0072] In one implementation, the plurality of line environment characteristic parameters are used as correction factors, and operating condition migration probability is predicted based on the plurality of spatiotemporal joint feature vectors to output multiple ultra-short-term net power demands, including:

[0073] Y510: Construct a discrete operating condition state space based on multi-dimensional train operating condition states; Y520: Under operating condition transition constraints, combine and enumerate the multi-dimensional train operating condition states to obtain Y sample operating condition transition vectors; Y530: Locally retrieve Y sets of sample joint feature vectors of the Y sample operating condition transition vectors; Y540: Calculate Y baseline transition probabilities based on the recurrence frequency of the Y sets of sample joint feature vectors; Y550: Use the Y sets of sample joint feature vectors as training data to construct a Y operating condition transition probability prediction model based on an LSTM model; Y560: Use the Y baseline transition probabilities as initial probability distributions and the Y operating condition transition probability prediction model as a dynamic corrector to fill the probability matrix of the discrete operating condition state space, completing the construction of the dynamic probability evolution model.

[0074] The train's operating states can be categorized into the following main operating conditions: traction steady state, where the train operates under constant traction force and maintains a stable speed; traction acceleration, where the train is accelerating and the traction force exceeds the resistance; braking steady state, where the train is in a steady braking state and its speed begins to decrease; braking enhancement, where the train is in a state of enhanced braking with strong braking force and rapid deceleration; and coasting, where the train no longer accelerates or decelerates and continues to move due to inertia. A discrete operating condition state space is established based on these five operating conditions. These states describe the train's state at different stages of operation. This discrete operating condition state space indicates the train's operating state at each moment and provides a foundation for subsequent operating condition transition modeling.

[0075] The transitions between each operating state are not arbitrary but subject to certain physical and operational constraints. For example, the transition from traction steady state to braking steady state is limited by factors such as driver operation, train speed, and track conditions. By combining and enumerating multi-dimensional operating states, all possible operating state transition paths are generated. These paths demonstrate the transition methods and sequences of the train between different operating states, forming multiple operating state transition vectors. These vectors describe the transition patterns of the train from one operating state to another during actual operation. By analyzing these vectors, we can understand the train's operating patterns and potential changes in operating conditions.

[0076] Each operating condition transition vector contains information about the train's transition from one operating condition to another. This information can be used to predict the train's behavior and performance under different operating conditions. Each operating condition transition vector can extract relevant features based on the train's historical operating data, environmental conditions, and operating modes, including train speed, traction force, braking force, and track conditions. By retrieving known sample operating condition transition vectors and their corresponding feature vectors, relevant sample joint feature vectors are obtained. These sample joint feature vectors contain key data about the train's operating state and transition process, serving as an important basis for prediction and modeling.

[0077] For each set of samples, the frequency of their occurrence in historical data is calculated, i.e., the recurrence frequency. For example, if a certain operating condition transition (e.g., from traction steady state to braking steady state) occurs more often in historical data, then its recurrence frequency is higher. Based on the recurrence frequency of each set of samples, the baseline transition probability of each operating condition transition is calculated. The baseline transition probability reflects the probability of the train changing from one operating condition state to another. High frequency operating condition transitions correspond to higher transition probabilities.

[0078] LSTM (Long Short-Term Memory) is a deep learning model capable of processing time-series data, particularly suitable for data with long-term dependencies. In this application scenario, LSTM is used to capture the time-series features of train transitions between different operating conditions. Using the joint feature vector of Y sets of samples as training data, the LSTM model learns how these features influence the probability of train transitions between operating conditions. Through training with historical sample data, the model can identify the temporal patterns of transitions between each operating condition. After training, the LSTM model can predict the probability distribution of a train transitioning to other operating conditions under its current condition.

[0079] Based on historical sample data, baseline transition probabilities are calculated for each operating condition. These baseline probabilities reflect the historical frequency of operating condition transitions. The baseline probabilities represent the initial probability distribution, describing the train's transition trend between different operating conditions. The prediction results of the LSTM model are used to dynamically correct the initial probability distribution; that is, at each time step, the LSTM model updates the transition probabilities based on real-time operating conditions and historical data, dynamically correcting the transition probabilities between each operating condition. In operating condition transition prediction, the transition probability between each operating condition and other operating conditions can be represented by a matrix, where each element corresponds to the train's transition probability from one operating condition to another. By combining the predicted probabilities of the LSTM with the baseline probabilities, the probability matrix is ​​gradually filled. Through continuous correction and updating of the probability matrix, a dynamic probability evolution model is formed. This model continuously evolves as the train operates, adaptively reflecting the real-time changes in operating condition transitions.

[0080] In one implementation, the plurality of line environmental characteristic parameters are used as correction factors, and operating condition migration probability is predicted based on the plurality of spatiotemporal joint feature vectors to output multiple ultra-short-term net power demands, including:

[0081] Y500-1: Interact to obtain the first real-time operating condition status of the first operating train; Y500-2: Based on the first real-time operating condition status, after setting state transition constraints in the dynamic probability evolution model, load the first spatiotemporal joint feature vector into the dynamic probability evolution model, perform operating condition transition probability prediction, and obtain multiple time-series predicted feature vectors under multiple predicted transition operating condition probabilities; Y500-3: Smoothly concatenate the first spatiotemporal joint feature vector and multiple time-series predicted feature vectors to obtain multiple multimodal spatiotemporal feature vectors; Y500-4: Perform power compensation calculation based on the multiple line environment feature parameters and multiple multimodal spatiotemporal feature vectors to obtain multiple initial net power demands; Y500-5: After serializing the multiple initial net power demands based on the operating condition risk attributes, perform risk probability weighted sorting correction based on the multiple predicted transition operating condition probabilities, and select the first ultra-short-term net power demand.

[0082] The train's operating status is monitored in real time by sensors on the train, including the train's current speed, traction, acceleration, and direction. Based on the real-time data collected, the train's current operating condition is determined, such as traction steady state or braking steady state, thus determining the first real-time operating condition.

[0083] In the dynamic probabilistic evolution model, by analyzing the first real-time operating condition, constraints for operating condition transitions are set. These constraints can be based on physical limitations, such as traction force limits or acceleration, or on the stability requirements of the train under different operating conditions, such as the inability to directly jump from the acceleration phase to the braking phase. A first spatiotemporal joint feature vector extracted from sensor data is input into the dynamic probabilistic evolution model. This feature vector represents the train's current state and its environmental information. After setting the state transition constraints, the dynamic probabilistic evolution model predicts the different operating conditions the train will transition to in the future based on the current operating condition and historical operating condition transition data. The model outputs multiple predicted operating condition probabilities, representing the various possible operating conditions and their probabilities of occurrence within a future time period. Based on the predicted operating condition transition probabilities, multiple time-series predicted feature vectors are generated. These vectors describe the dynamic characteristics of the train over a future period under multiple operating condition transitions.

[0084] The first spatiotemporal joint feature vector of the current train state is concatenated with multiple time-series predicted feature vectors obtained through a dynamic probabilistic evolution model. In this way, each spatiotemporal feature vector not only contains the current state information of the train, but also incorporates the predicted data of the future operating conditions. During the concatenation process, smoothing techniques are used to avoid discontinuous fluctuations in the feature vector concatenation. For example, weighted averaging, interpolation, or other smoothing algorithms are used to ensure that the concatenated feature vectors are smooth in time and space, thereby improving the stability and accuracy of the subsequent model.

[0085] By combining multimodal spatiotemporal feature vectors and line environmental parameters, the initial net power demand of the train under the current and predicted operating conditions is calculated. This calculation not only considers traction power but also the recovery of regenerative braking energy. If the train is in a braking state, the regenerative energy will be fed back to the power grid, while the traction power needs to be adjusted according to the predicted operating conditions. Based on different operating conditions, the initial net power demand under multiple different operating conditions is calculated. These demands reflect the net power required by the train under each operating condition, that is, traction power minus regenerative power.

[0086] Each operating condition has certain risk attributes, such as the instability of power demand and the ease of energy recovery. These attributes can be quantified using a serialization method. Through this serialization process, the risks of different operating conditions can be quantified, forming a sequence of operating condition risk attributes. Based on the predicted probability of transitioning to a new operating condition, the risk attributes of each operating condition are weighted and ranked according to their net power demand. For example, some operating conditions lead to higher power demand fluctuations, so their net power demand is given a higher weight. After weighted ranking, the first ultra-short-term net power demand is selected, which is the net power demand that best matches the current train status and environmental conditions. This prediction result serves as a reference in actual operation, helping to achieve optimized scheduling of power load.

[0087] In one implementation, power supply section topology attribution prediction is performed based on the multiple train operation status data and multiple train position vectors, and multiple power supply section topology codes are located, including:

[0088] Y610: Segment the electronic map based on the power supply interval boundary to obtain distributed power supply grid units, wherein each power supply grid unit is bound to a unique power supply interval code; Y620: Analyze multiple IMU inertial measurement outputs from the multiple train operation status data to obtain multiple initial acceleration vectors; Y630: Extract multiple IMU inertial prediction outputs from the multiple time-series prediction feature vectors to obtain multiple incremental acceleration vectors; Y640: Extract multiple velocity vectors and multiple current direction identifiers from the multiple train operation status data; Y650: Calculate the basic displacement distance based on the multiple velocity vectors, multiple initial acceleration vectors, and multiple incremental acceleration vectors, and then perform condition constraint correction in conjunction with the multiple current direction identifiers to output multiple ultra-short-term displacement distances; Y660: Predict multiple updated position coordinates based on the multiple ultra-short-term displacement distances and multiple train position vectors; Y670: Map the multiple updated position coordinates to the distributed power supply grid units to locate the multiple power supply interval topology codes.

[0089] The electronic map is segmented based on the actual physical boundaries of the power supply area, such as the distribution of lines and geographical features. It is divided into multiple small grid units, each representing a part of the power supply network. Each grid unit is assigned a unique power supply area code based on its location on the map and the needs of the power grid. This code ensures that each area has a unique identifier, which allows for precise control of power dispatch and power distribution.

[0090] IMU sensors measure the linear and angular acceleration information of the train using a triaxial accelerometer and gyroscope. By analyzing the output data of these sensors, the train's acceleration can be obtained. Based on the data from the IMU sensors, multiple initial acceleration vectors are calculated. Each vector represents the train's acceleration at a certain moment. The acceleration vectors include acceleration data in the X, Y, and Z directions in three-dimensional space.

[0091] The time-series prediction feature vector is derived from past and current train operation status information, combined with environmental factors, and includes predictions of train acceleration and other operating states at future points in time. Predictive outputs about train inertial data are extracted from multiple time-series prediction feature vectors, including predicted acceleration (IMU prediction output) and possible state changes (such as future speed and position changes). Based on the prediction results, incremental acceleration vectors are obtained, which represent the amount of acceleration change over a future period of time. These vectors represent the changes in train acceleration over a future period of time.

[0092] Multiple velocity vectors are extracted from the train's operational status data. These vectors represent the train's velocity at various points in time, including velocity components in the X, Y, and Z directions, and reflect the train's direction of motion. The current direction indicator shows whether the train is currently in traction or braking mode. If the train is braking, the current direction is negative, indicating that electrical energy is flowing back to the grid; if the train is in traction, the current direction is positive, indicating that electricity is being supplied to the train from the grid.

[0093] The train's current speed information represents its velocity. By combining this with time intervals, the train's displacement per unit time can be estimated. The train's acceleration describes the rate of change of velocity. Combined with the initial state, the contribution of acceleration to displacement can be calculated. Based on the changes in train acceleration, the impact of incremental acceleration on train displacement is calculated; incremental acceleration is an immediate reflection of velocity changes. The direction of the current indicates whether the train is in traction or braking mode, which affects the train's speed and acceleration changes. During braking, the train experiences additional negative acceleration, thus reducing displacement. In traction mode, the acceleration is positive, and the train accelerates and increases displacement. Combining this current direction information, the basic displacement calculation is adjusted. All calculation results are integrated to output multiple ultra-short-term displacement distances. These values ​​represent the predicted displacement of the train in the ultra-short term, which can be displacement estimates for the next tens of seconds to minutes.

[0094] By using multiple ultra-short-term displacement distances and multiple train position vectors, the updated position of the train in the future time is calculated. Specifically, the displacement distance of each train is added to the current coordinates to calculate the updated position at a certain time in the future. Based on the calculation results of the ultra-short-term displacements, combined with the current position of the train, multiple predicted updated position coordinates are obtained. These updated positions reflect the possible position of the train in the future within a short period of time (such as tens of seconds to minutes).

[0095] Multiple updated location coordinates are mapped to pre-divided distributed power supply grid cells. Each grid cell corresponds to a power supply area. The grid division can reflect the power demand of different areas. By mapping the predicted position of the train to the grid cells, the power supply area to which the train belongs is determined. Then, the corresponding power supply section topology code is assigned to the train. This code identifies the power supply area where the train is located and can interact with the power grid dispatching system.

[0096] In one implementation, based on the spatiotemporal coupling relationship between the topology codes of the multiple power supply sections and the multiple ultra-short-term net power demands, dynamic closed-loop coordinated control of regenerative energy in the remote traction power supply system is performed, including:

[0097] Y710: Aggregate the multiple ultra-short-term net power demands according to the topology codes of the multiple power supply sections to obtain P sets of ultra-short-term net power demands corresponding to P train power supply sections in the distributed power supply grid unit; Y720: Sum the P sets of ultra-short-term net power demands to obtain P ultra-short-term predicted total loads; Y730: Adjust the working mode and load ratio of the traction substations in the P train power supply sections according to the P ultra-short-term predicted total loads.

[0098] Multiple ultra-short-term net power demands are aggregated according to the topology code of their respective power supply sections. The train demands within each power supply section (identified by the topology code) are merged into a total demand, forming the ultra-short-term net power demand for each power supply section. Since the power supply area is divided into multiple grid cells, the aggregated power demand corresponds to P train power supply sections in these grids. The train demands within each power supply section are weighted and summed according to the grid position to obtain the ultra-short-term net power demand for each section.

[0099] By summing up the ultra-short-term net power demand of each power supply section, P ultra-short-term predicted total loads are obtained. These loads represent the total power demand of each power supply section. By summing up the total loads, the power grid system can predict the power demand of each power supply section in the ultra-short term, thereby planning and allocating power resources in advance.

[0100] Based on the predicted total load of the entire network, the operating modes and load allocation of substations are dynamically adjusted to ensure reasonable power distribution and load balance within the system. Specifically, when the predicted total load is positive, it indicates an increase in grid load demand, meaning more traction energy is needed to drive trains. In this case, the nearest traction substation is selected based on the train's distance to serve as the primary rectifier load provider. When the predicted total load is negative, it indicates an overload in the grid, requiring the absorption of regenerative energy to regulate the load. In this case, the traction substation furthest from the train's location is selected as the primary inverter load provider to absorb excess power into the grid.

[0101] In one implementation, adjusting the operating mode and allocating the load ratio of traction substations in the P train power supply sections based on the P ultra-short-term predicted total loads includes:

[0102] Y731: Retrieve the updated position coordinates of the first group of operating trains in reverse order based on the first group of ultra-short-term net power demand; Y732: Calculate the first set of double-end spatial distance values ​​between the double-end traction substations in the power supply section of the first train and the first set of updated position coordinates; Y733: If the first ultra-short-term predicted total load is negative, switch the double-end traction substations to inverter mode and then perform pre-intervention of inverter load proportional allocation for the first group of operating trains based on the first set of double-end spatial distance values; Y734: If the first ultra-short-term predicted total load is positive, switch the double-end traction substations to rectification mode and then perform pre-intervention of rectification load proportional allocation for the first group of operating trains based on the first set of double-end spatial distance values.

[0103] Based on the first set of ultra-short-term net power demand as input, the corresponding train composition is first determined, that is, which trains participated in the power demand calculation of the current power supply section. Then, based on the historical operation data, current speed, acceleration, direction of travel and track information of these trains, the latest position coordinates of each train are extracted. Combining the real-time status and historical position information of the trains, the updated position coordinates of each train at the next moment are dynamically calculated, thereby obtaining the first set of updated position coordinates.

[0104] Based on the first set of updated position coordinates, the spatial distance between the train and the substation is calculated. Here, "two ends" refers to the traction substations at both ends, because each power supply section is powered by two substations. By calculating the distance between the train and the two traction substations, a basis is provided for subsequent load allocation decisions, especially when switching load modes, so that a suitable substation can be selected for energy allocation.

[0105] If the first ultra-short-term forecast total load is negative, it indicates that the grid load is excessive, meaning that the excess power needs to be absorbed into the grid. At this time, the substation is switched to inverter mode. In inverter mode, the braking energy of the train is absorbed into the grid, and the double-ended traction substation feeds the regenerative energy of the train back to the grid through the inverter. The distribution of inverter load is determined according to the distance between the train and the double-ended traction substation. If the train is closer to a certain substation, that substation will bear more inverter load. This process is achieved through load proportional distribution, that is, dynamic distribution according to distance and energy absorption capacity.

[0106] If the first ultra-short-term forecast of total load is positive, it means that the power grid needs to supply more power to the train. At this time, the double-end traction substation switches to rectification mode to convert the power from the external power grid into the DC power required by the train. In rectification mode, the substation is responsible for converting the AC power from the external power grid into DC power and transmitting it to the train through the power supply line. By calculating the distance between the train and the double-end substation, the power supply load is allocated. The substation that is closer will provide more power to the train, ensuring the optimization of power supply efficiency and power transmission.

[0107] One implementation also includes:

[0108] Y735: If the first ultra-short-term predicted total load is positive and exceeds the maximum rectification capacity of the double-ended traction substation, activate the rectification function of the traction substation in the adjacent power supply section to provide power support; Y736: If the first ultra-short-term predicted total load is negative and exceeds the absorption capacity of the double-ended traction substation, activate the inverter function of the traction substation in the adjacent power supply section to absorb braking energy in coordination.

[0109] When the first ultra-short-term forecast total load is positive and exceeds the rectification capacity of the current traction substation (rectification capacity refers to the maximum current a traction substation can absorb from the grid), if the load demand exceeds this capacity, it may lead to insufficient power supply. To prevent the current substation from operating under overload, the rectification function of traction substations in adjacent power supply sections is automatically activated, allowing them to participate in power supply. The rectifiers of adjacent substations will help balance the load and ensure sufficient power supply to the trains. Through cross-regional power supply support, the power grid can achieve load balancing, prevent any single substation from operating under overload, and improve the stability and reliability of the system.

[0110] When the first ultra-short-term forecast total load is negative, it means the train is braking and feeding braking energy back into the grid. However, the current absorption capacity of the double-ended traction substations is limited and cannot absorb this extra energy. To prevent excess energy from being wasted, the inverter functions of traction substations in adjacent power supply sections are automatically activated. By activating the inverters in these adjacent substations, the excess braking energy is absorbed into the grid. Through this coordinated absorption, energy can be allocated between multiple power supply sections, avoiding energy waste and protecting the stable operation of the power grid.

[0111] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0112] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A dynamic collaborative control method for regenerative energy in a traction power supply system based on integrated power prediction, characterized in that, The method includes: Dynamic correction of train posture is performed based on incremental sensing data at the trackside end, resulting in multiple train position vectors for multiple operating trains. Based on the multiple train position vectors, dual-thread asynchronous matching is performed to obtain multiple line environment feature parameters and multiple multi-source sensor dynamic fusion weights; Multiple train operation status data are collected by pre-deploying multiple on-board heterogeneous sensing units on the multiple operating trains, wherein the train operation status data includes velocity vector, current direction identifier and IMU inertial measurement output; Based on the dynamic fusion weights of the multiple multi-source sensors, the multiple train operation status data are fused and spliced ​​to construct multiple spatiotemporal joint feature vectors. Using the multiple line environment characteristic parameters as correction factors, the operating condition migration probability is predicted based on the multiple spatiotemporal joint feature vectors, and multiple ultra-short-term net power demands are output. Based on the multiple train operation status data and multiple train position vectors, the topology of the power supply section is predicted, and the topology codes of multiple power supply sections are located. Based on the spatiotemporal coupling relationship between the topology codes of the multiple power supply sections and the multiple ultra-short-term net power demands, dynamic closed-loop collaborative control of regenerative energy in the remote traction power supply system is executed.

2. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 1, characterized in that, Based on incremental sensing data from the trackside end, dynamic correction of train posture is performed to obtain multiple train position vectors for multiple operating trains, including: RFID readers are deployed along the trackside pillars of the bidirectional track at a predetermined deployment density to construct an RFID positioning subarray. Based on the preset key mechanical node rules, the bidirectional track positioning and installation node array is traversed, fiber optic strain sensors are deployed, and a catenary tension monitoring subarray is constructed. Based on the orbital spatial properties, heterogeneous optical sensors are uniformly deployed along a unidirectional track to construct an optical ranging subarray. A heterogeneous sensing spatiotemporal synchronization protocol is adopted to drive the RFID positioning subarray, the contact wire tension monitoring subarray, and the optical ranging subarray in parallel to collect RFID location beacon signals, contact wire tension changes, and optical ranging data. Multiple initial positioning coordinates of multiple operating trains are analyzed based on the RFID location beacon signals; By integrating the catenary tension change data and optical ranging data, dynamic pose-assisted correction is performed on the multiple initial positioning coordinates to obtain the multiple train position vectors.

3. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 2, characterized in that, Based on the multiple train position vectors, a dual-thread asynchronous matching process is performed to obtain multiple line environment feature parameters and multiple multi-source sensor dynamic fusion weights, including: A pre-built dual-threaded processing architecture is provided, which includes a geographic information retrieval engine and a scene adaptation engine. The geographic information retrieval engine projects the multiple train location vectors onto an electronic map and retrieves the multiple line environmental feature parameters, including real-time gradient, curve radius, and line impedance parameters. After matching multiple operating scenario types based on the multiple train position vectors, the scenario adaptation engine extracts the dynamic fusion weights of the multiple multi-source sensors by interacting with the preset scenario weight configuration table. The geographic information retrieval engine and the scene adaptation engine are executed in parallel and asynchronously.

4. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 3, characterized in that, By integrating the catenary tension change data and optical ranging data, dynamic pose-assisted correction is performed on the multiple initial positioning coordinates to obtain the multiple train position vectors, including: Based on the spatial topological mapping relationship of the RFID positioning subarray, the catenary tension monitoring subarray and the optical ranging subarray, multiple tension gradient distributions and multiple optical reference distance deviations are segmented from the catenary tension change and optical ranging data using the multiple initial positioning coordinates. Multiple lateral offset vectors are generated by inversely calculating the topological relationship of the suspension points of the contact wire tension monitoring subarray based on the distribution of the multiple tension gradients. Based on the spatial geometric relationship of the multiple optical reference distance deviations in the electronic map, multiple longitudinal compensation vectors are generated; The multiple lateral offset vectors and multiple longitudinal compensation vectors are used to perform dynamic pose coupling correction of the multiple initial positioning coordinates to obtain the multiple train position vectors.

5. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 3, characterized in that, Using the multiple line environment characteristic parameters as correction factors, and based on the multiple spatiotemporal joint feature vectors, operating condition migration probability prediction is performed to output multiple ultra-short-term net power demands, including: Constructing a discrete operating condition state space based on multi-dimensional train operating condition states; Under the constraint of working condition transition, the multi-dimensional train working condition states are combined and enumerated to obtain Y sample working condition transition vectors. Locally retrieve the joint feature vector of Y groups of samples from the Y types of sample working condition transition vectors; Calculate Y baseline transition probabilities based on the recurrence frequency of the joint feature vector of the Y groups of samples; Using the joint feature vector of the Y groups of samples as training data, a Y-condition transition probability prediction model is constructed based on the LSTM model. Using the Y baseline transition probabilities as the initial probability distribution and the Y working condition transition probability prediction models as dynamic correctors, the probability matrix of the discrete working condition state space is filled to complete the construction of the dynamic probability evolution model.

6. The dynamic coordinated control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 5, characterized in that, Using the multiple line environment characteristic parameters as correction factors, and based on the multiple spatiotemporal joint feature vectors, operating condition migration probability prediction is performed to output multiple ultra-short-term net power demands, including: The first real-time operating status of the first running train is obtained interactively. After setting state transition constraints in the dynamic probability evolution model based on the first real-time operating condition state, the first spatiotemporal joint feature vector is loaded into the dynamic probability evolution model to perform operating condition transition probability prediction, thereby obtaining multiple time-series prediction feature vectors under multiple predicted operating condition probabilities. The first spatiotemporal joint feature vector and multiple temporal prediction feature vectors are smoothly concatenated to obtain multiple multimodal spatiotemporal feature vectors; Power compensation calculations are performed based on the multiple line environment characteristic parameters and multiple multimodal spatiotemporal characteristic vectors to obtain multiple initial net power requirements; After serializing the multiple initial net power demands based on the operating condition risk attributes, the risk probability weighted sorting and correction are performed based on the multiple predicted operating condition transition probabilities to select the first ultra-short-term net power demand.

7. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 6, characterized in that, Based on the multiple train operation status data and multiple train position vectors, the topology assignment of power supply sections is predicted, and the topology codes of multiple power supply sections are located, including: The electronic map is segmented based on the power supply interval boundaries to obtain distributed power supply grid units, wherein each power supply grid unit is bound to a unique power supply interval code; Multiple IMU inertial measurement outputs are analyzed from the multiple train operation status data to obtain multiple initial acceleration vectors; Multiple IMU inertial prediction outputs are extracted from the multiple time-series prediction feature vectors to obtain multiple incremental acceleration vectors; Multiple velocity vectors and multiple current direction identifiers are extracted from the multiple train operation status data; After calculating the basic displacement distance based on the multiple velocity vectors, multiple initial acceleration vectors and multiple incremental acceleration vectors, the working condition constraint correction is performed in combination with the multiple current direction indicators, and multiple ultra-short-term displacement distances are output. Multiple updated position coordinates are predicted based on the multiple ultra-short-term displacement distances and multiple train position vectors; Map the multiple updated location coordinates to the distributed power supply grid unit to locate the topology code of the multiple power supply intervals.

8. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 7, characterized in that, Based on the spatiotemporal coupling relationship between the topology codes of the multiple power supply sections and the multiple ultra-short-term net power demands, dynamic closed-loop coordinated control of regenerative energy in the remote traction power supply system is executed, including: Based on the topology coding of the multiple power supply sections, the multiple ultra-short-term net power demands are aggregated to obtain P sets of ultra-short-term net power demands corresponding to P train power supply sections in the distributed power grid unit; By summing the P groups of ultra-short-term net power demands, we obtain P ultra-short-term predicted total loads; Based on the P ultra-short-term predicted total loads, the operating modes of traction substations in the P train power supply sections are adjusted and the load ratios are allocated.

9. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 8, characterized in that, Based on the P ultra-short-term predicted total loads, the operating modes of traction substations in the P train power supply sections are adjusted and load ratios are allocated, including: Based on the first group of ultra-short-term net power demand, the first group of updated position coordinates of the first group of operating trains are retrieved in reverse order. Calculate the first set of double-end spatial distance values ​​between the double-end traction substations in the first train power supply section and the first set of updated position coordinates; If the first ultra-short-term predicted total load is negative, then after switching the double-ended traction substation to inverter mode, the inverter load of the first group of running trains will be pre-intervened based on the first set of double-ended spatial distance values. If the first ultra-short-term predicted total load is positive, then after switching the double-ended traction substation to rectification mode, the rectification load of the first group of running trains will be pre-intervened based on the first group of double-ended spatial distance values.

10. The dynamic collaborative control method for regenerative energy in a traction power supply system based on fusion power prediction as described in claim 9, characterized in that, Also includes: If the first ultra-short-term predicted total load is positive and exceeds the maximum rectification capacity of the dual-end traction substation, the rectification function of the adjacent power supply section traction substation is activated to provide power support. If the first ultra-short-term predicted total load is negative and exceeds the absorption capacity of the dual-end traction substation, the inverter function of the adjacent power supply section traction substation is activated to collaboratively absorb braking energy.

Images