Energy-saving operation optimization method for high-speed railway trains on long and steep slopes based on traction load prediction

Abstract

The present invention provides a method for optimizing the energy-saving operation of a high-speed railway train on a long and steep slope based on traction load prediction, and relates to the technical field of train operation optimization. The method for optimizing the energy-saving operation of a high-speed railway train on a long and steep slope based on traction load prediction comprises the following steps: establishing a traction load prediction model, obtaining traction load prediction data; and optimizing the energy-saving operation of the train according to the traction load prediction data. By optimizing the energy-saving operation of a high-speed railway train on a long and steep slope based on traction load prediction, on the basis of traction load prediction, the utilization of the regenerative braking energy generated by a braking train on a long and steep slope is promoted within the system, and the traction energy obtained by a traction train on a long and steep slope from an external power grid is reduced, which is conducive to reducing the operating cost of high-speed railways and achieving energy-saving operation of the system.

Description

Technical Field

[0001] The invention relates to the technical field of train operation optimization, and in particular to a method for optimizing the energy-saving operation of a high-speed railway long and steep slope train with traction load prediction. Background Art

[0002] Since my country's terrain is high in the west and low in the east, the railway lines under construction in the central and western regions, represented by the Sichuan-Tibet, Chengdu-Lanzhou, Chongqing-Kunming, Xi'an-Yanzhou, and Xi'an-Shijiazhuang railways, generally have dense long and steep slopes. When a train passes through a long and steep slope in the uphill direction, it needs a sufficiently high traction power, and the traction energy consumption increases dramatically. When a train passes through a long and steep slope in the downhill direction, the long-term continuous application of electric braking will generate abundant regenerative braking energy. The regenerative braking energy generated by the train braking on the long downhill slope can be used to supplement the traction energy demand of the train on the long uphill slope, saving external power supply.

[0003] Therefore, by predicting the traction load on high-speed railway lines with long and steep slopes, we can understand the behavioral characteristics and energy consumption requirements of the traction load within the power supply range, provide a reference for the railway department to formulate power consumption plans and optimize traffic organization, promote the utilization of regenerative braking energy generated by trains on long and steep slope lines within the traction power supply system, achieve energy-saving operation, and reduce the operating costs of high-speed railways. Summary of the invention

[0004] The purpose of the present invention is to provide a method for optimizing the energy-saving operation of high-speed railway long and steep slope trains based on traction load prediction, so as to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The method for optimizing energy-saving operation of high-speed railway long and steep slope trains based on traction load prediction is characterized by comprising the following steps:

[0007] Step 1: Establish a traction load prediction model and obtain traction load prediction data;

[0008] Step 2: Optimize train energy-saving operation based on traction load prediction data.

[0009] Furthermore, the step 1 establishes a traction load prediction model based on CNN-LSTM-Attention. The input of the model is four parameters: traction load historical data, line condition influencing factors, train parameter influencing factors, and operation organization influencing factors. The model structure is mainly divided into six layers, namely input layer, CNN layer, LSTM layer, fully connected layer, Attention layer, and output layer.

[0010] Furthermore, the train energy-saving operation optimization in step 2 includes single-train energy-saving operation and multi-train operation matching.

[0011] Furthermore, the single-train energy-saving operation steps include dividing the operating range, generating an ideal operating strategy, and allocating traction energy, and the multi-train operation matching steps include determining the ramp crossing number, transmitting regenerative braking energy, and overlapping braking / traction conditions.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] 1. The present invention uses the historical data of traction load, considers the factors affecting traction load, and establishes a CNN-LSTM-Attention traction load prediction model to obtain the traction load data of high-speed railway long and steep slopes; according to the traction load prediction data, a train energy-saving operation optimization scheme is proposed that coordinates the saving of traction energy and the traction load peak reduction, which is divided into two categories: single train energy-saving operation and multi-train operation matching. Through the optimization of high-speed railway long and steep slope train energy-saving operation based on traction load prediction, a solution can be proposed for the current problem that the traction energy consumption of high-speed railway long and steep slope trains is high and the regenerative braking energy is not fully utilized. On the basis of traction load prediction, the traction energy saving and traction load peak reduction are coordinated, and a train energy-saving operation optimization scheme is proposed to promote the utilization of the regenerative braking energy generated by the slope braking condition train within the system, reduce the traction energy consumption obtained by the slope traction condition train from the external power grid, and reduce the operating cost of high-speed railways. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the implementation steps of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] See also Figure 1 , embodiments of the present invention:

[0017] The embodiment of the present invention provides a method for optimizing the operation of a high-speed railway long and steep slope train based on traction load prediction, comprising the following steps:

[0018] Step 1: Establish a traction load prediction model and obtain traction load prediction data;

[0019] Since traction load data is not only related to the train operation process, but also affected by line conditions, train parameters, operation organization and other factors, the input of the prediction model is the multivariate time series of traction load historical data and traction load influencing factors, and there is correlation between the input variables. The convolutional neural network (CNN) is used to automatically extract the feature information of traction load historical data and traction load influencing factors, and then the long short-term memory (LSTM) network is used to capture the temporal dependency relationship between the extracted feature information. After being processed into a one-dimensional traction load feature vector through a fully connected layer, it is input into the attention model for training, and corresponding weights are assigned to different input features, thereby achieving high-precision traction load prediction.

[0020] The input of the traction load prediction model based on CNN-LSTM-Attention is: traction load historical data, line condition influencing factors, train parameter influencing factors, and operation organization influencing factors. The values ​​of the four parameters are connected in series to form a vector representation. The input of every four moments constitutes a characteristic vector of a unit traction load. The training samples of the traction load prediction model are constructed using a sliding window. The time step of the sliding window is set to 1, and the sliding window slides along the time axis with a size of 4*n. Every time step of sliding can generate a training sample of the traction load prediction model.

[0021] The model structure is mainly divided into six layers, as follows:

[0022] 1. Input layer, the traction load data x at time t s , Line condition influence factor l s , train parameter influence factor p s , Operational Organization Impact Factor r s The data is concatenated into a vector Y t The sliding window step size is set to 1. Each time the sliding window slides, a new time series data of the traction load and the influencing variables can be obtained. The new time series data of the traction load and the influencing variables are standardized and used as the input of the CNN-LSTM-Attention model.

[0023] 2. CNN layer, extracts features from the input train traction load data and the time series data of the influencing variables. The relative positions of the trains at 4 moments and the time series numbers of the influencing variables form a traction load feature matrix with a size of 4*4. This layer consists of 2 convolutional layers and 2 pooling layers. The number of convolution kernels used in each convolutional layer doubles layer by layer to increase the network depth, and the pooling layer selects the maximum pooling method. The reLU function is selected as the activation function of the convolutional layer. The CNN layer automatically extracts the features of the input traction load and the time series data of the influencing variables through convolution kernel operations.

[0024] 3. LSTM layer: LSTM has a strong ability to learn long time series and has the ability to automatically encode and select important information. A two-layer LSTM network is selected, and the feature vector extracted by the CNN layer is used as input to learn its internal change rules to capture the time dependency relationship between data and perform time series modeling. The two-layer LSTM network also helps to reduce the model training time.

[0025] 4. The fully connected layer integrates the traction load feature matrix to obtain a one-dimensional traction load feature vector, realizing the transformation of the data dimension, thereby completely passing the features extracted by the LSTM layer to the Attention layer.

[0026] 5. Attention layer: Attention training is performed on the traction load feature vector to obtain the weights of factors affecting the traction load prediction and reduce the weights of factors with weak correlation with the actual load in order to achieve more accurate traction load prediction.

[0027] 6. Output layer, obtain the predicted value of traction load at t+1 and get the final traction load prediction result.

[0028] Step 2: Optimize train energy-saving operation based on traction load prediction data;

[0029] According to the traction load prediction data based on CNN-LSTM-Attention, regenerative braking energy is used to supplement traction energy consumption, achieve energy-saving operation, and improve the traction load peak shaving effect. A train energy-saving operation optimization scheme that coordinates traction energy saving and traction load peak shaving is proposed, which is divided into two categories: single train energy-saving operation and multi-train operation matching:

[0030] 1. Single train energy-saving operation

[0031] The traction load results based on CNN-LSTM-Attention are used as the data basis. The goal of saving train traction energy consumption is achieved by allocating the minimum energy consumption to the typical operating stage and meeting the specified operating hours. Since the train does not consume traction energy during the acceleration-cruise-coasting-braking operating condition conversion, the optimal allocation of the minimum energy consumption in the acceleration and cruising conditions is mainly considered. The steps are as follows:

[0032] 1) Divide the train operation area into multiple sections, with the line slope unchanged and the speed limit the same as the basis for division;

[0033] 2) Generate an ideal operation strategy consisting of acceleration-cruise-coasting-braking in each section, which conforms to the specified operation time of the section operation as much as possible;

[0034] 3) According to the traction load prediction data, traction energy is allocated to the acceleration and cruising conditions, and the optimal allocation is solved using a heuristic genetic algorithm to obtain the energy-saving speed curve of a single train.

[0035] 2. Multi-train operation matching

[0036] Multi-car operation matching is to achieve the operation matching of traction trains and braking trains in the same power supply section, and transfer the regenerative braking energy generated by the braking train during a long downhill section to the traction train in the uphill section of the same power supply section. Specifically, by overlapping or extending the traction / braking time between trains in the same power supply section, the net value of train traction energy consumption can be reduced and the traction load peak reduction effect can be achieved. The steps are as follows:

[0037] 1) According to the train operation diagram, the up and down trains that may intersect on the slope section are analyzed, corresponding to the load curve mutation section in the traction load prediction data based on CNN-LSTM-Attention, and the train numbers with overlapping traction / braking conditions are determined;

[0038] 2) Calculate the transferable regenerative braking energy, only consider the regenerative braking energy utilization within the same power supply interval, and the excess regenerative braking energy will be dissipated through the resistor, and will not cause traction network voltage fluctuations during the energy transmission process;

[0039] 3) Coordinate the overlap of braking / traction conditions between multiple trains, and extend the overlap time of braking / traction conditions as much as possible by allocating interval redundant time on the basis of meeting the specified operating time, so as to maximize the transmission and consumption of regenerative braking energy within the same power supply interval.

[0040] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0041] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A high-speed railway long and steep slope train energy-saving operation optimization method based on traction load prediction, characterized in that: The following steps are involved: Step 1: Establish a traction load prediction model and obtain traction load prediction data; The step 1 establishes a traction load prediction model based on CNN-LSTM-Attention. The input of the model is: traction load historical data, line condition influencing factors, train parameter influencing factors, and operation organization influencing factors. The model structure is mainly divided into six layers, namely input layer, CNN layer, LSTM layer, fully connected layer, Attention layer, and output layer; the model structure is mainly divided into six layers, as follows: (1) Input layer, the traction load data x at time t s , Line condition influence factor l s , train parameter influence factor p s , Operational Organization Impact Factor r s The data is concatenated into a vector Y t , set the sliding window step size to 1. Each time the sliding window slides, a new time series data of traction load and influencing variables can be obtained. The new time series data of traction load and influencing variables are standardized and used as the input of the CNN-LSTM-Attention model; (2) CNN layer, which extracts features from the input train traction load data and the time series data of the influencing variables. The relative positions of the trains at four moments and the time series numbers of the influencing variables form a traction load feature matrix with a size of 4*4. The CNN layer consists of two convolutional layers and two pooling layers. The number of convolution kernels used in each convolutional layer doubles layer by layer to increase the network depth. The pooling layer selects the maximum pooling method and selects the reLU function as the activation function of the convolutional layer. The CNN layer automatically extracts the features of the input traction load and the time series data of the influencing variables through convolution kernel operations; (3) LSTM layer: select a two-layer LSTM network, take the feature vector extracted by the CNN layer as input, learn its internal change rules to capture the temporal dependency relationship between data, and perform time series modeling; (4) The fully connected layer integrates the traction load feature matrix to obtain a one-dimensional traction load feature vector, realizing the transformation of the data dimension, thereby completely transferring the features extracted by the LSTM layer to the Attention layer; (5) Attention layer, which performs attention training on the traction load feature vector to obtain the weights of factors affecting traction load prediction; (6) Output layer, obtain the predicted value of traction load at t+1 and obtain the final traction load prediction result; Step 2: Optimize train energy-saving operation based on traction load prediction data; According to the traction load prediction data based on CNN-LSTM-Attention, a train energy-saving operation optimization scheme is proposed that coordinates the saving of traction energy and the traction load peak reduction. It is divided into two categories: single-train energy-saving operation and multi-train operation matching: (1) Single train energy-saving operation The traction load results based on CNN-LSTM-Attention are used as the data basis. The goal of saving train traction energy consumption is achieved by allocating the minimum energy consumption to the typical working condition stage and meeting the specified running time. Since the train does not consume traction energy during the acceleration-cruise-coasting-braking working condition conversion, the coasting and braking working conditions mainly consider the optimal allocation of the minimum energy consumption in the acceleration and cruising working conditions. The steps are as follows: ① Divide the train operation area into multiple sections, with the line slope unchanged and the speed limit the same as the basis for division; ② Generate an ideal operation strategy consisting of acceleration-cruise-coasting-braking in each section, so as to comply with the specified operation time of the section operation as much as possible; ③ According to the traction load prediction data, traction energy is allocated to acceleration and cruising conditions, and the optimal allocation is solved using a heuristic genetic algorithm to obtain a single train energy-saving speed curve; (2) Multi-train operation matching By overlapping or extending the traction / braking time between trains in the same power supply section, the net value of train traction energy consumption can be reduced and the traction load peak reduction effect can be achieved. The steps are as follows: ① According to the train operation diagram, the up and down trains that may intersect on the slope section are analyzed, corresponding to the load curve mutation section in the traction load prediction data based on CNN-LSTM-Attention, and the train numbers with overlapping traction / braking conditions are determined; ② Calculate the transferable regenerative braking energy, consider the utilization of regenerative braking energy within the same power supply interval, and the excess regenerative braking energy will be dissipated through the resistor, and will not cause traction network voltage fluctuations during energy transmission; ③ Coordinate the overlap of braking / traction conditions among multiple trains, and extend the overlap time of braking / traction conditions as much as possible by allocating interval redundant time on the basis of meeting the specified operating time, so as to maximize the transmission and consumption of regenerative braking energy within the same power supply interval.

2. The energy-saving operation optimization method for high-speed railway long and steep slope trains based on traction load prediction according to claim 1 is characterized in that: The energy-saving operation steps of a single train include dividing the operating range, generating an ideal operating strategy, and allocating traction energy. The operation matching steps of multiple trains include determining the number of trains to pass on the slope, transmitting regenerative braking energy, and overlapping braking / traction conditions.

Images