An intelligent vehicle flow allocation method for railway marshalling yards

Abstract

The present invention relates to an intelligent flow allocation method for a railway marshalling station, including: for a designated railway station, obtaining the information requirements and marshalling-related data within the designated railway station, constructing a hierarchical model for flow allocation in the marshalling station, iteratively solving the dynamic flow allocation model based on an improved evolutionary algorithm and defined heuristic rules, generating the uncoupling sequence of arriving trains and the marshalling sequence of departing trains, and completing the dynamic flow allocation of trains; according to the uncoupling sequence and the marshalling sequence, optimizing the static flow allocation model based on a hybrid sine-cosine whale optimization algorithm to obtain the specific vehicle flow sources and marshalling contents of each departing train, completing the static flow allocation, and realizing the intelligent flow allocation of the designated railway station. The above method improves the uncoupling and shunting operation efficiency and the fulfillment rate of the flow allocation plan in the railway marshalling station by optimizing the specific marshalling contents of the departing trains, so that the vehicle flow of the departing trains comes from the same arriving train as much as possible.

Description

Technical Field

[0001] The present invention relates to the technology of train operation optimization, and particularly to an intelligent flow distribution method for railway marshalling yards. Background Art

[0002] Railway freight transportation is an important part of China's railway transportation. Marshalling yards are an important part of railway freight transportation, and the flow distribution of freight trains is the core of the work of marshalling yards and also the key technology to improve railway freight capacity.

[0003] Currently, some marshalling yards use the unified national railway station management information system (SMIS) and integrated automation systems, which have improved the informatization level of the preparation of various operation plans in marshalling yards through computer-aided means. However, in terms of operation scheduling, it mainly relies on station dispatchers to manually compile and adjust based on experience. Due to the large scale of the flow distribution problem and the complexity of on-site work, when dispatchers perform flow distribution, facing a large amount of information, they focus on whether the flow distribution plan is feasible and it is difficult to take into account the reasonable allocation of yard resources.

[0004] With the maturity and rapid development of big data, artificial intelligence, and computer network technologies, using efficient intelligent optimization methods to solve traditional flow distribution problems has become the focus of attention. How to combine new intelligent algorithms to quickly and real-time optimize the flow distribution of marshalling yards, improve the preparation speed and quality of plans, and reasonably allocate resources is a problem worthy of further consideration and research. Summary of the Invention

[0005] (I) Technical Problems to be Solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides an intelligent flow distribution method for railway marshalling yards.

[0007] (II) Technical Solutions

[0008] To achieve the above object, the main technical solutions adopted by the present invention include:

[0009] In a first aspect, an embodiment of the present invention provides an intelligent flow distribution method for railway marshalling yards, including:

[0010] S10. For a designated railway station, obtain the information requirements and marshalling-related data within the designated railway station, and construct a hierarchical model for flow distribution in the marshalling yard. The hierarchical model for flow distribution in the marshalling yard includes: a dynamic flow distribution model and a first constraint condition, and a static flow distribution model and a second constraint condition;

[0011] S20. Based on an improved evolutionary algorithm and defined heuristic rules, iteratively solve the dynamic flow distribution model to generate the uncoupling sequence of arriving trains and the marshalling sequence of departing trains, and complete the dynamic flow distribution of trains; the heuristic rules are the rules corresponding to the improved evolutionary algorithm;

[0012] S30. Solve the static vehicle distribution model based on the hybrid sine-cosine whale optimization algorithm according to the decoding order and grouping order, obtain the specific vehicle flow sources and grouping contents of each departing train, complete the static vehicle distribution, and achieve intelligent vehicle distribution at the designated railway station.

[0013] Optionally, the grouping-related data includes: the decoding method of the marshalling station, the number of decoding shunting locomotives, the number of marshalling shunting locomotives, the information of arriving trains and departing trains within the stage plan period;

[0014] The information requirements include: the standard time for arrival technical operations, the standard time for departure technical operations, the standard time for empty running operations, the standard time for coupling operations, the standard time for humping operations, and the standard time for marshalling operations;

[0015] Correspondingly, according to the information of arriving trains and departing trains within the stage plan period, obtain the arrival time and grouping content of arriving trains, the departure time and grouping requirements of departing trains, and the in-stock data of the marshalling yard.

[0016] Optionally, the S10 includes:

[0017] Estimate the humping operation time of the train according to the information requirements to obtain an estimated value of the train humping time.

[0018] Construct the objective functions Z1, Z2, Z3 of the dynamic vehicle distribution model and the objective function Z4 of the static vehicle distribution model in the hierarchical vehicle distribution model of the marshalling station according to the grouping-related data and the estimated value of the train humping time;

[0019] Z1 is to send the largest number of full-axle vehicles to the section via the marshalling station within the stage plan period;

[0020] Z2 is to maximize the sum of the priorities of departing trains to ensure marshalling quality;

[0021] Z3 is to reduce the waiting time for decoding of arriving trains and decode them according to the priorities of arriving trains;

[0022]

[0023]

[0024] where n is the number of arriving trains; m is the number of departing trains; s is the number of vehicle flow directions; s j is a 0-1 variable, is a 0-1 variable, is the number of vehicles with group number k allocated from arriving train d i to departing train f j ; is the grade of the departing train; is the grade of the arriving train; x ii′is a 0-1 variable; n j is the number of the car flow sources of the departing trains, that is, the number of the arriving trains that provide car flows for each departing train.

[0025] Optionally, the first constraint condition in S10 includes:

[0026] Train disintegration time constraint, train disintegration order constraint, train formation time constraint, train connection time constraint, formation content constraint, full axle constraint;

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034] t wb t(j) = t cf (j) - T cf -T bz (j), j = m, m - 1, m - 2 Formula 12;

[0035] t wb t(j) = min{max{t wb (j + 1), t wb (j + 2), t wb (j + 3)} - T bz t(j), t cf (j) - T cf -T bz (j)}, j = 1, 2, …, m - 3 Formula 13;

[0036]

[0037]

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046] The second constraint condition in S10 includes: the full-axle constraint, the train connection time constraint, and the formation content constraint.

[0047] Optionally, S20 includes:

[0048] The initial population generated based on the improved evolutionary algorithm includes the initial formation order population and the initial decomposition order population; the initial formation order population is generated according to the defined heuristic rules; the specific decomposition interval is obtained through the generated initial formation order population; the corresponding decomposition order for each formation order is generated through the decomposition interval; a number of formation orders and corresponding decomposition orders are obtained as the initial population and input into the improved evolutionary algorithm for solution, and the population evolves continuously through three processes: selection, crossover, and mutation; after reaching the preset number of evolutionary times, the optimized decomposition order and formation order are finally output;

[0049] Specifically, obtaining the initial formation order population includes:

[0050] S21. Adopt the method of integer coding, generate a formation order population with only one individual according to the principle of "first departure, first formation" of the train, and calculate the latest formation time t wb (j) and the waiting departure time t df (j), where t cf (j) is the departure time of train f j , T bz (j) is the formation time of the jth departing train, and T cf is the standard departure technical operation time;

[0051] S22. For each individual in the current formation order population, if t df (j) ≥ t df (j + 1), adjust the formation order of the corresponding two trains, and expand the adjusted individual into the current formation order population to obtain an initial formation order population generated according to the heuristic rules;

[0052] In addition, based on the defined decomposition interval and initialized based on the decomposition order, a chromosome in the initial population is obtained to get the initial decomposition order population.

[0053] Optionally, in the population selection step of the improved evolutionary algorithm, the fitness of each individual in the initial population is calculated according to the fitness function, and selection is performed based on the fitness of each individual. The selection operators include:

[0054] According to the Metropolis criterion, at temperature T, the probability of a temperature drop with an energy difference of dE is: where k represents a constant, dE < 0, and T represents temperature;

[0055] In the process of iteratively solving the dynamic traffic assignment model using the improved evolutionary algorithm, the probability of a chromosome being selected and replicated is: where λ is the number of iterations, p is the population size, and z max represents the maximum value of the fitness function of all individuals in the population;

[0056] After the population is selected, in the population crossover step of the improved evolutionary algorithm, the crossover probability calculation formula is:

[0057] where γ ∈ (0, 1) is the crossover probability coefficient, and f max is the ideal optimal solution for traffic assignment, and all trains depart fully loaded and on schedule;

[0058] Two parent chromosomes are randomly generated according to the crossover probability. On the two parent chromosomes, the same gene segments are randomly generated according to a certain rule probability. After crossover, these gene segments will be retained as excellent genes in the offspring chromosomes, and the other gene segments are obtained by excluding the excellent genes from the other parent chromosome in turn;

[0059] In the process of solving the dynamic traffic assignment model, the mutation operation adopts the swap mutation operator. A mutant chromosome is randomly generated according to the mutation probability, and then two gene mutation positions are randomly selected, and the genes at these two positions are swapped, while the other genes remain unchanged.

[0060] Optionally, S30 includes:

[0061] Input the breakup sequence of the arriving trains and the formation sequence of the departing trains into the static traffic assignment model, and use the hybrid sine-cosine whale optimization algorithm to solve the static traffic assignment model to obtain the specific formation content of the departing trains and clarify the connection relationship between the vehicle flows;

[0062] First, set the whale individual vector X = (x(1), x(2),..., x(n)), divide the whale individual vector into two equal-length segments, satisfying n = 2l, where l is the possible traffic flow sources of all departing trains within the stage plan. The first segment vector is used to represent the specific traffic flow sources of each train, including the arriving trains and the directions of the arriving trains; the second segment vector represents the selection order of the traffic flow sources, and set the individual position value range as [-ε, ε], where ε is the total number of departing trains within the stage plan.

[0063] Secondly, encode the first segment vector, and map the whale individual position vector to the remaining number of available trains through formula (6) to complete the process of converting the individual position to the number of trains.

[0064]

[0065] Among them, x(j) is the whale individual vector, s(j) is the number of elements in the set of remaining train numbers, and u(j) is the selection serial number in the set of remaining train numbers.

[0066] Use the maximum position rule to encode the second segment vector, and select the different traffic flow source order vectors representing the same formation train according to the element values. The larger the corresponding element value, the earlier the selection order.

[0067] Finally, based on the decoupling order and formation order output by the distribution model, determine the number of traffic flow sources of the departing trains as the fitness function. By imitating the way of whales hunting, establish a mathematical model for optimization from three aspects: searching, surrounding, and preying on prey, and combine the SCA algorithm to improve the WOA algorithm to obtain the specific traffic flow sources of each departing train and complete the static distribution.

[0068] Optionally, S30 further includes:

[0069] Improve the stage of whales surrounding prey:

[0070] Set a group of random solutions at the beginning and calculate the fitness of the solutions. At the same time, regard the solution with the highest fitness as the target, and the other whale individuals will approach the target. To improve local exploitation, introduce the SCA strategy into the local exploitation process of WOA, and its position update method can be expressed as:

[0071] X(t + 1) = X * (t) - A·D1

[0072]

[0073] A = C·a - a

[0074]

[0075] Among them, X* Let the position of the prey be \(X\), \(t\) be the current iteration number, \(T\) be the maximum iteration number. Two convergence factors \(a\) and \(r1\) are modified and non-linearly decrease from 2 to 0 with the iteration number. \(r2\in[0, 2\pi]\), \(C\in[0, 2]\), \(\theta\in[0, 1]\), all are random numbers, and \(\lambda\), \(\mu\) are constants.

[0076] Improve the prey-catching stage of the bubble net:

[0077] When whales hunt, they will hunt prey along a circular or "9"-shaped path. To enhance the diversity of the entire population and the ability to jump out of local optima, the SCA strategy is also used to improve the spiral update method as follows:

[0078]

[0079] \(D2 = |r3\cdot X * (t)-X(t)|\)

[0080] where \(b\) is a constant defining the spiral shape, \(l\in[0, 1]\), \(\theta\in[0, 1]\), \(r3\in[0, 2\pi]\), all are random numbers. A random number \(p\in[0, 1]\) is set. When \(p\geq0.5\), the bubble net prey-catching is performed; when \(p < 0.5\), the prey is surrounded.

[0081] Optionally, S30 includes:

[0082] Improve the prey-searching stage:

[0083] To further search for prey, WOA uses the change of \(A\) for exploration, that is, when \(|A| < 1\), the algorithm will perform precise search and the whale individual will move towards the current optimal individual. Mix SCA and WOA, and use the optimization method of SCA to improve the global exploration method of WOA by combining the strong global exploration ability of SCA:

[0084] \(X(t + 1)=X rand (t)-A\cdot D3

[0085]

[0086] where \(X rand (t)\) is the position of a randomly selected whale individual.

[0087] In a second aspect, an embodiment of the present invention further provides a computing device, including: a processor, a memory, and a computer program stored on the memory. The processor executes the computer program to implement a railway marshalling station intelligent flow allocation method as described in any one of the first aspects. The processor in this embodiment can execute all the methods in the first aspect above. For the sake of omission, they are not listed one by one here. The above methods are all programs executed by the processor.

[0088] (3) Beneficial effects

[0089] The method of the present invention optimizes the de - grouping sequence, grouping sequence and the origin of the train flow. Starting from the actual operation situation of the marshalling station, it introduces the uncertainty of the de - grouping time. At the same time, a hybrid intelligent algorithm is used to ensure the compilation speed and quality of the flow - allocation plan, enabling the trains within the stage plan to depart fully - loaded and on schedule.

[0090] According to the data of the arriving trains and departing trains at the marshalling station within the stage plan, considering the uncertainty of the train de - grouping time, while optimizing the de - grouping sequence of the arriving trains and the grouping sequence of the departing trains, a reasonable train flow - allocation plan for the marshalling station is obtained, and the specific origin of the train flow for the departing trains is given. Using a hybrid intelligent algorithm (i.e., the hybrid sine - cosine whale optimization algorithm), on the one hand, the solving speed is improved. On the other hand, this hybrid intelligent algorithm comprehensively considers various transport capacity resources within the marshalling station, realizes the rational utilization of resources, ensures that as many departing trains as possible within the stage plan are fully - loaded, on schedule and do not violate the compilation rules, and makes the train flow of the departing trains come from the same arriving train as much as possible, improving the shunting operation efficiency of the railway marshalling station and the fulfillment rate of the flow - allocation plan. Brief description of the drawings

[0091] Figure 1 It is the flowchart of the intelligent train flow - allocation method provided by the embodiment of the present invention;

[0092] Figure 2 It is the improved evolutionary algorithm diagram provided by the embodiment of the present invention;

[0093] Figure 3 It is the schematic diagram of the conversion of the train flow distribution vector provided by the embodiment of the present invention;

[0094] Figure 4 It is the SCWOA flowchart provided by the embodiment of the present invention;

[0095] Figure 5 It is the comparison diagram of the average fitness algorithm of the dynamic flow - allocation population;

[0096] Figure 6 It is the comparison diagram of the historical optimal solution algorithm of the dynamic flow - allocation;

[0097] Figure 7 It is the comparison diagram of the historical optimal solution algorithm of the static flow - allocation;

[0098] Figure 8 It is the schematic diagram of the flow - allocation result of the departing trains. Detailed implementation manners

[0099] In order to better explain the present invention for easy understanding, the present invention will be described in detail below in conjunction with the drawings through specific implementation manners.

[0100] The vehicle flow distribution in a marshalling station belongs to a combinatorial optimization problem. The solution space of such problems grows explosively as the number of variables increases, and it is difficult to obtain the optimal solution using traditional operational research enumeration methods. In recent years, the accuracy of the model and the solution efficiency have been improved to a certain extent by using intelligent computing, but there is still much room for improvement in the solution performance of a single algorithm. At the same time, the train operation data of most current vehicle flow distribution models is fixed and lacks flexibility.

[0101] Starting from the whole process of train operation, the present invention constructs a hierarchical vehicle flow distribution model considering the uncertainty of the sorting time, and designs a hybrid intelligent optimization algorithm (i.e., a hybrid sine-cosine whale optimization algorithm) for solving, which improves the solution performance and solution efficiency compared with a single algorithm. The improved evolutionary algorithm and the hybrid sine-cosine whale algorithm are used to solve the dynamic vehicle flow distribution model and the static vehicle flow distribution model respectively, and the sorting order and formation order of trains and the origin of vehicle flows are optimized to ensure the solution speed while taking into account the formation quality.

[0102] Embodiment 1

[0103] This embodiment provides an intelligent vehicle flow distribution method for a railway marshalling station, and its execution subject can be any computing device, such as Figure 1 As shown, the method of this embodiment may include the following steps:

[0104] S10. For a specified railway station, obtain the information requirements and formation-related data within the specified railway station, and construct a hierarchical vehicle flow distribution model for the marshalling station. The hierarchical vehicle flow distribution model for the marshalling station includes: a dynamic vehicle flow distribution model and a first constraint condition, and a static vehicle flow distribution model and a second constraint condition.

[0105] The formation-related data includes: the sorting method of the marshalling station, the number of sorting locomotives, the number of formation locomotives, the information of arriving trains and departing trains within the period of the stage plan.

[0106] The information requirements include: the standard time for arrival technical operations, the standard time for departure technical operations, the standard time for empty running operations, the standard time for coupling operations, the standard time for humping operations, the standard time for formation operations.

[0107] Correspondingly, according to the information of arriving trains and departing trains within the period of the stage plan, obtain the arrival time and formation content of arriving trains, the departure time and formation requirements of departing trains, and the in-station car data of the marshalling yard.

[0108] S20. Based on the improved evolutionary algorithm and the defined heuristic rules, iteratively solve the dynamic vehicle flow distribution model to generate the sorting order of arriving trains and the formation order of departing trains, and complete the dynamic vehicle flow distribution of trains; the heuristic rules are the rules corresponding to the improved evolutionary algorithm.

[0109] That is, the initial population generated based on the improved evolutionary algorithm includes the initial formation sequence population and the initial decomposition sequence population; the initial formation sequence population is generated according to the defined heuristic rules; the specific disintegration intervals are obtained through the generated initial formation sequence population; the corresponding decomposition sequences for each formation sequence are generated through the disintegration intervals; several formation sequences and the corresponding decomposition sequences are obtained as the initial population and input into the improved evolutionary algorithm for solution, and the population continuously evolves through the three processes of selection, crossover, and mutation; after reaching the preset number of evolutionary times, the optimized decomposition sequence and formation sequence are finally output.

[0110] S30. According to the decomposition sequence and the formation sequence, solve the static vehicle flow distribution model based on the hybrid sine-cosine whale optimization algorithm, obtain the specific vehicle flow sources and formation contents of each departing train, complete the static vehicle flow distribution, and realize the intelligent vehicle flow distribution of the designated railway station.

[0111] The method of this embodiment optimizes the decomposition sequence, the formation sequence, and the vehicle flow source. Starting from the actual operation situation of the marshalling station, the uncertainty of the disintegration time is introduced; at the same time, the hybrid intelligent algorithm is used to ensure the compilation speed and quality of the vehicle flow distribution plan, so that the trains within the stage plan are fully loaded and depart on time.

[0112] Embodiment 2

[0113] The following will Figures 2 to 8 describe the method of this embodiment in detail.

[0114] The static data of the marshalling station includes: the marshalling and disintegration method of the marshalling station, the number of disintegration shunting locomotives, the standard time of various technical operations in the station, etc.

[0115] The data of the arriving trains includes: the arrival time and the formation content.

[0116] The data of the departing trains includes: the departure time and the formation requirements.

[0117] Data such as the marshalling and disintegration method of the marshalling station, the number of disintegration shunting locomotives, the standard time of various technical operations in the station, the arrival time and the formation content of the arriving trains, and the departure time and the formation requirements of the departing trains are obtained in advance.

[0118] For example, according to the "Rules for Station Train Operation" of Shijiazhuang South Marshalling Station, obtain the marshalling and disintegration method of the marshalling station, the number of disintegration shunting locomotives, the number of formation shunting locomotives, the standard time of arrival technical operations, the standard time of departure technical operations, the standard time of empty running operations, the standard time of coupling operations, the standard time of humping operations, and the standard time of formation operations.

[0119] According to the information of the arriving trains and the departing trains within the stage plan period, obtain the arrival time and the formation content of the arriving trains, the departure time and the formation requirements of the departing trains, and the in-stock data of the marshalling yard.

[0120] Number the arriving trains according to their arrival times, and define the set of arriving trains D =

[0121] [d0, d1, d2…d n , where d0 is the in-stock cars in the classification yard. Define the set of the number of arriving trains D' = [0, 1, 2,..., n], and n is the total number of arriving trains within the stage plan. Similarly, number the departing trains according to their departure times, and define the set of departing trains F = [x1, f2, f3…f m , define the set of the number of departing trains, F' = [1, 2, 3,..., m], and m is the total number of departing trains within the stage plan. Define the set of car flow directions V = [v1, v2, v3,..., v s , define the set of the number of car flow directions V' = [1, 2, 3,..., s], D = [d0, d1, d2…d n and s is the total number of car flow directions.

[0122] Step 1: Considering the uncertainty of train classification and marshalling time, construct a hierarchical model for car distribution in the marshalling station based on the information requirements within the designated railway station and the data related to marshalling. The hierarchical model for car distribution includes a dynamic car distribution model and the first constraint condition, and a static car distribution model and the second constraint condition.

[0123] For example, the classification and marshalling time of a train can be estimated using the formula of the hierarchical model for car distribution. The classification and marshalling operation of a train consists of four processes: empty running, coupling, humping, and releasing. Among them, the first three are regarded as constants. Estimate the releasing time of the train, and then add them together to obtain the classification and marshalling time of the train.

[0124] Use an accurate estimation method to solve the classification and marshalling time of each train, making the planned operation time closer to the actual operation time, thereby improving the accuracy and controllability of the plan.

[0125] In the classification and marshalling operation of the marshalling station, the time of empty running, coupling, and humping at the same station generally varies little and can be regarded as constants. Since the releasing time varies greatly and has the greatest impact on the disintegration time, the estimation formula for the releasing operation time of the train in this embodiment is as follows:

[0126]

[0127] Among them, t lf (i) represents the estimated value of the train releasing time, and the subscript lf represents releasing, are the number of disintegration hooks of the non-forbidden releasing car group and the forbidden releasing car group of the train respectively. The parameters α1, α2, and α3 are dynamic parameters related to the current station.

[0128] Step 1-1: Obtain the objective function of the dynamic car distribution model.

[0129] First: To enable as many train formations to be successful as possible, the objective function Z1 is designed to maximize the number of fully-loaded vehicles sent from the marshalling station to the section within the stage time. At the same time, the objective function Z2 maximizes the sum of the priorities of the departing trains to ensure the marshalling quality. Finally, the objective function Z3 is to reduce the waiting time for the arriving trains to be unmarshalled and unmarshal them according to the priorities of the arriving trains as much as possible.

[0130]

[0131]

[0132]

[0133] Among them, n is the number of arriving trains; m is the number of departing trains; s is the number of vehicle flow directions; s j is a 0-1 variable. If the departing train f j can depart with a full load, the value is 1, otherwise it is 0; is a 0-1 variable. k represents the train direction, i represents the arriving train, and j represents the departing train;

[0134] If the car body of the arriving train d i in the k direction can be used as the vehicle flow source of the departing train f j the value is 1, otherwise it is 0; is the number of vehicles with group number k allocated from the arriving train d i to the departing train f j ; is the grade of the departing train; is the grade of the arriving train; x ii′ is a 0-1 variable. i' represents the arriving train. If the arriving train d i is unmarshalled earlier than the arriving train d i′ the value is 1, otherwise it is 0.

[0135] Step 1-2: The constraint conditions of the dynamic traffic assignment model (i.e., the first constraint conditions) are satisfied :

[0136] ① Train disintegration time constraint: When the train arrives at the yard, it needs to complete the arrival technical operations before it can be unmarshalled. Therefore, the earliest start time of train disintegration can be expressed as:

[0137]

[0138] The actual disintegration time should be greater than or equal to the earliest disintegration time. Therefore,

[0139]

[0140] In the double-push single-slide disintegration mode, the disintegration end time constraint is

[0141]

[0142] where: t zj (i) is the earliest start time for disintegrating train d i ; t dd (i) is the arrival time of the i-th arriving train; T dd is the standard operation time for arrival technical operations; t sj (i) is the actual start time for disintegrating train d i ; t jj (i) is the arrival train d i disintegration end time; T jt (i) is the standard operation time for other train formation and decoupling operations.

[0143] ② Train disintegration sequence constraint

[0144]

[0145]

[0146] Considering that the disintegration mode is double-push single-slide, the disintegration end sequence is constrained

[0147]

[0148]

[0149] where, t zj (i) is the earliest start time for disintegrating train d i ; t sj (i) is the arrival train d i actual disintegration start time; t jj (i) is the arrival train d i disintegration end time; t lf (i) represents the estimated value of the train humping time; x ii′ is a 0-1 variable. If arrival train d i is earlier than arrival train d i′ for formation and decoupling, the value is 1, otherwise 0.

[0150] ③ Train formation time constraint: Assuming there are three formation shunting locomotives at the end of the hump, the latest formation time is: t wb (j) = t cf (j) - T cf - T bz (j), j = m, m - 1, m - 2 Formula 12;

[0151] t wb (j) = min{max{t wb (j + 1), t wb (j + 2), twb (j + 3)} - T bz (j), t cf (j) - T cf -T bz (j)}, j = 1, 2, …, m - 3 Formula 13;

[0152] The train's waiting - to - depart time is:

[0153] In actual operation, the actual formation time of the train should be less than or equal to the latest formation time of the train:

[0154]

[0155] where t wb (j) is the latest formation time of the train; t cf (j) is the departure time of the train; T cf is the departure technical operation time; t df (j) is the waiting - to - depart time of the train; t sb (j) is the actual start formation time of the train.

[0156] ④ Train connection time constraint: The arriving train can only provide vehicle flow for the departing train formation if it is disintegrated before the departing train formation starts:

[0157]

[0158]

[0159] where l i,j is a 0 - 1 variable. If the arriving train d i and f j meet the vehicle flow connection condition in terms of time, the value is 1, otherwise it is 0.

[0160] ⑤ Formation content constraint: The departing train can only select the vehicle flow of the specified direction / car type in the train formation plan:

[0161]

[0162]

[0163]

[0164] where is the number of vehicles with group number k included in the arriving train d i ; is a 0 - 1 variable. If the formation content of the departing train f j includes vehicles with group number k, the value is 1, otherwise it is 0.

[0165] ⑥Full - axle constraint: A full - axle railway freight train means that it meets one of the maximum values of the axle load or the conversion length of the train. The axle load of the train is determined by the locomotive traction force and the slope within the section, while the conversion length is determined by the effective length of the arrival and departure tracks at the junction stations within the section and the braking distance of the train, etc. Both the axle load and the conversion length of the train are related to the number of cars in the train formation. In this paper, the full - axle constraint is relaxed to not exceed the maximum number of cars in the train formation, that is:

[0166]

[0167]

[0168]

[0169] Among them, is the number of cars with group number k allocated from arrival train d i to departure train f j ; M j is the number of cars required for departure train f j to be full - axle; s j is a 0 - 1 variable. If departure train f j can depart with a full - axle, the value is 1, otherwise 0.

[0170] Step 1 - 3: Obtain The objective function of the static traffic assignment model is :

[0171]

[0172] In the formula, n j is the number of traffic flow sources of the departure train, which is equal to the number of arrival trains providing traffic flow for each departure train.

[0173] Step 1 - 4: The constraint conditions of the static traffic assignment model (i.e., the first constraint conditions) are satisfied :

[0174] ① - 1 Full - axle constraint (corresponding to the above - mentioned full - axle constraint):

[0175]

[0176]

[0177]

[0178] Among them, is the number of cars with group number k allocated from arrival train d i to departure train f j ; M j is the number of cars required for departure train f j to be full - axle; s j is a 0 - 1 variable. If departure train fj If it is possible to depart with a full train, the value is 1; otherwise, it is 0.

[0179] ②-2 Train connection time constraint (corresponding to the above train connection time constraint)

[0180]

[0181]

[0182] Among them, l i,j is a 0-1 variable. If the arriving train d i and f j meet the vehicle flow connection condition in terms of time, the value is 1; otherwise, it is 0.

[0183] ①-3 Formation content constraint:

[0184]

[0185]

[0186]

[0187] Among them, is the number of vehicles with group number k included in the arriving train d i ; is a 0-1 variable. If the formation content of the departing train f j includes vehicles with group number k, the value is 1; otherwise, it is 0.

[0188] Step 2: Generate an initial population using the improved evolutionary algorithm, as Figure 2 shown. Use the improved evolutionary algorithm to iteratively solve the dynamic vehicle allocation model to obtain the optimized uncoupling sequence of the arriving trains and the formation sequence of the departing trains.

[0189] In this embodiment, it is improved on the basis of the evolutionary algorithm to obtain an improved evolutionary algorithm. The initial population described below is the initial population generated for the uncoupling sequence of the arriving trains and the formation sequence of the departing trains. The initial population of this embodiment includes an initial population of uncoupling sequences and an initial population of formation sequences.

[0190] That is to say, the initial population is generated according to the rules in the following text. First, generate the initial population of formation sequences, and then, based on the initial population of formation sequences, generate the corresponding initial population of uncoupling sequences. In this way, each individual includes a set of uncoupling sequences and the corresponding formation sequences. Finally, these initial individuals together are called the initial population. Input the initial population into the improved evolutionary algorithm and perform iterations according to the algorithm process. The final output result is the optimized uncoupling sequence and formation sequence.

[0191] The solution space of the flow distribution problem is all possible disintegration orders and formation orders, which means that the solution to the problem needs to include the numbers of all arriving trains and departing trains, and each number exists and only exists once. Therefore, it is decided to adopt the integer coding method, and the individual thus formed is C = [J, B], where J is the train disintegration order and B is the train formation order.

[0192] In each individual of the initial population, the disintegration order and the formation order are randomly generated, which is obviously unreasonable. Random generation is very likely to produce illegal individuals. In this embodiment, a heuristic rule for population initialization is designed to increase the proportion of individuals with high fitness in the generated initial population.

[0193] First, generate the formation order population. The specific steps are as follows:

[0194] First: Generate a formation order population with only one individual according to the principle that the trains depart first and are formed first, and calculate the latest formation time t wb (j) and the waiting departure time t df (j), where t cf (j) is the departure time of train f j , T bz (j) is the formation time of the j-th departing train, and T cf is the standard of departure technical operation time.

[0195] Second: For each individual in the current formation order population, if t df (j) ≥ t df (j + 1), adjust the formation order of the corresponding two trains, and expand the adjusted individual into the current formation order population. Finally, a formation order population generated according to the heuristic rule is obtained.

[0196] Of course, if the population size generated according to the heuristic rule is less than the required size, randomly select individuals from the generated population for supplementation.

[0197] For the method of generating the initial population of the disintegration order, this embodiment introduces the concept of the disintegration interval and uses the latest formation time of the departing train as the division standard. Define t zj (i) as the earliest start disintegration time of the arriving train d i , T jt (i) as the disintegration time of the i-th arriving train. If the earliest disintegration end time of the arriving train is less than its latest start formation time, that is, t zj (i) + T jt (i) < t wb (j), the car flow connection constraint is satisfied. The set of arriving trains that satisfy the constraint is called the disintegration interval KR j .

[0198] But KRj The elements in j-1 may be selected by the previous disintegration interval KR j (i.e., it indicates the disintegration interval KR corresponding to each departure train at this time j The elements in the set are known), and the elements in KR j that are not selected by the previous disintegration interval are combined into a new set, called the relevant arrival train R j . Since the formation order population has been determined at this time, the disintegration interval KR of each departure train

[0199] Disassembly sequence initialization method is selected sequentially starting from the first disintegration interval. When the relevant arrival train R j corresponding to this interval is an empty set, it turns to the next disintegration interval to continue the selection until all arrival trains are selected, forming a chromosome in the initial population. The initial population generated based on this rule can reduce the adverse sorting of genes in the chromosome and accelerate the convergence speed of the improved evolutionary algorithm.

[0200] Set the fitness function as:

[0201]

[0202] where l i,j is a 0-1 variable. If the arrival train d i and the departure train f j meet the vehicle flow connection condition in terms of time, the value is 1, otherwise it is 0; is the number of vehicles with group number k assigned to the arrival train d i and assigned to the departure train f j ; s j is a 0-1 variable. If the departure train d j departs on time with a full load, it is 1, otherwise it is 0; is the departure train level; is the arrival train level; x ii′ is a 0-1 variable. If the arrival train d i is earlier than the arrival train d i′ to be uncoupled and sorted, the value is 1, otherwise it is 0; γ1 and γ2 are constants.

[0203] The algorithm of the improved evolutionary algorithm is used after generating the initial population. Calculate the fitness of each individual in the initial population according to the fitness function, and select the individuals in the population according to the fitness function.

[0204] After the improved evolutionary algorithm generates the initial population, the population continuously evolves through three processes: selection, crossover, and mutation. After reaching the specified number of evolutionary iterations, the final evolved result, i.e., the disassembly order and the grouping order, is obtained. In the selection process, the fitness of each individual in the initial population is calculated according to the fitness function, and selection is performed based on the fitness of each individual. The design of the selection operator is as follows. Combining the roulette wheel method and the idea of simulated annealing, a heuristic random search process combined with the Metropolis solution method is applied to the design of the selection operator. According to the Metropolis criterion, at temperature T, the probability of a temperature drop with an energy difference of dE is:

[0205]

[0206] where k represents a constant, dE < 0, and T represents the temperature. Since the temperature is gradually decreasing, P(dE) also gradually decreases, which indicates that as the temperature decreases, the particles tend to be stable, and the acceptance of inferior values gradually decreases.

[0207] Applying this idea to the selection operator, the probability that a chromosome is selected and replicated during the iterative solution of the dynamic traffic assignment model by the evolutionary algorithm is:

[0208]

[0209] where λ is the number of iterations, p is the population size, and z max represents the maximum value of the fitness function of all individuals in the population.

[0210] After the population completes selection, the crossover process needs to be carried out. Next, the crossover operator is designed to improve the population diversity. The calculation formula for the crossover probability is:

[0211]

[0212] where γ ∈ (0, 1) is the crossover probability coefficient, and f max is the ideal optimal solution for traffic assignment, and all trains depart fully loaded and on schedule.

[0213] To avoid generating invalid individuals after using the crossover operator, including introducing duplicate genes and invalid genes, and to retain excellent genes as much as possible, the crossover operator adopts the order crossover method, i.e., Order Crossover (OX). Two parent chromosomes are randomly generated according to the crossover probability. On the two parent chromosomes, a gene segment is randomly generated according to a certain rule probability. This gene segment will be retained as an excellent gene in the offspring chromosome after crossover, and the other gene segments are sequentially obtained from the other parent chromosome after excluding the excellent genes.

[0214] The variation operation in a chromosome refers to changing the value of a certain gene or certain genes in the chromosome. Gene variation can accelerate the local search ability for the optimal solution and increase the diversity of the population at the same time. In dynamic vehicle flow allocation, the variation operation adopts the exchange mutation operator, randomly generates mutant chromosomes according to the mutation probability, then randomly selects two gene mutation positions, swaps the genes at these two positions, and keeps other genes unchanged. After the variation step, according to the set mutation probability, an individual in the population may change its decoding order through the operations in the variation. What is obtained is the population after the variation operation. The relationship with the previous one is that the population is continuously iterated through operations such as selection, crossover, and variation.

[0215] In this embodiment, in order to improve the quality of the initial population, heuristic rules are designed, relevant train theories are proposed, by constructing a set of relevant trains, some invalid solutions are removed, the competitiveness of the initial individuals is improved, dynamic vehicle flow allocation is performed on the marshalling station, and the optimized decoding order and marshalling order are generated.

[0216] Based on the solution results of the dynamic model, that is, the decoding order of the arriving trains and the marshalling order of the departing trains.

[0217] Step 4: After determining the decoding order of the arriving trains and the marshalling order of the departing trains (i.e., determining the order), use the hybrid sine-cosine whale optimization algorithm for static vehicle flow allocation. For the content and selection order of train resource selection, design the encoding and decoding methods, and solve the specific vehicle flow sources and marshalling contents of the departing trains in the population.

[0218] A hybrid intelligent optimization algorithm can be optionally used to solve the specific marshalling content. For example, input the results of the decoding order of the arriving trains and the marshalling order of the departing trains into the static vehicle flow allocation model, and use a hybrid intelligent optimization algorithm to solve the static vehicle flow allocation model, and finally obtain the specific marshalling content of the departing trains, and clarify the connection relationship between vehicle flows.

[0219] This hybrid intelligent optimization algorithm combines two intelligent optimization algorithms, namely the whale algorithm and the sine-cosine algorithm.

[0220] The design of the whale optimization algorithm (WOA) is inspired by the foraging behavior of a group of whales. In a group of whales, its foraging behavior is often accompanied by cooperation among individuals. The most special thing in the foraging behavior of a group of whales is the bubble net foraging behavior, that is, after a group of whales discovers prey, they will gradually surround the prey in a spiral path, the activity range of the prey gradually shrinks, and finally the prey is captured. The whale optimization algorithm studies from the unique foraging behavior of a group of whales and simulates the optimization process of the algorithm.

[0221] The main idea of the sine cosine algorithm (SCA) is to comprehensively use the sine and cosine functions for exploration and development during the algorithm optimization process. This position update method enables the algorithm to explore a larger area during exploration and perform precise searches within a smaller range during local development.

[0222] This hybrid intelligent optimization algorithm comprehensively utilizes the position update method of SCA and applies it to the position update and spiral update of WOA to improve the overall optimization level of WOA. This algorithm combines the WOA algorithm and the SCA algorithm into a new algorithm called the sine cosine whale optimization algorithm (SCWOA).

[0223] First, set the whale individual vector X = (x(1), x(2),..., x(n)). Divide the whale individual vector into two equal-length segments, satisfying n = 2l, where l is the possible traffic flow sources of all departing trains within the stage plan. The first segment vector is used to represent the specific traffic flow sources of each train, including the arriving trains and the directions of the arriving trains. The second segment vector represents the selection order of the traffic flow sources, and set the individual position value range as [-ε, ε], where ε is the total number of departing trains within the stage plan.

[0224] Encode the first-segment traffic flow source vector. Map the whale individual position vector to the remaining number of available trains through the following formula, thereby completing the process of converting the individual position to the number of trains.

[0225]

[0226] Among them, x(j) is the whale individual vector, s(j) is the number of elements in the set of remaining train numbers, and u(j) is the selection serial number in the set of remaining train numbers. Figure 3 The conversion process assigned to the first-segment individual vector. As Figure 3 shown, O 11 represents the first traffic flow source of the first train for marshalling operations, O 12 represents the second traffic flow source of the first train for marshalling operations. Each traffic flow source has a corresponding set of remaining train numbers, which indicates how many trains can remain for the corresponding traffic flow source after the flow distribution is completed. The selection serial number of the traffic flow source in the figure corresponds to the list serial number of the set of remaining train numbers below. The selection serial number of O 11 is 1, that is, select the first element 0 in the set of remaining train numbers, indicating that after the first train completes marshalling, the remaining number of trains for this traffic flow source is 0.

[0227] The encoding of the selection sequence vector is implemented using the largest position value (LPV) rule. Each sequence vector also corresponds to a car flow source. The sequence vectors representing different car flow sources of the same formation train are selected according to the magnitude of the element values. The larger the corresponding element value, the earlier the selection order.

[0228] Since the upper-layer algorithm has solved the optimized decoupling sequence and formation sequence, in order to minimize the number of shunting operation hooks for disintegration and formation, the departure trains should try to use the car flows of the same arriving train as much as possible. Based on this, the number of car flow sources of the departure train is designed as the fitness function, that is

[0229]

[0230] where n j represents the number of car flow sources of the departure train f j , γ3 is a constant. s j is the number of departure trains with successfully allocated car flows, γ3 is the penalty coefficient, which is a constant.

[0231] In the optimization process of any swarm intelligence optimization algorithm, there are two stages: global exploration and local exploitation. Appropriately balancing the two stages can enable the algorithm to explore a larger range in the initial optimization, and at the same time improve its search accuracy for the local range in the later stage, and improve the overall accuracy of the algorithm solution. Whether it is WOA or SCA, the convergence factor in the iterative process changes linearly, and this linear change cannot fully exert the exploration and exploitation ability of the algorithm. Therefore, in order to adapt to the algorithm optimization process, a non-linear convergence factor adjustment strategy is proposed to adjust the convergence factor in the algorithm. At the same time, this algorithm establishes a mathematical model for optimization from three aspects: searching, surrounding, and preying on prey by imitating the way of whale groups preying, and improves the WOA algorithm by combining with the SCA algorithm.

[0232] (1) Improve the stage of whales surrounding prey

[0233] At the beginning of the algorithm, a set of random solutions are set and the fitness of the solutions is calculated. At the same time, the solution with the highest fitness is regarded as the target, and the other whale individuals will approach the target. To improve the local exploitation ability of the algorithm, the SCA strategy is introduced into the local exploitation process of WOA, and its position update method can be expressed as:

[0234] X(t + 1) = X * (t) - A·D1 Formula 39;

[0235]

[0236] A = C·a - a Formula 41;

[0237]

[0238] Among them, X * is the prey position, t is the current iteration number, T is the maximum iteration number, two convergence factors a and r1 are modified and nonlinearly decrease from 2 to 0 with the iteration number, r2 ∈ [0, 2π], C ∈ [0, 2], θ ∈ [0, 1], all are random numbers, and λ, μ are constants.

[0239] (2) Improved bubble net predation stage

[0240] When the whale group preys, it will prey on the prey along a circular or "9"-shaped path. Although this logarithmic spiral trajectory can accelerate the overall convergence speed, it will cause the individuals in the group to aggregate together at a relatively fast speed, thereby causing the entire population to lose diversity and ultimately increasing the probability of falling into a local optimum. Therefore, to enhance the diversity of the entire population and the ability to jump out of the local optimum, the SCA strategy is also used to improve the spiral update method as follows:

[0241]

[0242] D2 = |r3·X * (t) - X(t)| Formula 44;

[0243] Among them, b is a constant defining the spiral shape, l ∈ [0, 1], θ ∈ [0, 1], r3 ∈ [0, 2π], all are random numbers. SCA avoids the rapid loss of population diversity by using the random occurrence of positive spiral update movement and cosine spiral update movement, enabling it to smoothly transition from global exploration to local exploitation and finally iteratively converge to the optimal solution. WOA simulates the entire whale predation path through these two mechanisms. To simulate two behaviors occurring simultaneously, a random number p ∈ [0, 1] is set. When p ≥ 0.5, bubble net predation is carried out, and when p < 0.5, the prey is surrounded.

[0244] (3) Improved prey searching stage

[0245] To further search for prey, WOA uses the change of A for exploration, that is, when |A| < 1, the algorithm will perform precise search and the whale individual will move towards the current optimal individual. Different from the exploitation stage, when |A| ≥ 1, the whale individual will move away from the current prey and update its position according to a random whale individual. Mix SCA and WOA, use the optimization method of SCA, combine the strong global exploration ability of SCA, and improve the global exploration method of WOA. Using SCA with a control parameter decreasing exponentially has higher computational accuracy than the original algorithm and improves the solution efficiency of the algorithm. Therefore, combined with the nonlinear convergence factor strategy, the original formula is improved as follows:

[0246] X(t + 1) = Xrand (t)-A·D3 formula 45;

[0247]

[0248] where X rand( t) is the position of a randomly selected whale individual.

[0249] The algorithm flowchart of SCWOA is as Figure 4 shown.

[0250] Through this solution, by using the improved evolutionary algorithm to solve the dynamic vehicle allocation model, the decomposition sequence of the arriving trains and the formation sequence of the departing trains can be obtained, and the dynamic vehicle allocation of the trains can be completed. Then, these two groups of sequences are input into the static vehicle allocation model, and the hybrid sine-cosine whale optimization algorithm is used for solution to obtain the specific vehicle flow sources of each departing train, and the static vehicle allocation is completed.

[0251] Experimental example

[0252] For the simulation of the actual operation scenario, the comparison results of different algorithms under dynamic vehicle allocation are as Figure 5 and Figure 6 , in this invention, the improved evolutionary algorithm is used to solve the model, and the solution quality is higher under the same number of iterations. The comparison results of different algorithms under static vehicle allocation are shown in Appendix Figure 7 , and the specific vehicle allocation results are shown in Figure 8 . Through the result comparison, it can be seen that on the one hand, the hybrid sine-cosine whale algorithm uses a non-linear strategy to process the convergence factor, and on the other hand, it combines the sine-cosine optimization algorithm and the whale optimization algorithm, improves the local search ability, and reduces the possibility of the algorithm "prematurely converging" and falling into the local optimal solution. By optimizing the specific formation content of the departing trains, the vehicle flow of the departing trains is made to come from the same arriving train as much as possible, which improves the decomposition and shunting operation efficiency of the railway marshalling station and the fulfillment rate of the vehicle allocation plan.

[0253] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0254] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions.

[0255] It should be noted that in the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in the claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention can be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a claim listing several means, several of these means can be embodied by the same hardware. The use of the terms first, second, third, etc. is for convenience only and does not denote any order. These terms can be construed as part of the name of the element.

[0256] In addition, it should be noted that in the description of this specification, the descriptions of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without conflict, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0257] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications after learning the basic creative concept. Therefore, the claims should be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0258] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention should also include these modifications and variations.

Claims

1. An intelligent car distribution method for a railway marshalling station, characterized in that, Including: S10. For a designated railway station, obtain the information requirements and formation-related data within the designated railway station, and construct a hierarchical model for marshalling yard car distribution. The hierarchical model for marshalling yard car distribution includes: a dynamic car distribution model and a first set of constraints, and a static car distribution model and a second set of constraints; S20. Based on an improved evolutionary algorithm and defined heuristic rules, iteratively solve the dynamic car distribution model to generate the uncoupling sequence of arriving trains and the formation sequence of departing trains, and complete the dynamic car distribution of trains; the heuristic rules are the rules corresponding to the improved evolutionary algorithm; S30. According to the uncoupling sequence and the formation sequence, solve the static car distribution model based on the hybrid sine-cosine whale optimization algorithm to obtain the specific car flow sources and formation contents of each departing train, complete the static car distribution, and achieve intelligent car distribution for the designated railway station; The formation-related data includes: marshalling yard uncoupling method, number of uncoupling shunting locomotives, number of formation shunting locomotives, information on arriving trains and departing trains during the stage plan period; The information requirements include: standard time for arrival technical operations, standard time for departure technical operations, standard time for empty journey operations, standard time for coupling operations, standard time for humping operations, standard time for formation operations; Correspondingly, according to the information on arriving trains and departing trains during the stage plan period, obtain the arrival time and formation content of arriving trains, the departure time and formation requirements of departing trains, and the in-stock data of the marshalling yard; S10 includes: estimating the train humping operation time according to the information requirements to obtain an estimated value of the train humping time, Constructing the objective functions Z1, Z2, Z3 of the dynamic car distribution model and the objective function Z4 of the static car distribution model in the hierarchical model for marshalling yard car distribution according to the formation-related data and the estimated value of the train humping time; Z1 is to maximize the number of fully loaded cars sent from the marshalling yard to the section during the stage plan period; Z2 is to maximize the sum of the priorities of departing trains to ensure formation quality; Z3 is to reduce the waiting time for uncoupling of arriving trains and uncouple them according to the priorities of arriving trains; where n is the number of arriving trains; m is the number of departing trains; s is the number of vehicle flow directions; s j is a 0-1 variable, is a 0-1 variable, is the number of vehicles with group number k allocated from the arriving train d i to the departing train f j ; is the level of the departing train; is the level of the arriving train; x ii′ is a 0-1 variable; n j is the number of vehicle flow sources of the departing train, that is, the number of arriving trains providing vehicle flow for each departing train; The first set of constraints in S10 includes: train disintegration time constraints including formulas 5 to 7; train disintegration sequence constraints including formulas 8 to 11, train formation time constraints including formulas 12 to 15, train connection time constraints including formulas 16 and 17, formation content constraints including formulas 18 to 20, full-load constraints including formulas 21 to 23; t zj (i) arrival train d i earliest starting disintegration time; t dd (i) arrival time of the i-th arrival train; T dd is the arrival technical operation time standard; t sj (i) arrival train d i actual starting disintegration time t jj (i) is the arrival train d i The end time of disintegration; T jt (i) is the standard operation time for other train formation and decomposition operations; t lf (i) represents an estimated value of the rolling time of the train t wb (j) = t cf (j) - T cf -T bz (j), j = m, m - 1, m - 2 Formula 12; t wb (j) = min{max{t wb (j + 1), t wb (j + 2), t wb (j + 3)} - T bz (j), t cf (j) - T cf -R bz (j)}, j = 1, 2, …, m - 3; Formula 13 t wb (j) is the latest formation time of the train; t cf (j) is the train f j departure time; T cf is the standard departure technical operation time; T bz (j) is the formation time of the j-th departing train; t df (j) is the train waiting departure time; t sb (j) is the actual train formation start time; l i,j is a 0-1 variable; For the arriving train d i contains the number of vehicles with group number k; is a 0-1 variable; For arriving train d i The number of cars with group number k allocated to departing train f j ; M j For the departure train f j The number of vehicles required for full axle load; The second set of constraints in S10 includes: the full-load constraints, the train connection time constraints, and the formation content constraints; The S20 includes: The initial population generated based on the improved evolutionary algorithm includes the initial formation order population and the initial decomposition order population; generating the initial formation order population according to the defined heuristic rules; obtaining specific disintegration intervals through the generated initial formation order population; generating the corresponding decomposition order for each formation order through the disintegration intervals; obtaining several formation orders and corresponding decomposition orders as the initial population and inputting them into the improved evolutionary algorithm for solution, and the population continuously evolves through the three processes of selection, crossover, and mutation; after reaching the preset number of evolutionary times, finally outputting the optimized decomposition order and formation order; Specifically, obtaining the initial formation order population includes: S21. Adopt the integer coding method to generate a formation sequence group with only one individual according to the principle that the trains with earlier departure are coded earlier, and calculate the latest formation time \(t\) of the departing train wb (j) and the waiting departure time \(t\) df (j). S22. For each individual in the current formation sequence population, if t df (j) ≥ t df (j + 1), adjust the formation sequences of the corresponding two trains, and expand the adjusted individual into the current formation sequence population to obtain an initial population of formation sequences generated according to the heuristic rule; In addition, based on the defined disintegration intervals, initializing based on the disintegration order to obtain a chromosome in the initial population, and obtaining the initial disintegration order population; In the population selection link of the improved evolutionary algorithm, calculating the fitness of each individual in the initial population according to the first fitness function, and making selections according to the fitness of each individual. The selection operators include: According to the Metropolis criterion, at a temperature T, the probability of a temperature decrease with an energy difference dE is: where k represents a constant, dE < 0, and T represents the temperature; The first fitness function is: γ1 and γ2 are constants; In the process of using the improved evolutionary algorithm to iteratively solve the dynamic traffic assignment model, the probability of a chromosome being selected and replicated is as follows: where λ is the number of iterations, p is the population size, and z max represents the maximum value of the fitness function of all individuals in the population; After the population is selected, in the population crossover link of the improved evolutionary algorithm, the crossover probability calculation formula is as follows: Among them, γ ∈ (0, 1) is the crossover probability coefficient, and f max is the ideal optimal solution of flow allocation, and all trains depart fully loaded and on time; Randomly generating two parent chromosomes according to the crossover probability, and then randomly generating the same gene segments according to a certain rule probability on the two parent chromosomes. These gene segments will be retained as excellent genes in the offspring chromosomes after crossover, and the other gene segments are sequentially obtained from the other parent chromosome after excluding the excellent genes; In the process of solving the dynamic traffic assignment model, the mutation operation adopts the swap mutation operator, randomly generating a mutant chromosome according to the mutation probability, and then randomly selecting two gene mutation positions, and swapping the genes at these two positions while keeping the other genes unchanged; S30 includes: Inputting the decomposition order of the arriving trains and the formation order of the departing trains into the static traffic assignment model, and using the hybrid sine-cosine whale optimization algorithm to solve the static traffic assignment model to obtain the specific formation content of the departing trains and clarify the connection relationship between the vehicle flows; First, set the whale individual vector X = (x(1), x(2), …, x(n)). Divide the whale individual vector into two segments of equal length, satisfying n = 2l, where l is the possible traffic flow sources of all departing trains within the stage plan. The first segment vector is used to represent the specific traffic flow sources of each train, including the arriving trains and the directions of the arriving trains. The second segment vector represents the selection order of the traffic flow sources, and set the value range of the individual position as [- ε, ε], where ε is the total number of departing trains within the stage plan; Secondly, encoding the first-segment vector, and mapping the whale individual position vector to the remaining number of available trains through formula (6) to complete the process of converting the individual position to the number of trains; Among them, x(j) is the whale individual vector, s(j) is the number of elements in the set of remaining train numbers, and u(j) is the selection serial number in the set of remaining train numbers; Encoding the second-segment vector using the maximum position rule, and selecting the different vehicle flow source order vectors representing the same formation train according to the magnitude of the element values. The larger the corresponding element value, the earlier the selection order; Finally, based on the decomposition order and formation order output by the traffic assignment model, determining that the number of vehicle flow sources of the departing trains is the second fitness function, establishing a mathematical model for optimization from three aspects: searching, surrounding, and preying on prey by imitating the way of whales hunting, and improving the WOA algorithm in combination with the SCA algorithm to obtain the specific vehicle flow sources of each departing train and complete the static traffic assignment; S30 also includes: Improving the stage of whales surrounding prey: At the beginning, a set of random solutions is set and the fitness of the solutions is calculated. At the same time, the solution with the highest fitness is regarded as the target, and the rest of the whale individuals will approach the target. To improve local exploitation, the SCA strategy is introduced into the local exploitation process of WOA, and its position update method is expressed as: X(t + 1) = X * (t) - A·D1 A = C·a - a Among them, X * is the prey position, t is the current iteration number, T is the maximum iteration number, two convergence factors a and r1 are modified and nonlinearly decrease from 2 to 0 with the iteration number, r2 ∈ [0, 2π], C ∈ [0, 2], θ ∈ [0, 1], all are random numbers, and λ and μ are constants; Improve the bubble-net predation stage: When the whale group preys, it will prey on the prey along a circular or "9"-shaped path. To enhance the diversity of the entire population and the ability to jump out of local optima, the SCA strategy is also used to improve the spiral update method as follows: D2 = |r3·X * (t) - X(t)| Among them, b is a constant defining the spiral shape, l ∈ [0, 1], θ ∈ [0, 1], r3 ∈ [0, 2π], all are random numbers. A random number p ∈ [0, 1] is set. When p ≥ 0.5, bubble-net predation is performed. When p < 0.5, the prey is surrounded; S30 includes: improving the stage of searching for prey: To further search for prey, WOA uses the change of A to conduct exploration, that is, when |A| < 1, the algorithm will conduct precise search, and the whale individuals will move towards the current optimal individual; SCA and WOA are mixed, and the optimization method of SCA is used to improve the global exploration method of WOA by combining the strong global exploration ability of SCA: X(t + 1) = X rand (t) - A·D3 Among them, X rand (t) is the position of a randomly selected whale individual.

2. A computing device, characterized in that, including: It includes a processor, a memory, and a computer program stored on the memory, characterized in that the processor executes the computer program to implement a railway marshalling station intelligent flow allocation method as described in any one of claims 1.

Images