A method for predicting the workload of station shunting operations

Abstract

The present invention discloses a method for predicting the workload of shunting operations at a station. First, time series data for measuring the workload of shunting operations are constructed through wireless shunting locomotive signals and monitoring system STP data; each subsequence data of the constructed time series data is distributed to Hadoop cluster nodes, the DTW distance matrix is ​​divided into multiple sub-matrices, and the time series DTW distance is obtained by a parallel calculation method; the scene with the highest similarity to the current data in the historical data is searched according to the obtained DTW distance, and the workload of shunting operations at the station is predicted based on this. The method can effectively predict the workload of shunting operations at the station in a specific period of time, and appropriately allocate transportation resources based on this, providing a judgment basis for relevant railway departments in transportation scheduling and command.

Description

Technical Field

[0001] The invention relates to the technical field of railway transportation research, and in particular to a method for predicting the workload of station shunting operations. Background Art

[0002] Balanced transportation is an important principle of railway transportation organization. Each station basically has fixed transportation resources, but this method will cause waste or tension of transportation resources, that is, unreasonable use of transportation resources. Therefore, it is one of the important topics of railway transportation research to reasonably use transportation resources according to the busyness of station operations at different times, improve station operation organization methods and shunting operation efficiency. Wireless shunting locomotive signal and monitoring system (STP, Shunting Train Protection) is an important auxiliary equipment for shunting safety protection. Through its ground equipment, it processes the relevant signals, turnouts, and track circuit section information of the centralized interlocking station shunting operation. The train and the ground adopt wireless communication, and the train operation monitoring device is connected to the on-board equipment to realize the shunting locomotive signal display and train speed monitoring and control. The ground host computer stores all the data generated during the shunting operation process. STP data includes interlocking data (real-time display of station equipment), locomotive data (locomotive speed mileage, etc.), operation order data (number of unhooking, unhooking trains, etc.) and other important data.

[0003] The existing technology measures the workload of shunting operations by simply counting macro indicators such as the number of uncoupling and coupling trains and the number of uncoupling and coupling times each year. It cannot measure the workload of shunting operations in all aspects from the perspective of big data, and lacks a solution to predict the workload of shunting operations at stations. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting the workload of station shunting operations, which can effectively predict the workload of station shunting operations in a specific period of time, and accordingly properly allocate transportation resources, providing a basis for judgment for relevant railway departments in transportation scheduling and command.

[0005] The objective of the present invention is achieved through the following technical solutions:

[0006] A method for predicting the workload of station shunting operations, the method comprising:

[0007] Step 1: construct time series data for measuring the workload of shunting operations through wireless shunting locomotive signals and monitoring system STP data; wherein the STP data is ground signal interlocking and locomotive operation status data related to shunting operations recorded during the operation of the station STP, specifically including interlocking data, control data, locomotive data, work order data, and tracking data;

[0008] Step 2: Distribute each subsequence data of the constructed time series data to the Hadoop cluster nodes, divide the dynamic time warping DTW distance matrix into multiple sub-matrices, and use parallel computing methods to obtain the DTW distance of the time series data;

[0009] Step 3: Search the scene with the highest similarity to the current data in the historical data based on the obtained DTW distance, and use this as a basis to predict the workload of the station shunting operation.

[0010] It can be seen from the technical solution provided by the present invention that the above method can effectively predict the shunting workload of the station in a specific period of time, and appropriately allocate transportation resources based on this, providing a basis for judgment for relevant railway departments in transportation scheduling and command. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0012] Figure 1 A schematic flow chart of a method for predicting the workload of station shunting operations provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0013] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings 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, which does not constitute a limitation of the present invention. 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.

[0014] like Figure 1 The figure is a schematic flow chart of a method for predicting the workload of station shunting operations provided by an embodiment of the present invention, the method comprising:

[0015] Step 1: Construct time series data to measure the workload of shunting operations through wireless shunting locomotive signals and monitoring system STP data;

[0016] Among them, the STP data is the ground signal interlocking and locomotive operation status data related to shunting operations recorded during the operation of the station STP, specifically including interlocking data, control data, locomotive data, work order data, and tracking data.

[0017] In the specific implementation, factors related to the workload of shunting operations are obtained through locomotive data and work order data in STP data, and normalized to calculate the influencing factors of shunting workload per unit day;

[0018] The final value for measuring the workload of shunting operations is then derived by taking into account the number of occupied throat switches, open routes, open shunting signals, and section occupancy in the interlocking data, combined with the calculated shunting workload influencing factors per unit day.

[0019] For example, the above process of constructing time series data to measure the workload of shunting operations is as follows:

[0020] Since the workload of shunting operations is related to many factors, we can first count the mileage, number of uncoupling trains, and number of shunting operation hooks of all shunting operations in the station every day on a daily basis; the mileage is obtained from the locomotive data, and the number of uncoupling trains and the number of shunting operation hooks are obtained from the operation order data. Since the unit ranges of the three types of data are different, they are first normalized separately. The normalization method used is maximum-minimum normalization, and the formula is:

[0021]

[0022] In the formula, x′ is the dimensionless value of [0,1] after normalization; x is the original value of the data; minA is the minimum value of the range of x; maxA is the maximum value of the range of x;

[0023] According to the actual situation, the shunting operation mileage of a single shunting car in a station in a day generally does not exceed 400 kilometers, so the maximum shunting mileage is the number of shunting cars in the station multiplied by 400 kilometers, and the minimum value is 0; the maximum number of unhooking trains in a shunting operation in a station in a day is 40,000, and the minimum value is 0; the maximum number of hooks in a shunting operation in a day is, and the minimum value is 0. Then calculate the normalized values ​​of the three types of data respectively, assign weights to different data, and then add them together to get the value of measuring the workload of the station's shunting operation in a day. The formula is:

[0024] y=εa+γb+(1-ε-γ)c

[0025] In the formula, y is the influencing factor of the shunting workload per unit day; ε is the weight coefficient of the shunting kilometers; a is the normalized shunting kilometers of the station; γ is the weight coefficient of the number of detached and trailer trains in the shunting operation; b is the normalized number of detached and trailer trains in the station shunting operation; c is the normalized number of shunting operation hooks;

[0026] The influencing factor y of the shunting workload per day is calculated. Because y is determined by the shunting kilometers per day, the number of uncoupling and trailering trains, and the number of shunting operations, these data are relatively macroscopic and cannot always reflect the busyness of the station's shunting operations, so it is used as an influencing factor.

[0027] The data that can always reflect the busyness of shunting operations are interlocking data that reflect the changes in the station in real time. Here, the number of occupied throat switch groups, the number of open routes, the number of open shunting signal machines, and the number of occupied sections are used to measure the workload of station shunting operations. First, these four types of data are normalized, then accumulated according to the weights, and finally the impact factor y of the shunting workload per unit day is added to obtain the final value for measuring the workload of shunting operations. The formula is:

[0028] m=η(ωd+αe+θf+(1-ω-α-θ)g)+ρy′

[0029] Wherein, m is the final value for measuring the workload of station shunting operations; η is the weight coefficient of interlocking data; ω is the weight coefficient of the occupied number of throat switch groups; d is the normalized occupied number of throat switch groups; α is the weight coefficient of the number of open routes; e is the normalized number of open routes; θ is the weight coefficient of the open number of shunting signals; f is the normalized open number of shunting signals; g is the normalized number of section occupancy; ρ is the weight coefficient of the influencing factor; y′ is the normalized value of the influencing factor y.

[0030] Step 2: Distribute each subsequence data of the constructed time series data to the Hadoop cluster nodes, divide the dynamic time warping (DTW) distance matrix into multiple sub-matrices, and use parallel computing methods to obtain the DTW distance of the time series data;

[0031] In this step, Hadoop is a distributed architecture for big data; DTW distance is a method for measuring the similarity between two time series of different lengths, and is mainly used in template matching, such as isolated word speech recognition (recognizing whether two speech segments represent the same word), gesture recognition, data mining, and information retrieval.

[0032] The calculation formula of DTW distance matrix is:

[0033]

[0034] Where D dtw is the DTW distance matrix; X is a time series of length m; Y is a time series of length n; i = 1, 2, ..., m; j = 1, 2, ..., n; x i is the i-th point of the X sequence; y jis the jth point of the Y sequence; min is the minimum function; Rest is the remaining sequence after removing the first point of the sequence;

[0035] Since the STP interlocking data is stored once every 200 milliseconds and the data volume is huge, the efficiency of using the DTW algorithm directly is too low to be accepted. This embodiment uses big data methods such as Hadoop and MapReduce to design a high-performance station shunting workload prediction algorithm. Specifically:

[0036] First, the constructed time series data is set to X = {x1,, x2, x3, ..., xm}, with a length of m, and the time series data to be searched for similarity is set to R = {r1, r2, r3, ..., rn}, with a length of n; in order to take advantage of Hadoop parallel computing to improve efficiency, both X and R are split into several subsequences, X is divided into a subsequences of length m / a, and the subsequences are x0, x1, ..., xa-1; R is divided into b subsequences of length n / b, and the subsequences are r1, r2, ..., rb-1;

[0037] Using Hadoop distributed cache mechanism, each subsequence data is distributed to Hadoop cluster nodes, and a×b submatrices D(p,q) of size [m / a]×[n / b] are constructed, where p=1,2,…,a; q=1,2,…,b;

[0038] Then calculate the DTW distance of D(1,1), store the last row and column of D(1,1) in HDFS (Hadoop Distributed File System); read and calculate the last row and column of D(1,1) from HDFS, and insert them into the 0th row of D(2,1) and the 0th column of D(1,2) respectively for DTW distance calculation; finally, store the last row and column of D(2,1) and D(1,2) in HDFS; then get the last row of D(2,1) and insert it into the 0th row of D(3,1); read the last row of D(1,2) and store it into the 0th row of D(2,2); read the last column of D(2,1) and insert it into the 0th column of D(2,2); read the last column of D(1,2) and insert it into the 0th column of D(1,3); after the above operations are completed, the DTW distances of D(3,1), D(2,2) and D(1,3) are obtained through Hadoop parallel calculation;

[0039] Through the above process, and so on: read the last row of the submatrix D(p-1, q) and the last column of D(p, q-1), insert them into the 0th row and 0th column of D(p, q) respectively, insert the last row of D(p, q) into the 0th row of D(p+1, q), and insert the last column of D(p, q) into the 0th column of D(p, q+1); then use Hadoop to parallelly calculate the DTW distance of each submatrix (one MapReduce process calculates one submatrix);

[0040] During the calculation process, the calculated DTW distance is accumulated and the distance threshold is judged. If the distance threshold is exceeded, the calculation is terminated; otherwise, it is iterated p+q-1 times, and the DTW distance of D(p,q) is calculated and output in parallel;

[0041] The distance threshold is set according to actual needs, and is a threshold used to determine whether two time series data are similar.

[0042] Step 3: Search the scene with the highest similarity to the current data in the historical data based on the obtained DTW distance, and use this as a basis to predict the workload of the station shunting operation.

[0043] In this step, the similarity between the two curves is determined based on the calculated DTW distance between the previous time series data curve and the current time series data curve.

[0044] If it is judged to be similar, the trend of the current time series data curve is predicted based on the time series data curves of previous years, thereby predicting the workload of station shunting operations.

[0045] It is worth noting that the contents not described in detail in the embodiments of the present invention belong to the prior art known to professional and technical personnel in the field.

[0046] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims. The information disclosed in the background technology section of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or in any form that the information constitutes prior art known to those skilled in the art.

Claims

1. A method for predicting the workload of station shunting operations, characterized in that: The method comprises: Step 1: construct time series data for measuring the workload of shunting operations through wireless shunting locomotive signals and monitoring system STP data; wherein the STP data is ground signal interlocking and locomotive operation status data related to shunting operations recorded during the operation of the station STP, specifically including interlocking data, control data, locomotive data, work order data, and tracking data; In step 1, factors related to the workload of shunting operations are obtained through locomotive data and work order data in the STP data, and normalized to calculate the shunting workload influencing factor per unit day; Then, the final value for measuring the workload of shunting operations is obtained by combining the number of occupied throat switches, number of open routes, number of open shunting signals, and number of occupied sections in the interlocking data with the calculated shunting workload influencing factors per unit day. Step 2: Distribute each subsequence data of the constructed time series data to the Hadoop cluster nodes, divide the dynamic time warping DTW distance matrix into multiple sub-matrices, and use parallel computing methods to obtain the DTW distance of the time series data; Step 3: Search the scene with the highest similarity to the current data in the historical data based on the obtained DTW distance, and use this as a basis to predict the workload of the station shunting operation.

2. The method for predicting the workload of station shunting operations according to claim 1 is characterized in that: The process of step 1 is specifically as follows: First, take the day as the unit, and count the mileage, number of uncoupling trains, and number of shunting operation hooks of all shunting operations in the station every day; the mileage is obtained from the locomotive data, and the number of uncoupling trains and the number of shunting operation hooks are obtained from the operation order data. Since the unit ranges of the three types of data are different, they are first normalized separately. The normalization method used is maximum-minimum normalization, and the formula is: Where x′ is the dimensionless value of [0,1] after normalization; x is the original value of the data; minA is the minimum value of the range of x; maxA is the maximum value of the range of x; Then, the normalized values ​​of the three types of data are calculated respectively, and weights are assigned to different data. After that, the values ​​are added together to get the value that measures the workload of the station's shunting operations in one day. The formula is: y=εa+γb+(1-ε-γ)c In the formula, y is the influencing factor of the shunting workload per unit day; ε is the weight coefficient of the shunting kilometers; a is the normalized shunting kilometers of the station; γ is the weight coefficient of the number of detached and trailer trains in the shunting operation; b is the normalized number of detached and trailer trains in the station shunting operation; c is the normalized number of shunting operation hooks; The impact factor y of the shunting workload per day is calculated, and then the number of occupied throat switch groups, the number of open routes, the number of open shunting signal machines, and the number of occupied sections are used to measure the workload of station shunting operations. These four types of data are first normalized, then accumulated according to the weights, and finally the impact factor y is added to obtain the final value for measuring the workload of shunting operations. The formula is: m=η(ωd+αe+θf+(1-ω-α-θ)g)+ρy′ Wherein, m is the final value for measuring the workload of station shunting operations; η is the weight coefficient of interlocking data; ω is the weight coefficient of the occupied number of throat switch groups; d is the normalized occupied number of throat switch groups; α is the weight coefficient of the number of open routes; e is the normalized number of open routes; θ is the weight coefficient of the open number of shunting signals; f is the normalized open number of shunting signals; g is the normalized number of section occupancy; ρ is the weight coefficient of the influencing factor; y′ is the normalized value of the influencing factor y.

3. The method for predicting the workload of station shunting operations according to claim 1 is characterized in that: In step 2, The calculation formula of DTW distance matrix is: Where D dtw is the DTW distance matrix; X is a time series of length m; Y is a time series of length n; i = 1, 2, ..., m; j = 1, 2, ..., n; x i is the i-th point of the X sequence; y j is the jth point of the Y sequence; min is the minimum function; Rest is the remaining sequence after removing the first point of the sequence; First, the constructed time series data is set to X = {x1, x2, x3, ..., xm}, with a length of m, and the time series data to be searched for similarity is set to R = {r1, r2, r3, ..., rn}, with a length of n; both X and R are split into several subsequences, X is divided into a subsequences of length m / a, and the subsequences are x0, x1, ..., xa-1; R is divided into b subsequences of length n / b, and the subsequences are r1, r2, ..., rb-1; Using Hadoop distributed cache mechanism, each subsequence data is distributed to Hadoop cluster nodes, and a×b submatrices D(p,q) of size [m / a]×[n / b] are constructed, where p=1,2,…,a; q=1,2,…,b; Then calculate the DTW distance of D(1,1), store the last row and column of D(1,1) in HDFS (Hadoop Distributed File System); read and calculate the last row and column of D(1,1) from HDFS, and insert them into the 0th row of D(2,1) and the 0th column of D(1,2) respectively for DTW distance calculation; finally, store the last row and column of D(2,1) and D(1,2) in HDFS; then get the last row of D(2,1) and insert it into the 0th row of D(3,1); read the last row of D(1,2) and store it into the 0th row of D(2,2); read the last column of D(2,1) and insert it into the 0th column of D(2,2); read the last column of D(1,2) and insert it into the 0th column of D(1,3); after the above operations are completed, the DTW distances of D(3,1), D(2,2) and D(1,3) are obtained through Hadoop parallel calculation; Through the above process, and so on: read the last row of the submatrix D(p-1, q) and the last column of D(p, q-1), insert them into the 0th row and 0th column of D(p, q) respectively, insert the last row of D(p, q) into the 0th row of D(p+1, q), and insert the last column of D(p, q) into the 0th column of D(p, q+1); then use Hadoop to parallelize the DTW distance of each submatrix; During the calculation process, the calculated DTW distance is accumulated and the distance threshold is judged. If the distance threshold is exceeded, the calculation is terminated; otherwise, it is iterated p+q-1 times, and the DTW distance of D(p,q) is calculated and output in parallel.

4. The method for predicting the workload of station shunting operations according to claim 1, characterized in that: The process of step 3 is specifically as follows: The similarity between the two curves is determined based on the calculated DTW distance between the historical time series data curve and the current time series data curve. If it is determined to be similar, the trend of the current time series data curve is predicted based on the time series data curves of previous years, thereby predicting the workload of the station shunting operation.

Images