Train control model training method and apparatus

Abstract

Provided in the present invention are a train control model training method. The training method comprises: collecting historical operation data of a train which operates at the highest operation efficiency; on the basis of the historical operation data, using a machine learning regression algorithm to establish a train operation simulation model; on the basis of an initial sequence, generating a plurality of correction sequences; respectively inputting the plurality of correction sequences into the train operation simulation model, so as to obtain operation parameters respectively corresponding to the plurality of correction sequences; on the basis of the operation parameters respectively corresponding to the plurality of correction sequences, obtaining a target correction sequence among the plurality of correction sequences; and on the basis of the target correction sequence among the plurality of correction sequences, performing training to obtain a train control model.

Description

Training method and device for train control model

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application filed with the China Patent Office on November 28, 2023, with application number 202311604309.7 and application name “Training method and device for train control model”, the entire contents of which are incorporated by reference into this application. Technical Field

[0003] The present invention relates to the field of rail transportation technology, and in particular to a training method and device for a train control model. Background Art

[0004] At present, most of the train control methods actually used in the rail transit field pursue the highest efficiency and lack attention to energy conservation. During implementation, they are highly dependent on human experience and cannot ensure the application of train control methods that take into account both operational efficiency and energy conservation.

[0005] Summary of the Invention

[0006] The present invention provides a training method and device for a train control model, which is used to solve the defect that the train control method actually used in the prior art cannot take into account both operating efficiency and energy saving. Through an intelligent algorithm, it is achieved that the train energy consumption is reduced while taking into account the operating efficiency.

[0007] The present invention provides a training method for a train control model, comprising:

[0008] Collect historical operating data of trains running at the highest efficiency;

[0009] Based on the historical operation data, a train operation simulation model is established using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence consisting of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles;

[0010] Based on the initial sequence, a plurality of modified sequences are generated, wherein the plurality of modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage; the initial sequence is a stage sequence consisting of stages corresponding to a train operating at the highest operating efficiency in a plurality of consecutive control cycles;

[0011] Inputting the multiple correction sequences into the train operation simulation model respectively to obtain operation parameters corresponding to the multiple correction sequences respectively;

[0012] obtaining a target correction sequence among the multiple correction sequences based on the operating parameters respectively corresponding to the multiple correction sequences;

[0013] Training a train control model based on a target correction sequence among the multiple correction sequences;

[0014] The operating parameters include one or more of the following:

[0015] Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0016] According to a train control model training method provided by the present invention, the train operation simulation model is established based on the historical operation data using a machine learning regression algorithm, including:

[0017] Taking the historical operation data as input, the gradient descent method is used to solve the parameters of the speed level position sub-model, the position speed sub-model and the energy consumption level position speed sub-model to construct the train operation simulation model;

[0018] The speed level sub-model is expressed as: The input of the speed level sub-model includes the simulation instantaneous speed corresponding to the previous control cycle and the level sequence consisting of the levels corresponding to the control cycles from m control cycles ago to the current control cycle The output of the speed level sub-model is the simulation instantaneous speed corresponding to the current control cycle The model parameter of the speed level position sub-model is a -m ,...,a -1 ,a0,b,t tick Indicates the duration of a control cycle;

[0019] The position-velocity sub-model is expressed as: The input of the position-speed sub-model includes the simulation instantaneous position corresponding to the previous control cycle. The speed sequence consisting of the speeds corresponding to the control cycles from m control cycles ago to the current control cycle and the level sequence consisting of the levels corresponding to the control cycles from m control cycles ago to the current control cycle The output of the position and speed sub-model is the simulation instantaneous position corresponding to the current control cycle The model parameter of the position and velocity sub-model is c -m ,...,c -1 ,c0,d -m,...,d -1 ,d0,e,t tick Indicates the duration of a control cycle;

[0020] The energy consumption level velocity sub-model is expressed as:

[0021] The input of the energy consumption level speed sub-model includes the speed corresponding to the current control cycle The level corresponding to the current control cycle The output of the energy consumption level speed sub-model is the energy consumption corresponding to the current control cycle. The model parameters of the energy consumption level speed sub-model are f1, f2, f3, g0, g1, g2, h0, h1, h2.

[0022] According to a train control model training method provided by the present invention, obtaining a target correction sequence among the multiple correction sequences based on the operating parameters corresponding to the multiple correction sequences respectively includes:

[0023] Eliminating, from the plurality of correction sequences, correction sequences that meet a first condition based on the operating parameters respectively corresponding to the plurality of correction sequences;

[0024] Based on the total energy consumption of the remaining correction sequences, the correction sequences with abnormal total energy consumption are eliminated from the remaining correction sequences to obtain the target correction sequence;

[0025] The first condition includes at least one of the following:

[0026] The total running time corresponding to the correction sequence exceeds a preset running time threshold;

[0027] Among the multiple consecutive control cycles corresponding to the correction sequence, there is at least one control cycle whose speed is lower than a preset speed threshold.

[0028] According to a train control model training method provided by the present invention, the method of eliminating correction sequences with abnormal total energy consumption from the remaining correction sequences based on the total energy consumption of the remaining correction sequences to obtain a target correction sequence includes:

[0029] sorting the remaining correction sequences from largest to smallest based on the running times corresponding to the remaining correction sequences to obtain a first-order correction sequence;

[0030] Determining a correction sequence for abnormal energy consumption based on the order of energy consumption corresponding to the correction sequences of the first order;

[0031] The correction sequence with abnormal total energy consumption is eliminated from the remaining correction sequences to obtain a target correction sequence.

[0032] According to a method for training a train control model provided by the present invention, the training of the train control model based on a target correction sequence among the multiple correction sequences includes:

[0033] The train control model is obtained by training using the relevant parameters of the control cycle corresponding to the traction stage in the target correction sequence as training samples and whether the traction stage is corrected to the inertia stage as a label;

[0034] Among them, the relevant parameters of the control cycle include at least one of the following: remaining running time, remaining distance, slope of the next N slope sections, length of the next N slope sections, current target speed limit, current ceiling speed limit, braking deceleration, traction acceleration, and tolerable impact rate, where N is an integer greater than 2.

[0035] The present invention also provides a train control method, comprising:

[0036] Obtain real-time relevant parameters of the train;

[0037] Inputting the real-time relevant parameters of the train into a train control model to obtain a target level output by the train control model;

[0038] Controlling the train operation based on the target level;

[0039] The training method of the train control model includes:

[0040] Collect historical operating data of trains running at the highest efficiency;

[0041] Based on the historical operation data, a train operation simulation model is established using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence consisting of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles;

[0042] Based on the initial sequence, a plurality of modified sequences are generated, wherein the plurality of modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage; the initial sequence is a stage sequence consisting of stages corresponding to a train operating at the highest operating efficiency in a plurality of consecutive control cycles;

[0043] Inputting the multiple correction sequences into the train operation simulation model respectively to obtain operation parameters corresponding to the multiple correction sequences respectively;

[0044] obtaining a target correction sequence among the multiple correction sequences based on the operating parameters respectively corresponding to the multiple correction sequences;

[0045] Training a train control model based on a target correction sequence among the multiple correction sequences;

[0046] The operating parameters include one or more of the following:

[0047] Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0048] The present invention also provides a training device for a train control model, comprising:

[0049] The collection module is used to collect historical operating data of trains running at the highest operating efficiency;

[0050] a model building module, configured to establish a train operation simulation model based on the historical operation data using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence consisting of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles;

[0051] A sequence generation module is configured to generate a plurality of modified sequences based on an initial sequence, wherein the plurality of modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage; the initial sequence is a stage sequence consisting of stages corresponding to a train operating at the highest operating efficiency in a plurality of consecutive control cycles;

[0052] a parameter acquisition module, configured to input the plurality of correction sequences into the train operation simulation model respectively, and obtain operation parameters corresponding to the plurality of correction sequences respectively;

[0053] a target correction sequence acquisition module, configured to obtain a target correction sequence from the plurality of correction sequences based on the operating parameters respectively corresponding to the plurality of correction sequences;

[0054] a training module, configured to train a train control model based on a target correction sequence among the plurality of correction sequences;

[0055] The operating parameters include one or more of the following:

[0056] Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0057] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the training method for the train control model described in any one of the above items, or implements the above train control method.

[0058] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it implements the training method of the train control model described in any one of the above items, or implements the above train control method.

[0059] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any one of the above-mentioned methods for training a train control model or the above-mentioned train control method.

[0060] The train control model training method and device provided by the present invention are easy to transplant and highly robust because they only correct the traction level in the operating data; they can automatically generate a large number of train control model inputs by using historical operating data of trains running at the highest operating efficiency; the obtained target correction sequence can take into account both efficiency and energy saving compared to the initial sequence when running at the highest operating efficiency; they can achieve model optimization iteration as the input is updated; because they do not rely on human experience and judgment, they can identify certain operating modes that take into account both efficiency and energy saving that humans may miss, thereby reducing train energy consumption while taking into account operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0062] FIG1 is a flow chart of a method for training a train control model provided by the present invention;

[0063] FIG2 is a flow chart of a train control method provided by the present invention;

[0064] FIG3 is a schematic structural diagram of a training device for a train control model provided by the present invention;

[0065] FIG4 is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0066] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0067] First, let’s introduce the following contents:

[0068] To meet people's demands for energy conservation and environmental protection, energy conservation has become a key goal in urban rail transit. However, currently used automatic train control methods mostly pursue maximum efficiency, resulting in large errors in operating time and a lack of attention to energy conservation. Furthermore, they are unable to achieve optimization iterations over time and with the accumulation of operating data. Furthermore, they are highly dependent on human experience and are prone to falling into local optimal solutions. They are unable to identify operating methods that balance efficiency and energy conservation from the vast amount of operating data that humans may have missed. Therefore, research on how to optimize current train operation methods and, further, how to achieve operational efficiency while reducing train energy consumption through intelligent algorithms is an urgent issue to be addressed.

[0069] There are three main control modes during train operation: traction, braking, and coasting. Since coasting mode is more energy-efficient than traction mode, coasting is best used to minimize energy consumption while still achieving operational goals.

[0070] To overcome the above-mentioned drawbacks, the present invention provides a method and apparatus for training a train control model, which achieve the following advantages:

[0071] 1. Take into account both running time and energy saving;

[0072] 2. It can identify certain operating modes that balance efficiency and energy conservation that humans may miss;

[0073] 3. Since only the traction level is modified, it is easy to transplant and compatible with different types of control methods;

[0074] 4. Ability to achieve optimization iteration as operating data is updated;

[0075] 5. Using the collected operating data, a large amount of training input can be automatically generated.

[0076] The following describes the train control model training method and device of the present invention with reference to the accompanying drawings.

[0077] FIG1 is a flow chart of a method for training a train control model provided by the present invention. As shown in FIG1 , the method includes the following steps:

[0078] Step 100 , collecting historical operation data of trains running at the highest operating efficiency;

[0079] On one or more trains on a real line, the existing train control method is used to make the train run at the highest operating efficiency for several days, and the train operation log is collected, that is, the historical operation data of the train during the operation. It should be pointed out that the operation at the highest operating efficiency in the present invention refers to the situation where the train is running at the highest operating efficiency under the condition that the train does not exceed the speed limit during the operation. Among them, the maximum delay of the train traction and braking conversion is m control cycles, and the time corresponding to each control cycle is t tick The collected historical operating data includes the train's instantaneous position s0, instantaneous speed v0, level l0, instantaneous power p0, and other data information within each of multiple consecutive control cycles. Level can also be called control level, output level, etc.

[0080] The train corresponds to a level within a control cycle. Therefore, the level l0 of the train within this control cycle can also be called the current level l0. When collecting the position, speed, and power of the train within a control cycle, it is only necessary to select a time point within this control cycle for collection. Optionally, the train's instantaneous position s0, instantaneous speed v0, level l0, instantaneous power p0 and other data information can be collected at the start time of this cycle, or at other time points. The present invention is not limited to this. The instantaneous position s0, instantaneous speed v0, level l0, and instantaneous power p0 can also be referred to as the current position s0, current speed v0, current level l0, and current power p0.

[0081] Step 110: Based on the historical operation data, a train operation simulation model is established using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence composed of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles; wherein the operating parameters include one or more of the following: total operating time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0082] In some embodiments, based on the historical operation data collected in the above step 100, that is, the instantaneous position s0, instantaneous speed v0, level l0, instantaneous power p0 and other data information of the train in each cycle of multiple consecutive control cycles, the above historical operation data is processed by a machine learning regression algorithm, and the following sub-model is defined and a train operation simulation model is established, wherein the cost function adopts the mean square value, and the model is optimized using the gradient descent method, and finally the various model parameters in the following train operation simulation model are obtained, wherein the level sequence composed of the levels corresponding to each cycle of multiple consecutive control cycles of the train constitutes an initial sequence, and the instantaneous position s0, instantaneous speed v0, level l0, instantaneous power p0 and the like of the train are the operating parameters of the train.

[0083] The speed level sub-model is defined as follows:

[0084] The input of the speed level sub-model is the simulation instantaneous speed of the previous control cycle. and the output level from m cycles ago to the level of the current control cycle The output of the speed level sub-model is the simulation instantaneous speed of the current control cycle is the model parameter, t tick Indicates the duration of a control cycle.

[0085] The position and velocity sub-model is defined as follows:

[0086] The input of the position-speed sub-model is the instantaneous position of the simulation in the previous control cycle. From the simulation instantaneous speed m cycles ago to the simulation instantaneous speed of the current control cycle and the level from m cycles ago to the level of the current control cycle t tick Represents the duration of a control cycle. The output of the position-speed sub-model is the simulated instantaneous position of the current control cycle. c -m ,...,c -1 ,c0,d -m ,...,d -1 ,d0,e are model parameters.

[0087] The energy consumption level speed sub-model is defined as follows. Substitute the result of the first formula into the second formula, and the result of the second formula into the third formula to obtain the final energy consumption result:

[0088] Among them, based on the results of the measured data modeling, the energy consumption level speed sub-model is obtained by adopting the segmented description method, and its input is the simulation instantaneous speed of the current control cycle Simulation level of the current control cycle The output of the energy consumption level speed sub-model is the simulated instantaneous power of the current control cycle The model parameters are f1, f2, f3, g0, g1, g2, h0, h1, h2. As described above, based on the above train operation simulation model, namely the speed level position sub-model, the position speed sub-model and the energy consumption level position speed sub-model, a train operation simulation model can be established to simulate the actual operation of the train.

[0089] Step 120: Generate multiple modified sequences based on the initial sequence, wherein the multiple modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage; the initial sequence is a stage sequence consisting of stages corresponding to the train operating at the highest operating efficiency in multiple consecutive control cycles;

[0090] The initial sequence is defined as the sequence of levels corresponding to each control cycle obtained in step 100, when the train is operating at maximum efficiency over multiple consecutive control cycles. This initial sequence is then processed as follows. Since these multiple control cycles are defined based on the maximum delay between the train's traction and braking transitions, there is at least one traction control cycle. The level of at least one non-coasting control cycle in this initial sequence is modified to convert it to coasting, generating a new sequence, which can be referred to as a modified sequence. By repeating the above modification operations, new modified sequences can be continuously generated, ultimately resulting in a series of modified sequences.

[0091] Specifically, for example, if the maximum delay between traction and braking is three control cycles, the initial sequence is traction-coasting-traction-braking. Repeated level correction operations on this initial sequence yield three corrected sequences: coasting-coasting-traction-braking; coasting-coasting-coasting-braking; and traction-coasting-coasting-braking.

[0092] Optionally, multiple modified sequences may be generated according to the initial sequence in a binary tree-generating sequence manner.

[0093] Step 130: input the multiple correction sequences into the train operation simulation model to obtain the operation parameters corresponding to the multiple correction sequences respectively;

[0094] By inputting the above-mentioned multiple correction sequences into the above-mentioned train operation simulation model respectively, simulation operation parameters corresponding to the multiple correction sequences can be obtained.

[0095] Record the instantaneous position, instantaneous speed, and level of each correction sequence in each control cycle. Since the train's signal system stores a route map, the current slope, the length and value of the upcoming slope, the current ceiling speed limit, and the signal system's own control parameters, including braking deceleration, traction acceleration, and tolerable impact rate, can be obtained from the map based on the instantaneous position. Furthermore, the target speed limit can be obtained based on the operation plan. These values ​​are fixed, and further information can be obtained about the slope values ​​of the N upcoming slope segments, the slope lengths of the N upcoming slope segments, the current slope value, the current remaining slope length, and the total energy consumption and total operating time of the correction sequence over all control cycles, where N is an integer greater than 2.

[0096] Step 140: obtaining a target correction sequence from the plurality of correction sequences based on the operating parameters corresponding to the plurality of correction sequences;

[0097] In some embodiments, based on the operating parameters corresponding to the multiple correction sequences respectively, the correction sequences that meet the first condition are eliminated from the multiple correction sequences;

[0098] Based on the total energy consumption of the remaining correction sequences, the correction sequences with abnormal total energy consumption are eliminated from the remaining correction sequences to obtain the target correction sequence;

[0099] The first condition includes at least one of the following:

[0100] The total running time corresponding to the correction sequence exceeds a preset running time threshold;

[0101] Among the multiple consecutive control cycles corresponding to the correction sequence, there is at least one control cycle whose speed is lower than a preset speed threshold.

[0102] Based on the multiple correction sequences obtained in step 130, correction sequences that meet at least one of the following conditions are eliminated: correction sequences whose running time exceeds a preset running time threshold; correction sequences whose speed is lower than a preset speed threshold in at least one of the multiple consecutive control cycles corresponding to the correction sequence. Furthermore, based on the total energy consumption of the remaining correction sequences, correction sequences with abnormal total energy consumption are eliminated from the remaining correction sequences to obtain a target correction sequence. Eliminating correction sequences whose running time exceeds the preset running time threshold is to ensure that the train's operating efficiency is within a reasonable range; eliminating correction sequences whose speed is lower than the preset speed threshold in at least one of the multiple consecutive control cycles corresponding to the correction sequence is to ensure the normal operation of the train and avoid situations such as slow speed causing mid-trip stops; eliminating correction sequences with abnormal total energy consumption from the remaining correction sequences means, for example, eliminating correction sequences with high total energy consumption when the running time is short, or correction sequences with total energy consumption higher than the total energy consumption when the train is operating at maximum efficiency. This is done to optimize a target correction sequence that ensures both high efficiency and energy conservation for the train.

[0103] Specifically, for example, the following evaluation index is designed to evaluate the correction sequence.

[0104] Among them, T and E represent the running time and energy consumption of the train when it is running at the highest operating efficiency, that is, under the control of the initial sequence level. The running time T r and energy consumption E rIndicates the running time and energy consumption of the corrected sequence obtained according to step 130. The corrected sequences with α < K are excluded. Optionally, K ranges from 3 to +∞, and the larger K is, the better the energy-saving effect is.

[0105] In some embodiments, based on the running times respectively corresponding to the remaining corrected sequences, the remaining corrected sequences are sorted from largest to smallest to obtain the corrected sequences in the first order.

[0106] Based on the arrangement order of the energy consumptions respectively corresponding to the corrected sequences in the first order, determine the corrected sequences with abnormal energy consumption.

[0107] Exclude the corrected sequences with abnormal total energy consumption from the remaining corrected sequences to obtain the target corrected sequences.

[0108] Specifically, when arranging the corrected sequences obtained in step 130 from largest to smallest in terms of running time, it is reasonably speculated that the total energy consumption of the train should be from largest to smallest, but there are abnormal situations where the running time of the train is small but the energy consumption is high. In addition, there may also be abnormal corrected sequences where the total energy consumption is higher than the total energy consumption when the train runs at the highest operating efficiency. In these cases, the corrected sequences with abnormal energy consumption situations need to be excluded, and the remaining corrected sequences obtained therefrom are the target corrected sequences. As described above, the target corrected sequences are a combination of control levels with relatively balanced and excellent operating efficiency and energy consumption. That is, when the train is controlled using the levels in the target corrected sequences in multiple operating cycles, the energy consumption generated during operation is lower than the energy consumption generated when the train runs at the highest operating efficiency, and the operating efficiency is within a reasonable range.

[0109] Specifically, sort the remaining corrected sequences after exclusion from smallest to largest in terms of running time, determine whether the order of energy consumption of each corrected sequence is from smallest to largest, and exclude the samples whose energy consumption does not conform to the sorting to form an optimal training set as the input of the train control model in step 150. For example, exclude corrected sequence 2 in Table 1 below, and the remaining corrected sequences 1, corrected sequence 3, and corrected sequence 4 are the target corrected sequences.

[0110] Step 150, based on the target corrected sequences among the multiple corrected sequences, train to obtain a train control model.

[0111] In some embodiments, using the relevant parameters of the control cycle corresponding to the traction level in the target corrected sequence as training samples and whether to change from the traction level to the inert level as labels, train to obtain a train control model.

[0112] Among them, the relevant parameters of the control cycle include at least one of the following: remaining running time, remaining distance, slope of the next N slope sections, length of the next N slope sections, current target speed limit, current ceiling speed limit, braking deceleration, traction acceleration, and tolerable impact rate, where N is an integer greater than 2.

[0113] Specifically, a train control model is trained using the relevant parameters of the control cycle corresponding to the traction gear in the target correction sequence as training samples, with whether the traction gear is corrected to the inertia gear as a label. The relevant parameters of the control cycle include at least one of the following: remaining run time, remaining distance, the gradient of the N preceding gradient segments, the length of the N preceding gradient segments, the current target speed limit, the current ceiling speed limit, the braking deceleration, the traction acceleration, and the tolerable impact rate, where N is an integer greater than 2. Specifically, according to step 130, the instantaneous position, instantaneous speed, gear, the gradient of the N preceding gradient segments, the length of the N preceding gradient segments, the current gradient value, the current remaining gradient length, the target speed limit, the ceiling speed limit, the braking deceleration, the traction acceleration, the tolerable impact rate, the energy consumption, and the run time of the target correction sequence in each control cycle are collected, where N is an integer greater than 2. Based on the operating condition transition point at which the target correction sequence is located, the remaining run time and remaining distance in each traction control cycle can be calculated. The remaining run time, remaining distance, the slope values ​​of the next N slope segments, the slope lengths of the next N slope segments, the target speed limit, the ceiling speed limit, the braking deceleration, the traction acceleration, and the tolerable impact rate of the target correction sequence are used as training samples, where N is an integer greater than 2. Furthermore, the gear correction status of each sample is labeled. For example, in one embodiment, (0) indicates that the traction gear is corrected to coasting, and (1) indicates that the traction gear is maintained.

[0114] According to the above-mentioned train control model training method, since only the traction level in the operating data is corrected, it is easy to transplant and has high robustness; using the historical operating data of the train running at the highest operating efficiency, a large number of train control model inputs can be automatically generated; the obtained target correction sequence can take into account both efficiency and energy saving compared with the initial sequence when running at the highest operating efficiency; the model can be optimized and iterated as the input is updated; since it does not rely on human experience and judgment, it can identify certain operating methods that take into account both efficiency and energy saving that humans may miss.

[0115] Optionally, based on a target correction sequence from the multiple correction sequences, training the train control model further includes selecting an appropriate neural network structure comprising neurons in an input layer, a hidden layer, and an output layer. Optionally, the input and hidden layer neurons use a Relu activation function, and the output layer neurons use a Sigmoid activation function. The output of the deep learning model is 0 / 1. Furthermore, an appropriate loss function is selected based on actual needs to measure the difference between the model's predictions and the true labels.

[0116] Optionally, based on the target correction sequence in the multiple correction sequences, training the train control model also includes updating network parameters through a backpropagation algorithm and an optimization algorithm (such as stochastic gradient descent) to minimize the loss function.

[0117] Optionally, a validation dataset can be used to evaluate the train control model during training. Based on the performance indicators of the validation dataset, such as accuracy, precision, and recall, the train control model can be fine-tuned, such as by adjusting the network structure and hyperparameters.

[0118] Optionally, when the train condition, road conditions, or train control technology are changed, which affects the train operation data, the historical operation data of the train running at the highest efficiency can be re-collected, thereby easily updating and iterating the train operation simulation model and train control model in the present invention.

[0119] Optionally, the train operation simulation model and the train control model in the present invention can be continuously optimized by collecting more historical operation data of trains running at the highest operation efficiency as input to step 100 .

[0120] The train control method provided by the present invention is described below with reference to FIG. 2 . The train control method described below and the training method of the train control model described above can be referenced to each other.

[0121] On the other hand, the present invention also provides a train control method, as shown in FIG2 , comprising:

[0122] Step 200, obtaining relevant parameters of the train in real time;

[0123] Step 210: Input the real-time relevant parameters of the train into the train control model to obtain the target level output by the train control model;

[0124] Step 220, controlling the train operation based on the target level;

[0125] The training method of the train control model includes:

[0126] Collect historical operating data of trains running at the highest efficiency;

[0127] Based on the historical operation data, a train operation simulation model is established using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence consisting of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles;

[0128] Based on the initial sequence, a plurality of modified sequences are generated, wherein the plurality of modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage; the initial sequence is a stage sequence consisting of stages corresponding to a train operating at the highest operating efficiency in a plurality of consecutive control cycles;

[0129] Inputting the multiple correction sequences into the train operation simulation model respectively to obtain operation parameters corresponding to the multiple correction sequences respectively;

[0130] obtaining a target correction sequence among the multiple correction sequences based on the operating parameters respectively corresponding to the multiple correction sequences;

[0131] Training a train control model based on a target correction sequence among the multiple correction sequences;

[0132] The operating parameters include one or more of the following:

[0133] Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0134] Specifically, when applying the above-mentioned train control method to a train in actual operation, it is necessary to first train the train using the train control model training method to obtain the train control model of the train. For details, please refer to the above and will not be repeated here. In the case where the train control model of the train is trained, the train control method specifically includes the following steps: collecting and obtaining the real-time relevant parameters of the train on the running train. Specifically, the relevant parameters of the train include: remaining running time, remaining distance, the slope of the N slope sections ahead, the length of the N slope sections ahead, the current target speed limit, the current ceiling speed limit, the braking deceleration, the traction acceleration, the tolerable impact rate, and N is an integer greater than 2; inputting the real-time relevant parameters of the train into the train control model to obtain the target level output by the train control model; and controlling the operation of the train based on the target level.

[0135] The present invention also provides a training device 300 for a train control model, as shown in FIG3 , comprising:

[0136] The collection module 310 is used to collect historical operation data of trains running at the highest operating efficiency;

[0137] a model building module 320 for building a train operation simulation model based on the historical operation data using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence consisting of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles;

[0138] The sequence generation module 330 is configured to generate a plurality of modified sequences based on the initial sequence, wherein the plurality of modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage. The initial sequence is a stage sequence consisting of stages corresponding to the train operating at the highest operating efficiency in multiple consecutive control cycles.

[0139] a parameter acquisition module 340 for inputting the plurality of correction sequences into the train operation simulation model to obtain operation parameters corresponding to the plurality of correction sequences;

[0140] a target correction sequence acquisition module 350, configured to obtain a target correction sequence from the plurality of correction sequences based on the operating parameters corresponding to the plurality of correction sequences;

[0141] A training module 360 ​​is configured to train a train control model based on a target correction sequence among the plurality of correction sequences;

[0142] The operating parameters include one or more of the following:

[0143] Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0144] FIG4 illustrates a schematic diagram of the physical structure of an electronic device. As shown in FIG4 , the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440. The processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call logic instructions in the memory 430 to execute a train control method, which includes:

[0145] Obtain real-time relevant parameters of the train;

[0146] Inputting the real-time relevant parameters of the train into a train control model to obtain a target level output by the train control model;

[0147] Controlling the train operation based on the target level;

[0148] The training method of the train control model includes:

[0149] Collect historical operating data of trains running at the highest efficiency;

[0150] Based on the historical operation data, a train operation simulation model is established using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence consisting of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operating parameters of the train in the multiple consecutive control cycles;

[0151] Based on the initial sequence, a plurality of modified sequences are generated, wherein the plurality of modified sequences are different from each other and are obtained by modifying at least one traction stage in the initial sequence to a coasting stage; the initial sequence is a stage sequence consisting of stages corresponding to a train operating at the highest operating efficiency in a plurality of consecutive control cycles;

[0152] Inputting the multiple correction sequences into the train operation simulation model respectively to obtain operation parameters corresponding to the multiple correction sequences respectively;

[0153] obtaining a target correction sequence among the multiple correction sequences based on the operating parameters respectively corresponding to the multiple correction sequences;

[0154] Training a train control model based on a target correction sequence among the multiple correction sequences;

[0155] The operating parameters include one or more of the following:

[0156] Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

[0157] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0158] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned train control model training method or the above-mentioned train control method.

[0159] On the other hand, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned train control model training method or the above-mentioned train control method.

[0160] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0161] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for training a train control model, characterized in that: include: Collect historical operation data of trains running at the highest operating efficiency; Based on the historical operation data, a train operation simulation model is established by using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence composed of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes the operation parameters of the train in the multiple consecutive control cycles; Based on the initial sequence, a plurality of modified sequences are generated, wherein the plurality of modified sequences are different from each other, and the modified sequence is obtained by modifying at least one traction level in the initial sequence to an idling level; the initial sequence is a level sequence composed of levels corresponding to the train running at the highest operating efficiency in a plurality of consecutive control cycles; Inputting the multiple correction sequences into the train operation simulation model respectively to obtain the operation parameters corresponding to the multiple correction sequences respectively; Based on the operating parameters respectively corresponding to the multiple correction sequences, a target correction sequence among the multiple correction sequences is obtained; Based on the target correction sequence among the multiple correction sequences, training a train control model; The operating parameters include one or more of the following: Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

2. The method for training a train control model according to claim 1, characterized in that: The method of establishing a train operation simulation model based on the historical operation data and using a machine learning regression algorithm includes: Taking the historical operation data as input, the parameters of the speed level position sub-model, the position speed sub-model and the energy consumption level position speed sub-model are solved by the gradient descent method to construct the train operation simulation model; The speed level position model is expressed as: The input of the speed level sub-model includes the simulation instantaneous speed corresponding to the previous control cycle and the level sequence consisting of the levels corresponding to the control cycles from m control cycles ago to the current control cycle The output of the speed level sub-model is the simulation instantaneous speed corresponding to the current control cycle The model parameter of the speed level position sub-model is a -m ,...,a -1 ,a0,b,t tick Indicates the duration of a control cycle; The position-velocity submodel is expressed as: The input of the position-speed sub-model includes the simulation instantaneous position corresponding to the previous control cycle. The speed sequence consisting of the speeds corresponding to the control cycles from m control cycles ago to the current control cycle and the level sequence consisting of the levels corresponding to the control cycles from m control cycles ago to the current control cycle The output of the position and speed sub-model is the simulated instantaneous position corresponding to the current control cycle. The model parameter of the position and velocity sub-model is c -m ,...,c -1 ,c0,d -m ,...,d -1 ,d0,e,t tick Indicates the duration of a control cycle; The energy consumption level velocity sub-model is expressed as: The input of the energy consumption level speed sub-model includes the corresponding speed The level corresponding to the current control cycle The output of the energy consumption level speed sub-model is the energy consumption corresponding to the current control cycle. The model parameters of the energy consumption level speed sub-model are f1, f2, f3, g0, g1, g2, h0, h1, h2.

3. The method for training a train control model according to claim 1, characterized in that: The obtaining a target correction sequence among the multiple correction sequences based on the operating parameters respectively corresponding to the multiple correction sequences includes: Based on the operating parameters respectively corresponding to the multiple correction sequences, eliminating the correction sequences that meet the first condition from the multiple correction sequences; Based on the total energy consumption of the remaining correction sequences, the correction sequences with abnormal total energy consumption are eliminated from the remaining correction sequences to obtain the target correction sequence; The first condition includes at least one of the following: The total running time corresponding to the correction sequence exceeds a preset running time threshold; Among the multiple continuous control cycles corresponding to the correction sequence, there is at least one control cycle whose speed is lower than a preset speed threshold.

4. The method for training a train control model according to claim 1, characterized in that: The total energy consumption based on the remaining correction sequences is used to remove the correction sequences with abnormal total energy consumption from the remaining correction sequences to obtain the target correction sequence, including: Based on the running times respectively corresponding to the remaining correction sequences, the remaining correction sequences are sorted from large to small to obtain a correction sequence of the first order; Determine a correction sequence for abnormal energy consumption based on the arrangement order of energy consumptions respectively corresponding to the correction sequences of the first order; The correction sequence with abnormal total energy consumption is eliminated from the remaining correction sequences to obtain a target correction sequence.

5. The method for training a train control model according to claim 1, characterized in that: The training of a train control model based on a target correction sequence among the multiple correction sequences includes: The train control model is obtained by training by using the relevant parameters of the control cycle corresponding to the traction level in the target correction sequence as training samples and whether the traction level is corrected to the inertia level as a label; Among them, the relevant parameters of the control cycle include at least one of the following: remaining running time, remaining distance, slope of the N slope sections ahead, length of the N slope sections ahead, current target speed limit, current ceiling speed limit, braking deceleration, traction acceleration, and tolerable impact rate, where N is an integer greater than 2.

6. A train control method, characterized in that: include: Obtain relevant parameters of the train in real time; Inputting the real-time relevant parameters of the train into the train control model to obtain the target level output by the train control model; Based on the target level, controlling the operation of the train; The training method of the train control model includes: Collect historical operation data of trains running at the highest operating efficiency; Based on the historical operation data, a train operation simulation model is established by using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence composed of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes the operation parameters of the train in the multiple consecutive control cycles; Based on the initial sequence, a plurality of modified sequences are generated, wherein the plurality of modified sequences are different from each other, and the modified sequence is obtained by modifying at least one traction level in the initial sequence to an idling level; the initial sequence is a level sequence composed of levels corresponding to the train running at the highest operating efficiency in a plurality of consecutive control cycles; Inputting the multiple correction sequences into the train operation simulation model respectively to obtain the operation parameters corresponding to the multiple correction sequences respectively; Based on the operating parameters respectively corresponding to the multiple correction sequences, a target correction sequence among the multiple correction sequences is obtained; Based on the target correction sequence among the multiple correction sequences, training a train control model; The operating parameters include one or more of the following: Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

7. A training device for a train control model, characterized in that: include: A collection module for collecting historical operation data of trains running at the highest operating efficiency; A model building module, used to build a train operation simulation model based on the historical operation data using a machine learning regression algorithm, wherein the input of the train operation simulation model includes a level sequence composed of levels corresponding to the train in multiple consecutive control cycles, and the output of the train operation simulation model includes operation parameters of the train in the multiple consecutive control cycles; A sequence generation module is used to generate multiple modified sequences based on an initial sequence, wherein the multiple modified sequences are different from each other, and the modified sequence is obtained by correcting at least one traction level in the initial sequence to an idling level; the initial sequence is a level sequence composed of levels corresponding to the train running at the highest operating efficiency in multiple consecutive control cycles; A parameter acquisition module, used to input the multiple correction sequences into the train operation simulation model respectively, and obtain the operation parameters corresponding to the multiple correction sequences respectively; a target correction sequence acquisition module, configured to obtain a target correction sequence among the multiple correction sequences based on the operating parameters respectively corresponding to the multiple correction sequences; A training module, used for training a train control model based on a target correction sequence among the multiple correction sequences; The operating parameters include one or more of the following: Total running time, total energy consumption, speed corresponding to each control cycle, and position corresponding to each control cycle.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the training method of the train control model as described in any one of claims 1 to 5, or implements the train control method as described in claim 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the training method of the train control model as described in any one of claims 1 to 5, or implements the train control method as described in claim 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the training method of the train control model as described in any one of claims 1 to 5, or implements the train control method as described in claim 6.

Images