Ultra-short-term dynamic prediction method for traction load of high-speed railway

Abstract

The invention relates to the technical field of traction load prediction, in particular to a high-speed railway traction load ultra-short-term dynamic prediction method. Acquiring an original data set and a driving running diagram of the traction substation; extracting a dynamic traction load sequence based on a window state, and establishing a known sample; selecting characteristic indexes to classify the dynamic load data of the motor train unit; identifying the moment when the motor train unit enters the power supply interval; the vehicle type and the vehicle running process of the vehicle are obtained according to the driving running chart, and the vehicle type and the vehicle running process are input into the dynamic load database to obtain the corresponding dynamic load curve of the motor train unit. The control parameters of the control system can be effectively adjusted according to the prediction data of the traction load, a reasonable and effective scheduling strategy is formulated, and therefore the problems caused by the traction load peak value and regenerative braking energy are solved to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the technical field of traction load prediction, and in particular to an ultra-short-term dynamic prediction method for traction load of a high-speed railway. Background Art

[0002] With the continuous development of electrified railways, the total length of electrified railways is increasing rapidly; however, the rapid growth of railways has also brought some negative effects. For example, traction loads often generate high load peaks, which will impact the power grid and even affect the safe and stable operation of trains; or the regenerative braking energy (RBE) generated during train operation, if not recovered and injected into the power grid in time, will not only aggravate the load imbalance problem of the power grid, but also be charged additional fees by the power grid company.

[0003] In order to improve the daily operation efficiency of the traction power supply system, eliminate the peak of traction load and recover the regenerative braking energy (RBE), many solutions have been proposed, such as combining passive filters, static VAR compensators, railway power conditioners and other devices. These devices can improve the above two problems, but the most effective device is a control device with supercapacitors and railway power conditioners (SC-RPC) as the core. This type of control device can reduce the impact of regenerative braking energy on the power grid by absorbing this energy when the EMU generates regenerative braking energy; then release the energy when the traction load peak occurs, reduce the energy required by the power grid to supply the EMU, and eliminate the load peak. Since the traction load is random, volatile and has a large fluctuation amplitude, this brings challenges to the control of the power quality control device: on the one hand, the strong random volatility of the load requires the control system to have the ability to respond quickly; on the other hand, the large load fluctuation amplitude leads to a high load peak power, which puts higher requirements on the large signal stability of the control system.

[0004] Therefore, in order to improve the effect of the control device, it is necessary to predict the traction load. According to the predicted data of the traction load, the control parameters of the control system can be effectively adjusted, and a reasonable and effective scheduling strategy can be formulated to solve the problems caused by the traction load peak and regenerative braking energy to the maximum extent. In summary, it is necessary to propose a method for ultra-short-term prediction of dynamic load of traction substation. Summary of the invention

[0005] The purpose of the present invention is to provide an ultra-short-term dynamic prediction method for high-speed railway traction load, aiming to realize the prediction of traction load. According to the predicted data of traction load, the control parameters of the control system can be effectively adjusted, and a reasonable and effective scheduling strategy can be formulated, so as to solve the problems caused by traction load peak and regenerative braking energy to the greatest extent.

[0006] To achieve the above object, the present invention adopts a high-speed railway traction load ultra-short-term dynamic prediction method, comprising the following steps:

[0007] Obtain the original data set and train operation diagram of the traction substation;

[0008] Extract dynamic traction load based on window status and establish known samples;

[0009] Select characteristic indicators to classify the dynamic load data of EMUs;

[0010] Identify the moment when the EMU enters the power supply section;

[0011] The vehicle type and vehicle operation process of the vehicle are obtained according to the driving operation diagram, and the dynamic traction load is predicted according to the superposition characteristics of the active power of the traction dynamic load.

[0012] Among them, in the step of obtaining the original data set and the train operation diagram of the traction substation:

[0013] The original data set includes the total dynamic load data of the traction substation for three days, where the active power data is recorded as P h , current data is recorded as I h .

[0014] Among them, in the step of extracting dynamic traction load based on window state and establishing known samples:

[0015] Enter the measured dynamic load data of the traction substation P = {P1, P2, P3....., P n}, and set the data length to m, segment the entire data P by length m, and divide it into k sub-data segments {seg1, seg2, seg3...., segk}; where P i is the data set measured every hour, k = n / m;

[0016] For each data segment seg, calculate the average load value of each data segment seg and set the corresponding threshold P threshold , compare the average load value with the threshold;

[0017] Extract the complete data segments in the loaded state and save each data segment, mark the start and end times, and form a data sample;

[0018] According to the train operation diagram, the complete data in the data sample is saved, and the corresponding vehicle model, formation, running direction and stop conditions are marked to establish a known sample.

[0019] For each data segment seg, the average load value of the data segment is calculated, and the corresponding threshold P is set. threshold, in the step of comparing the average load value with the threshold:

[0020] The average load value is calculated as:

[0021]

[0022] When P mean <P threshold , it is in no-load state, marked as 0;

[0023] When P mean >P threshold , it is in loaded state, marked as 1.

[0024] For each data segment seg, the average load value of the data segment is calculated and the corresponding threshold P is set. threshold , after the step of comparing the average load value to the threshold:

[0025] By merging the data segments in the loaded state, that is, merging all adjacent data segments with the same label 1 into a complete data segment, the dynamic load curve data of the EMU can be completely obtained.

[0026] Among them, in the step of selecting characteristic indicators to classify the dynamic load data of the EMU:

[0027] Obtain known samples, calculate and save the load characteristic indicators of the known samples, re-measure the dynamic load data of a whole day in the same traction substation, and extract the loaded state data according to the dynamic traction load extraction method based on the window state to establish the target sample;

[0028] According to the four influencing factors and the corresponding characteristic indicators, four classification trainers are set up respectively, and corresponding labels are set respectively;

[0029] The known samples and the corresponding characteristic indicators are input into the four classifiers in turn to realize the classification of the trainer, and the characteristic indicators generated by each type of load in the known samples are matched with the load type and saved;

[0030] Input the target samples into the four trained classifiers in turn and perform recognition;

[0031] For incorrect target samples, load identification is performed according to the train operation diagram to check whether there are new EMU load types.

[0032] Among them, in the step of inputting the target samples into the four trained classifiers in sequence and performing recognition:

[0033] If the target sample has the same characteristic index as a certain type of load in the known sample, then the correct target sample is identified as a certain type of load in the known sample;

[0034] If all existing load types in the database are not satisfied, the wrong target sample will be identified.

[0035] Among them, in the step of identifying the time when the EMU enters the power supply section:

[0036] Obtain accurate train operation diagrams, dynamic load database of EMUs, and all-day current load data I measured on the day of the train operation diagram. At the same time, record the time when each EMU enters the power supply section during the data measurement process. The resolution of the acquired data is 20kHz.

[0037] The current load data with a resolution of 20kHz is converted into current load data with a resolution of 400Hz by using an anti-aliasing filter combined with downsampling; the filter is used to remove high-frequency components before downsampling;

[0038] Use a sliding window with a size of 7s and a step length of 7s to divide the current load data I, and analyze the data in the window to see whether the data in the window contains the moment when the vehicle enters the power supply interval. Save the analyzed data segments and labels accordingly to establish data samples;

[0039] A neural network model is established with the BiLSTM model as the core. The training set is input into the neural network model for training, and the validation set is used to test the effect of the trained model. The test set is input into the trained model to detect the accuracy of the model.

[0040] Among them, in the step of dividing the current load data I by using a sliding window with a size of 7s and a step length of 7s, analyzing the start time and the end time of the window, and forming a data sample by corresponding the analyzed data segments and labels:

[0041] If it contains the moment when the EMU enters the power supply section, the data segment is stored and labeled as 1;

[0042] If the data segment does not contain the moment when any EMU enters the power supply section, the data segment will be stored and labeled as 0.

[0043] Among them, in the step of obtaining the vehicle type and vehicle operation process of the vehicle according to the driving operation diagram, inputting the vehicle type and vehicle operation process into the dynamic load database to obtain the corresponding dynamic load curve of the EMU:

[0044] If the power supply section is in a no-load operation state, according to the superposition characteristics of the active power of the traction dynamic load, when the EMU is identified to be driving in, the dynamic load curve of the EMU can be superimposed on the no-load curve, so as to predict the dynamic load curve of the traction substation when only a single EMU is running in the power supply section;

[0045] If the power supply section is in a loaded operation state, when a new EMU is identified to enter, the dynamic load curve is superimposed on the originally predicted load curve, and the dynamic load curve of the traction substation when multiple vehicles are running can be predicted.

[0046] The present invention provides an ultra-short-term dynamic prediction method for high-speed railway traction load, which obtains the original data set and train operation diagram of the traction substation; extracts the dynamic traction load sequence based on the window state and establishes a known sample; selects characteristic indicators to classify the dynamic load data of the EMU; identifies the moment when the EMU enters the power supply section; obtains the vehicle type and vehicle operation process of the vehicle according to the train operation diagram, inputs the vehicle type and vehicle operation process into the dynamic load database to obtain the corresponding EMU dynamic load curve, and superimposes the load curve at the moment when the EMU enters the power supply section according to the superposition characteristics of the active power of the traction dynamic load. In the above manner, the prediction of the traction load is realized, and the control parameters of the control system can be effectively adjusted according to the predicted data of the traction load, and a reasonable and effective scheduling strategy can be formulated, thereby solving the problems caused by the traction load peak and regenerative braking energy to the greatest extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 It is a flow chart of the steps of the ultra-short-term dynamic prediction method of high-speed railway traction load of the present invention.

[0049] Figure 2 It is a step flow chart of S200 of the present invention.

[0050] Figure 3 It is a schematic diagram of the steps of S200 of the present invention.

[0051] Figure 4 It is a step flow chart of S300 of the present invention.

[0052] Figure 5 It is a step schematic diagram of S300 of the present invention.

[0053] Figure 6 It is a step flow chart of S400 of the present invention. DETAILED DESCRIPTION

[0054] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0055] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0056] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0057] See also Figure 1 to Figure 6 The present invention provides a method for ultra-short-term dynamic prediction of high-speed railway traction load, comprising the following steps:

[0058] S100: Acquire the original data set and train operation diagram of the traction substation.

[0059] In this embodiment, the original data set and the driving diagram of the traction substation are first obtained. The original data set includes the total dynamic load data of the traction substation for three days, wherein the active power data is recorded as P h , current data is recorded as I h .

[0060] S200: Extract dynamic traction load based on window status and establish known samples.

[0061] In this embodiment, the dynamic traction load is extracted based on the window state and a known sample is established. The specific process is as follows:

[0062] S201: Input the measured dynamic load data of the traction substation and divide it into sections.

[0063] Among them, the dynamic load data of the traction substation measured in 2 to 3 days is entered as P = {P1, P2, P3...., P n}, and set the data length to m, segment the entire data P by length m, and divide it into k sub-data segments {seg1, seg2, seg3...., segk}; where P i is the data set measured every hour, k = n / m;

[0064] S202: For each data segment, calculate the average load value of the data segment, set a corresponding threshold, and compare the average load value with the threshold.

[0065] For each data segment seg, the average load value of the data segment is calculated and the corresponding threshold P is set. threshold , compare the average load value with the threshold;

[0066] The average load value is calculated as:

[0067]

[0068] When P mean <P threshold , it is no-load state, marked as 0; when P mean >P threshold , it is in loaded state, marked as 1;

[0069] By merging the loaded data segments, that is, merging all adjacent data segments with the same label 1 into a complete data segment, the dynamic load curve data of the EMU can be completely obtained.

[0070] Among them, due to the noise caused by grid voltage fluctuation or measurement error in the measured data, the average load value in some no-load states is greater than the threshold. However, since this state exists for a relatively short time, the wrong judgment can be eliminated by judging the length of the loaded state.

[0071] S203: extract the complete data segments in the loaded state and save each data segment, marking the start and end times to form a data sample.

[0072] S204: According to the driving diagram, the complete data in the data sample is saved, and the corresponding vehicle type, formation, running direction and stop status are marked to establish a known sample.

[0073] S300: Select characteristic indicators to classify the dynamic load data of the EMU.

[0074] In this implementation, for a certain fixed interval, the factors that affect the dynamic load characteristics of the EMU are mainly: EMU model, formation, running direction and stop situation. These four influencing factors constitute the category to which the sample belongs. The dynamic load of the EMU with the same influencing factors has basically the same load waveform change trend and load characteristics. In order to intuitively analyze different influencing factors from the data, characteristic indicators that reflect the influencing factors are selected. When identifying the formation and model, the maximum power and minimum power are selected as characteristic indicators; when identifying the running direction, the traction power, regeneration power and skewness factor characteristic indicators are used; when judging the stop situation, the minimum load, regeneration power and kurtosis factor are used as characteristic indicators. Therefore, from the above four aspects, the dynamic load data of the EMU is classified according to the above eight characteristic indicators. The specific process is as follows:

[0075] S301: Obtain a known sample, re-measure the dynamic load data of a whole day in the same traction substation, and extract the loaded state data according to the dynamic traction load extraction method based on the window state, establish a target sample, and then calculate and save the load characteristic index of the known sample.

[0076] The calculation formula of relevant indicators is as follows:

[0077] Traction power:

[0078] Regeneration power:

[0079] Among them, P i is the active power measured per second, StP i >0;

[0080] Skewness Factor:

[0081] Kurtosis factor:

[0082] Where X ma is the average amplitude of power; is the average value of power.

[0083] S302: According to the four influencing factors and the corresponding characteristic indicators, four classification trainers are set respectively, and corresponding labels are set respectively.

[0084] S303: Input the known samples and the corresponding characteristic indicators into the four classifiers in sequence to implement the trainer classification, and correspond and save the characteristic indicators and load types generated by each type of load in the known samples.

[0085] S304: Input the target sample into the four trained classifiers in sequence and perform recognition.

[0086] Among them, if the characteristic indicators of the target sample and the type of load in the known sample are the same, the correct target sample is identified as a certain type of load in the known sample; if it does not meet all the load types existing in the database, the wrong target sample is identified.

[0087] S305: Due to the determinism of the train operation diagram, all load types of the EMUs passing through a traction substation every day can be determined by the train operation diagram to check whether there are any newly introduced load types. Therefore, for incorrect target samples, load identification is performed according to the train operation diagram to check whether there are any newly added EMU loads.

[0088] If there is no new load type, the wrong target sample is considered bad data and is deleted;

[0089] If there is a newly introduced load, its load type is determined according to the driving diagram and new characteristic indicators are found, and a new classifier is established and trained.

[0090] In summary, for all the extracted samples, after being processed by the classification trainer, the correct data in the samples can be saved and the corresponding vehicle types, formations, running directions and stop conditions can be marked to establish a dynamic load database with complete information.

[0091] S400: Identify the time when the EMU enters the power supply section.

[0092] In this embodiment, the specific process of identifying the time when the EMU enters the power supply section is as follows:

[0093] S401: Obtain an accurate train operation diagram, a dynamic load database of the EMU, and the all-day current load data I measured on the day of the train operation diagram. At the same time, during the data measurement process, the moment when each EMU enters the power supply section is recorded. The resolution of the acquired data is 20kHz.

[0094] S402: Taking into account the recognition time of the model and the working efficiency of the detection equipment, the current load data with a resolution of 20kHz is converted into current load data with a resolution of 400Hz by using an anti-aliasing filter combined with a downsampling method; the filter is used to remove high-frequency components before downsampling to prevent aliasing effects.

[0095] S403: Use a sliding window with a size of 7s and a step length of 7s to divide the current load data I, analyze the start time and end time of the window, and form data samples by corresponding the analyzed data segments and labels.

[0096] If it contains the moment when the EMU enters the power supply section, the data segment is stored and labeled as 1; if it does not contain the moment when any EMU enters the power supply section, the data segment is stored and labeled as 0.

[0097] S404: Establish a neural network model with the BiLSTM model as the core. Divide the data samples into training set, validation set and test set in the ratio of 7:1.5:1.5. Input the training set into the neural network model for training, and use the validation set to test the effect of the trained model. Finally, input the test set into the trained model to detect the accuracy of the model.

[0098] 500: Obtain the vehicle type and vehicle operation process of the vehicle according to the driving operation diagram, input the vehicle type and vehicle operation process into the dynamic load database to obtain the corresponding dynamic load curve of the EMU.

[0099] In this embodiment, whenever a EMU is identified to enter the power supply section, the vehicle type and vehicle operation process of the vehicle are obtained according to the driving diagram, and the vehicle type and vehicle operation process are input into the dynamic load database to obtain the corresponding EMU dynamic load curve; if the power supply section is in a no-load operation state, according to the superposition characteristics of the active power of the traction dynamic load, the EMU dynamic load curve can be superimposed on the no-load curve when the EMU is identified to enter, and the dynamic load curve of the traction substation when only a single EMU is running in the power supply section can be predicted; if the power supply section is in a loaded operation state, then when a new EMU is identified to enter, the dynamic load curve is superimposed on the originally predicted load curve, and the dynamic load curve of the traction substation when multiple vehicles are running can be predicted.

[0100] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the contents disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.

[0101] It should be understood that the present application is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof.

Claims

1. A method for ultra-short-term dynamic prediction of high-speed railway traction load, characterized in that: The steps include: Obtain the original data set and train operation diagram of the traction substation; Extract dynamic traction load based on window status and establish known samples; Select characteristic indicators to classify the dynamic load data of EMUs; Identify the moment when the EMU enters the power supply section; The vehicle type and vehicle operation process of the vehicle are obtained according to the driving operation diagram, and the dynamic traction load is predicted according to the superposition characteristics of the active power of the traction dynamic load.

2. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 1, characterized in that: In the steps of obtaining the original data set and train operation diagram of the traction substation: The original data set includes the total dynamic load data of the traction substation for three days, where the active power data is recorded as P h , current data is recorded as I h .

3. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 1, characterized in that: In the step of extracting dynamic traction load based on window state and establishing known samples: Enter the measured dynamic load data of the traction substation P = {P1, P2, P3....., P n }, and set the data length to m, segment the entire data P by length m, and divide it into k sub-data segments {seg1, seg2, seg3...., segk}; where P i is the data set measured every hour, k = n / m; For each data segment seg, calculate the average load value of each data segment seg and set the corresponding threshold P threshold , compare the average load value with the threshold; Extract the complete data segments in the loaded state and save each data segment, mark the start and end times, and form a data sample; According to the train operation diagram, the complete data in the data sample is saved, and the corresponding vehicle model, formation, running direction and stop conditions are marked to establish a known sample.

4. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 3, characterized in that: For each data segment seg, calculate the average load value of the data segment and set the corresponding threshold P threshold , in the step of comparing the average load value with the threshold: The average load value is calculated as: When P mean <P threshold , it is in no-load state, marked as 0; When P mean >P threshold , it is in loaded state, marked as 1.

5. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 4, characterized in that: For each data segment seg, calculate the average load value of the data segment and set the corresponding threshold P threshold , after the step of comparing the average load value to the threshold: By merging the loaded data segments, that is, merging all adjacent data segments with the same label 1 into a complete data segment, the dynamic load curve data of the EMU can be completely obtained.

6. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 1, characterized in that: In the step of selecting characteristic indicators to classify the dynamic load data of the EMU: Obtain known samples, calculate and save the load characteristic indicators of the known samples, re-measure the dynamic load data of a whole day in the same traction substation, and extract the loaded state data according to the dynamic traction load extraction method based on the window state to establish the target sample; According to the four influencing factors and the corresponding characteristic indicators, four classification trainers are set up respectively, and corresponding labels are set respectively; The known samples and the corresponding characteristic indicators are input into the four classifiers in turn to realize the classification of the trainer, and the characteristic indicators generated by each type of load in the known samples are matched with the load type and saved; Input the target samples into the four trained classifiers in turn and perform recognition; For incorrect target samples, load identification is performed according to the train operation diagram to check whether there are new EMU load types.

7. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 6, characterized in that: In the step of inputting the target samples into the four trained classifiers in sequence and performing recognition: If the target sample has the same characteristic index as a certain type of load in the known sample, then the correct target sample is identified as a certain type of load in the known sample; If all existing load types in the database are not satisfied, the wrong target sample will be identified.

8. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 1, characterized in that: In the step of identifying the time when the EMU enters the power supply section: Obtain accurate train operation diagrams, dynamic load database of EMUs, and all-day current load data I measured on the day of the train operation diagram. At the same time, record the time when each EMU enters the power supply section during the data measurement process. The resolution of the acquired data is 20kHz. The current load data with a resolution of 20kHz is converted into current load data with a resolution of 400Hz by using an anti-aliasing filter combined with downsampling; the filter is used to remove high-frequency components before downsampling; Use a sliding window with a size of 7s and a step length of 7s to divide the current load data I, and analyze the data in the window to see whether the data in the window contains the moment when the vehicle enters the power supply interval. Save the analyzed data segments and labels accordingly to establish data samples; A neural network model is established with the BiLSTM model as the core. The training set is input into the neural network model for training, and the validation set is used to test the effect of the trained model. The test set is input into the trained model to detect the accuracy of the model.

9. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 8, characterized in that: In the step of dividing the current load data I by using a sliding window with a size of 7s and a step length of 7s, analyzing the start time and the end time of the window, and forming data samples by corresponding the analyzed data segments and labels: If it contains the moment when the EMU enters the power supply section, the data segment is stored and labeled as 1; If the data segment does not contain the moment when any EMU enters the power supply section, the data segment will be stored and labeled as 0.

10. The ultra-short-term dynamic prediction method for high-speed railway traction load according to claim 1, characterized in that: In the step of obtaining the vehicle type and vehicle operation process of the vehicle according to the driving operation diagram, inputting the vehicle type and vehicle operation process into the dynamic load database to obtain the corresponding dynamic load curve of the EMU: If the power supply section is in a no-load operation state, according to the superposition characteristics of the active power of the traction dynamic load, when the EMU is identified to be driving in, the dynamic load curve of the EMU can be superimposed on the no-load curve, so as to predict the dynamic load curve of the traction substation when only a single EMU is running in the power supply section; If the power supply section is in a loaded operation state, when a new EMU is identified to enter, the dynamic load curve is superimposed on the originally predicted load curve, and the dynamic load curve of the traction substation when multiple vehicles are running can be predicted.

Images