Hybrid power locomotive energy management method based on parrot optimization algorithm

Abstract

The invention discloses a hybrid power locomotive energy management method based on a parrot-shaped algorithm, and the method comprises the steps: building a multi-objective optimization system, optimizing the parrot-shaped algorithm, training based on test data, and verifying the optimization performance, and the multi-objective optimization system comprehensively considers the key performance indexes of fuel economy, dynamic property and battery life. According to the method, the kinetic model of the fuel cell hybrid power locomotive is constructed, the Parrot optimization algorithm is introduced, the designed behavior conversion factor is combined, and the global search capability and the local optimization precision are considered, so that the optimization quality of an energy management strategy is effectively improved; under the complex working condition environment, collaborative optimization of multi-target performance such as locomotive economy, the service life of an energy storage system, the dynamic response capacity and the overall efficiency of the system can be achieved, and the operation efficiency and the system stability of the hybrid power locomotive are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electrical engineering and rail transit, and particularly relates to a hybrid locomotive energy management method based on a north pole parrot optimization algorithm. BACKGROUND

[0002] The background technology of hydrogen fuel hybrid locomotive energy management is driven by the dual needs of global rail transit low-carbon transformation and hydrogen energy technology breakthrough. The core is to solve the problems of slow dynamic response of hydrogen fuel cells, one-way energy flow and full working condition efficiency optimization through multi-energy system collaborative control. With the advancement of the "double carbon" strategy, the high carbon emission of traditional diesel locomotives (such as the annual carbon emission of shunting locomotives exceeding one thousand tons) and the range anxiety of pure electric drive form a sharp contrast. Hydrogen fuel cells have become a key path for decarbonization in rail transit due to their zero emission and high energy density (mass specific energy of 350-600 Wh / kg). However, due to the "soft" output characteristics of proton exchange membrane fuel cells (PEMFC) (power ramp rate of about 0.1 MW / s, which cannot match the instantaneous MW-level power demand of locomotives) and the defect that braking energy cannot be directly fed back, the industry generally adopts a hybrid architecture of "fuel cell + power battery / super capacitor". Through hierarchical voltage control (setting three protection thresholds of under-voltage / stabilization / over-voltage and connecting a bidirectional DC / DC converter), rule-based power distribution (adjusting the output ratio of the dual-source according to the load demand in low / middle / high power zones), and multi-time scale optimization (second-level working condition prediction, hundred-millisecond-level model prediction, and millisecond-level real-time response hierarchical control), energy efficient utilization is achieved. For example, the Coradia iLint train of Alstom in France achieves a system comprehensive efficiency of 48% through this technology, and the domestic CNR ZM scheme extends the service life of fuel cells to more than 20,000 hours. In the future, with the integration of edge computing and digital twin technology, the energy management system will further develop towards intelligence (online prediction of membrane electrode life), multi-energy flow collaboration (hydrogen-electric-thermal comprehensive utilization), and standardization (modular design under SAE / GB / T specifications), providing technical support for large-scale hydrogen energy replacement in rail transit.

[0003] Arctic puffin optimization (APO) is a meta-heuristic optimization algorithm based on the survival and predatory behavior of arctic puffins proposed in 2024. The APO algorithm includes two stages: air flight (exploration) and underwater foraging (development). In the air flight stage, the APO algorithm introduces Levy flight and speed factor mechanisms to improve convergence speed. In the underwater foraging stage, the APO algorithm uses strategies such as synergy effect and adaptive change factor to ensure that the algorithm can effectively use the current best position. In addition, the behavior transition factor is used to realize the dynamic transition between the exploration and development stages. In order to allow the algorithm to fully explore in the early stage, the behavior transition factor is redesigned, effectively balancing global exploration and local development. In order to verify the effectiveness of the proposed algorithm, it is applied to typical test functions and compared with other algorithms. The results show that by changing the behavior transition factor, the proposed algorithm has a good trade-off in solving global optimization problems or local optimization problems. SUMMARY

[0004] The present application provides a hybrid locomotive energy management method based on the Arctic puffin optimization algorithm, which can effectively solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solutions: a hybrid locomotive energy management method based on the Arctic puffin optimization algorithm, including multi-objective optimization system construction, Arctic puffin algorithm optimization, training based on test data, and optimization performance verification.

[0006] The multi-objective optimization system comprehensively considers the key performance indicators of fuel economy, power performance, and battery life, and constructs an energy management optimization model for hybrid locomotives.

[0007] The Arctic puffin algorithm is used to solve the model, improving search efficiency and global optimization ability.

[0008] The measured operating data is introduced to train and calibrate the optimization model, enhancing the adaptability and accuracy of the algorithm under actual working conditions.

[0009] Finally, the effectiveness and superiority of the optimization method are verified through simulation comparison and performance evaluation, realizing efficient and intelligent management of the energy distribution strategy of hybrid locomotives.

[0010] According to the above technical solutions, the four-dimensional objective function mathematical modeling of the multi-objective optimization system construction is established as follows:

[0011] For locomotive economy target:

[0012]

[0013] wherein, FuelCost(P eng (t)) is the function of hydrogen fuel cost and fuel cell output power P eng (t), EmissionCost(P eng (t)) is the function of hydrogen fuel cost and fuel cell output power P eng (t), both of which can be obtained by experiment or manufacturer's data;

[0014] λ1, λ2 are weight coefficients, hydrogen fuel cost and emission cost are calculated by table lookup method, through experience analysis, λ1 is taken as 0.3, λ2 is taken as 0.7, and N refers to N control strategy combinations;

[0015] Based on the change rate of battery discharge percentage DOD and residual capacity percentage SOC after locomotive operation discharge ends, a health degree model is constructed to obtain a storage life target:

[0016]

[0017] Deep discharge penalty: inhibit overcharge and overdischarge through the square term of DOD;

[0018] Dynamic current inhibition: SOC change rate term limits battery charge and discharge rate;

[0019] Parameter calibration: based on the accelerated aging test data of LFP battery, γ and δ are weight coefficients, generally γ = 0.6 and δ = 0.4 so that the deviation of the model from the actual life is less than 8%;

[0020] The relative deviation of demand power P req (t) and actual output power P out (t) is used to measure the response speed to obtain a power response target;

[0021]

[0022] Dynamic response threshold: requires 95% demand power tracking within 0.5 seconds;

[0023] Penalty mechanism: when the power deviation exceeds 10%, the objective function value decays exponentially.

[0024] According to the above technical scheme, the product of the efficiencies of each subsystem is used to calculate the comprehensive efficiency to obtain a system efficiency target, wherein, η eng (t) is the hydrogen fuel cell efficiency, η bat (t) is the battery efficiency, and η sc (t) is the super capacitor efficiency;

[0025]

[0026] Build constraints:

[0027] Power balance constraints:

[0028] P eng (t)+P bat (t)+P sc (t) = P req (t) (1.5)

[0029] Energy storage SOC constraints:

[0030] 0.2≤SOC bat (t)≤0.9,0.3≤SOC sc (t)≤0.8 (1.6)

[0031] Hydrogen fuel cell power constraints:

[0032] P eng,min ≤P eng (t)≤P eng,max (1.7).

[0033] According to the above technical solution, the optimization of the Arctic Puffin algorithm includes the following steps:

[0034] Modeling foraging behavior:

[0035] Simulate the two movement modes of puffins when foraging: "flying in the air" and "foraging underwater", which correspond to the local search and global exploration phases of the algorithm respectively. Before this, the variables need to be initialized.

[0036] Population initialization:

[0037] In the original APO algorithm, the initial population of the Arctic Puffin algorithm is an N×D matrix, X=[X1,X2,…,X N ] T , where X i is a D-dimensional vector, that is, X i =[X i,1 ,X i,2 ,…,X i,D ], which represents a combination of all the undetermined control parameters of energy management. The population initialization process is described by the following formula:

[0038]

[0039] in, represents the current position of the i-th Atlantic puffin, rand generates a random number between 0 and 1, ub and lb represent the upper and lower limits of the search space respectively, and N is the population size;

[0040] Air flight:

[0041] Arctic terns usually fly in groups, which can improve flight efficiency and the opportunity for cooperative hunting. When flying in the air, they first conduct an aerial search, during which they focus on detecting potential prey while remaining vigilant for potential predators nearby. The position update formula for this stage is as follows:

[0042]

[0043] R=round(0.5×(0.05+rand))×α(1.10)

[0044] α~Normal(0,1)(1.11)

[0045] represents the position of the i-th Arctic tern in the current population, that is, the i-th combination of all pending parameters in the control strategy, where r is a random integer between 1 and N-1, and does not include i, represents a randomly selected Arctic tern from the current population, L(D) represents a random number generated by Levy flight, D is the dimension, and α is a random sequence following a standard normal distribution.

[0046] According to the above technical scheme, when the Arctic terns search for suitable prey in the air, they will dive to catch the prey, at which time they will quickly change the flight direction to accelerate the diving speed. The position update formula for this stage is as follows:

[0047]

[0048] S=tan((rand-0.5)×π)(1.13)

[0049] where the Arctic tern adjusts its displacement during the diving process through the speed coefficient S;

[0050] In order to obtain better results in the exploration stage, the candidate solutions generated by the two processes of aerial search and diving hunting are combined, and then these candidate solutions are sorted and the top N are selected to form a new population. The merging process is as follows:

[0051]

[0052] where sort is to sort the new population in ascending order according to the fitness value;

[0053] Underwater foraging:

[0054] When Atlantic puffins dive underwater, they enter the underwater foraging phase. First, they adopt a collective strategy, gathering around schools of fish and foraging together. The position update formula for this process is as follows:

[0055]

[0056] F is the cooperation factor, which is used to adjust the foraging behavior of Atlantic puffins. In the original APO algorithm, F = 0.5, where r1, r2 and r3 are random integers between 1 and N-1, and and are three candidate solutions randomly selected from the current population, and r1≠r2≠r3,

[0057] After gathering foraging, the Atlantic puffin will change its underwater position in order to find more prey. The position update formula of this process is as follows:

[0058]

[0059] Among them, t represents the current iteration number, and T is the total number of iterations.

[0060] According to the above technical solution, when the Atlantic puffin is hunting underwater, it will encounter other predators. At this time, the Atlantic puffin will quickly change its position to avoid danger. The following is the escape strategy of the Atlantic puffin:

[0061]

[0062] represents the position of the i-th Atlantic puffin in the current population, are two candidate solutions randomly selected from the current population, and β is a uniformly distributed random number between 0 and 1;

[0063] In order to obtain the optimal results of the above three cases, the algorithm merges the candidate positions in the three different position equations into a new solution, then sorts these solutions according to their fitness values ​​and selects the top N individuals. The formula is described as follows:

[0064]

[0065] Behavior Conversion Factor

[0066] In the early stages of the algorithm, the Atlantic puffin tends to conduct global exploration through frequent aerial flights; in the later stages, the Atlantic puffin focuses on underwater foraging for local development. The traditional Arctic puffin algorithm does not have a clear rule to switch between these two behavior modes. Therefore, a behavior conversion factor B is designed to achieve the transition from global exploration to local development. The specific definition of behavior conversion factor B is as follows:

[0067] B = 2 x log(l / rand) x (1-t / T) (1.24)

[0068] Where: t represents the current iteration number, T is the total number of iterations, rand is a random number between 0 and 1. Based on the behavior conversion factor B and the parameter C, the search strategy of the algorithm in the iteration process is controlled. When B is greater than C, the algorithm adopts global exploration strategy, and when B is less than or equal to C, the algorithm turns to local development.

[0069] According to the above technical scheme, the training process based on test data is as follows: first, the historical running data is preprocessed and input into the algorithm, each sea parrot individual represents an energy allocation strategy, and the iteration optimization is carried out by simulating the foraging behavior of sea parrot group. In the early stage of the algorithm, the individual is in the "air flight" mode for global exploration, and the Levy flight mechanism is used to jump randomly in the solution space to cover a wider area, and at the same time, the diving strategy is used to quickly lock the potential high-quality solution;

[0070] With the deepening of iteration, the algorithm switches to "underwater foraging" mode, the individual learns from the optimal strategy of the neighborhood and adjusts cooperatively, and dynamic disturbance is introduced to avoid falling into local optimum;

[0071] In order to balance exploration and development, the algorithm uses behavior conversion factor to automatically adjust search intensity with iteration number, focusing on global scanning in early stage and strengthening local optimization in later stage;

[0072] After each iteration, the individual fitness is evaluated according to the multi-objective functions such as energy consumption, battery life, emission cost, etc., and the inferior strategy is eliminated and the optimal solution is retained;

[0073] Finally, the optimal strategy parameters are output, and their generalization ability is verified through test set. Compared with traditional algorithms, they need to show better convergence speed and strategy stability. In practical application, the parameters are embedded into locomotive control system to realize real-time dynamic power distribution, and periodic update of training data can be used to adapt to long-term working condition changes.

[0074] According to the above technical scheme, the optimization performance verification is verified by joint simulation verification of Pytorch+Matlab / Simulink, and the convergence of four optimization objectives is analyzed respectively.

[0075] Compared with the prior art, the present application has the beneficial effects that: the structure of the present application is reasonable, safe and convenient to use, by constructing a dynamic model of the fuel cell hybrid locomotive and introducing the Arctic Penguin optimization algorithm, combining the designed behavior conversion factor, taking into account the global search ability and local optimization precision, the optimization quality of the energy management strategy is effectively improved, under complex working condition environment, the multi-objective performance of the locomotive economy, energy storage system life, power response ability and system overall efficiency can be optimized, the running efficiency and system stability of the hybrid locomotive are significantly improved, the energy consumption and maintenance cost are reduced, and the present application has good engineering application prospect and popularization value. BRIEF DESCRIPTION OF DRAWINGS

[0076] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation of the present application.

[0077] In the drawings:

[0078] Fig. 1 is the Arctic Penguin algorithm flowchart of the present application;

[0079] Fig. 2 is the economic target and energy storage life target optimization convergence curve diagram of the present application;

[0080] Fig. 3 is the power response target and system efficiency target optimization convergence curve diagram of the present application; DETAILED DESCRIPTION

[0081] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, and it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and do not constitute a limitation of the present application.

[0082] Embodiment: As shown in the present application provides a technical scheme, a hybrid locomotive energy management method based on Arctic Penguin optimization algorithm, characterized by: including multi-objective optimization system construction, Arctic Penguin algorithm optimization, training based on test data, optimization performance verification; Figs. 1-3 The multi-objective optimization system comprehensively considers the key performance indicators of fuel economy, power and battery life, and constructs an energy management optimization model for the hybrid locomotive;

[0083] The Arctic Penguin algorithm is used to solve the model, and the search efficiency and global optimization ability are improved;

[0084] The measured running data is introduced to train and calibrate the optimization model, and the adaptability and accuracy of the algorithm under actual working conditions are enhanced;

[0085]

[0086] ​Finally, the effectiveness and superiority of the optimization method are verified through simulation comparison and performance evaluation, and efficient and intelligent management of the energy distribution strategy of hybrid locomotives is achieved.

[0087] According to the above technical solution, the mathematical modeling of the four-dimensional objective function in the construction of the multi-objective optimization system and the composite model of hydrogen fuel cost and emission cost are established as follows:

[0088] This solution breaks through the limitations of traditional single-objective optimization and constructs a four-dimensional objective function that includes economy, environmental protection, energy storage life, and dynamic response. There is a strong coupling relationship between the sub-objectives:

[0089] Regarding the economic goals of locomotives:

[0090]

[0091] Among them, FuelCost(P eng (t)) is the hydrogen fuel cost and the fuel cell output power P eng (t), EmissionCost(P eng (t)) is the hydrogen fuel cost and the fuel cell output power P eng (t) can be obtained through experiments or data provided by manufacturers;

[0092] λ1 and λ2 are weight coefficients. The hydrogen fuel cost and emission cost are calculated by table lookup method. Through empirical analysis, λ1 is taken as 0.3 and λ2 is taken as 0.7. N refers to the number of control strategy combinations.

[0093] A health model is constructed based on the battery discharge percentage DOD and remaining capacity percentage SOC change rate after the locomotive is discharged, and the energy storage life target is obtained:

[0094]

[0095] Deep discharge penalty: Overcharge and overdischarge are suppressed by the square term of DOD (for example, the penalty term is 0.64 when DOD = 80%, and 0.25 when DOD = 50%).

[0096] Dynamic current suppression: The SOC change rate item limits the battery charge and discharge rate (for example, a penalty is triggered when the SOC changes by more than 5% within 10 seconds);

[0097] Parameter calibration: Based on the accelerated aging test data of LFP batteries, γ and δ are weight coefficients. Generally, γ = 0.6 and δ = 0.4 are used to make the deviation between the model and the actual life less than 8%;

[0098] By the required power P req (t) and actual output power P outThe relative deviation of (t) measures the response speed, and the dynamic response target is obtained;

[0099]

[0100] Dynamic response threshold: 95% demand power tracking is required to be achieved within 0.5 seconds (corresponding to the subway train traction force step response standard);

[0101] Penalty mechanism: when the power deviation exceeds 10%, the objective function value decays exponentially (for example, the score is reduced to 0.64 when the deviation is 20%).

[0102] According to the above technical solution, the comprehensive efficiency is calculated based on the product of the efficiency of each subsystem, and the system efficiency target is obtained, wherein, η eng (t) is the hydrogen fuel cell efficiency, η bat (t) is the battery efficiency, η sc (t) is the super capacitor efficiency;

[0103]

[0104] Constraint condition is constructed:

[0105] Power balance constraint:

[0106] P eng (t) + P bat (t) + P sc (t) = P req (t) (1.5)

[0107] Energy storage SOC constraint:

[0108] 0.2≤SOC bat (t)≤0.9,0.3≤SOC sc (t)≤0.8 (1.6)

[0109] Hydrogen fuel cell power constraint:

[0110] P eng,min ≤P eng (t)≤P eng,max (1.7).

[0111] According to the above technical solution, the arctic sea parrot algorithm optimization includes the following steps:

[0112] Foraging behavior modeling:

[0113] The two motion modes of "air flight" and "underwater foraging" of the sea parrot group when foraging are simulated, which correspond to the local search and global exploration stages of the algorithm respectively, and before that, the variables (population) need to be initialized;

[0114] Population initialization:

[0115] In the original APO algorithm, the initial population of the Arctic Puffin algorithm is an N×D matrix, X=[X1,X2,…,X N ] T , where X i is a D-dimensional vector, that is, X i =[X i,1 ,X i,2 ,…,X i,D ], which represents a combination of all the undetermined control parameters of energy management. The population initialization process is as follows:

[0116]

[0117] in, represents the current position of the i-th Atlantic puffin, rand generates a random number between 0 and 1, ub and lb represent the upper and lower limits of the search space respectively, and N is the population size;

[0118] Air flight:

[0119] Atlantic puffins typically fly in groups, which improves flight efficiency and the chances of cooperative hunting. While in the air, they first conduct an aerial search. During this phase, they focus on detecting potential prey while remaining alert to nearby potential predators. The position update formula for this phase is as follows:

[0120]

[0121] R=round(0.5×(0.05+rand))×α (1.10)

[0122] α~Normal(0,1) (1.11)

[0123] represents the position of the i-th Arctic puffin in the current population, that is, the i-th combination of all the undetermined parameters in the control strategy, where r is a random integer between 1 and N-1, excluding i, represents a randomly selected Arctic puffin from the current population (control parameter combination), L(D) represents a random number generated by Levy flight, D is the dimension, and α is a random sequence obeying the standard normal distribution.

[0124] According to the above technical solution, when an Atlantic puffin searches for suitable prey in the air, it will dive to catch it. At this time, it will quickly change its flight direction to speed up the dive. The position update formula at this stage is as follows:

[0125]

[0126] S = tan((rand - 0.5) × π)(1.13)

[0127] The Arctic puffin uses the speed coefficient S to adjust its displacement during the dive;

[0128] In order to obtain better results in the exploration phase, the candidate solutions generated by the aerial search and diving predation processes are merged, and then these candidate solutions are sorted and the top N are selected to form a new population. The merging process is as follows:

[0129]

[0130] Among them, sort is to sort the new population from small to large according to the fitness value;

[0131] Underwater foraging:

[0132] When Atlantic puffins dive underwater, they enter the underwater foraging phase. First, they adopt a collective strategy, gathering around schools of fish and foraging together. The position update formula for this process is as follows:

[0133]

[0134] F is the cooperation factor, which is used to adjust the foraging behavior of Atlantic puffins. In the original APO algorithm, F = 0.5, where r1, r2 and r3 are random integers between 1 and N-1 (excluding i), and and are three candidate solutions randomly selected from the current population, and r1≠r2≠r3,

[0135] After gathering foraging, the Atlantic puffin will change its underwater position in order to find more prey. The position update formula of this process is as follows:

[0136]

[0137]

[0138] Among them, t represents the current iteration number, and T is the total number of iterations.

[0139] According to the above technical solution, when the Atlantic puffin is hunting underwater, it will encounter other predators. At this time, the Atlantic puffin will quickly change its position to avoid danger. The following is the escape strategy of the Atlantic puffin:

[0140]

[0141] represents the position of the i-th Atlantic puffin in the current population, are two candidate solutions randomly selected from the current population, and β is a uniformly distributed random number between 0 and 1;

[0142] In order to obtain the optimal results of the above three cases, the algorithm merges the candidate positions in the three different position equations into a new solution, then sorts these solutions according to their fitness values ​​and selects the top N individuals. The formula is described as follows:

[0143]

[0144] Behavior Conversion Factor

[0145] In the early stages of the algorithm, the Atlantic puffin tends to conduct global exploration through frequent aerial flights; in the later stages, the Atlantic puffin focuses on underwater foraging for local development. The traditional Arctic puffin algorithm does not have a clear rule to switch between these two behavior modes. Therefore, a behavior conversion factor B is designed to achieve the transition from global exploration to local development. The specific definition of behavior conversion factor B is as follows:

[0146] B=2×log(1 / rand)×(1-t / T)(1.24)

[0147] Where: t represents the current iteration number, T is the total number of iterations, and rand is a random number between 0 and 1. The algorithm's search strategy during the iteration process is controlled based on the behavior conversion factor B and parameter C (C = 0.5 in the original APO algorithm). When B is greater than C, the algorithm adopts a global exploration strategy. When B is less than or equal to C, the algorithm turns to local development.

[0148] According to the technical solution, the training process based on the test data is as follows: first, the historical operation data (such as speed curve, load change, battery state) is preprocessed and input into the algorithm, each individual of the sea parrot represents an energy distribution strategy (such as power ratio of hydrogen fuel cell and storage battery, super capacitor, charging and discharging threshold), and the population foraging behavior of the sea parrot is simulated for iterative optimization - in the early stage of the algorithm, the individual performs global exploration in the "air flight" mode, randomly jumps in the solution space by using the Levy flight mechanism to cover a wider area, and quickly locks the potential high-quality solution by using the diving strategy; as the iteration goes deep, the algorithm switches to the "underwater foraging" mode, the individual learns from the neighborhood optimal strategy and adjusts cooperatively, and dynamic disturbance is introduced to avoid falling into local optimum; in order to balance exploration and development, the algorithm uses a behavior conversion factor to automatically adjust the search intensity with the number of iterations, and focuses on global scanning in the early stage and strengthens local optimization in the later stage; after each iteration, the individual adaptability is evaluated according to the multi-objective functions such as energy consumption, battery life and emission cost, and the inferior strategy is eliminated and the optimal solution is retained; finally, the optimal strategy parameters are output, and the generalization ability is verified through the test set, and compared with the traditional algorithm (such as particle swarm optimization), the convergence speed and strategy stability need to be better, and in actual application, the parameters are embedded into the locomotive control system to realize real-time dynamic power distribution, and the training data can be updated periodically to adapt to long-term working condition changes.

[0149] According to the technical solution, the optimization performance verification is verified by joint simulation verification of Pytorch+Matlab / Simulink, and the convergence of four optimization objectives is analyzed. The results show that the fuel consumption can be reduced by 4.7% (compared with the constant power distribution method); the battery cycle life is prolonged by 6.2% (based on the rain flow counting method); and the dynamic response time is shortened to 0.32s (21% shorter than NSGA-III optimization).

[0150] Finally, it should be pointed out that: the above-mentioned only for the preferred examples of the present application, and not for limiting the present application, although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm, characterized by: Including the construction of a multi-objective optimization system, optimization of the Arctic Puffin algorithm, training based on test data, and optimization performance verification; The multi-objective optimization system comprehensively considers the key performance indicators of fuel economy, power and battery life to build an energy management optimization model for hybrid locomotives; Use the Arctic Puffin algorithm to solve the model, improving search efficiency and global optimization capabilities; By introducing measured operating data to train and calibrate the optimization model, the adaptability and accuracy of the algorithm under actual working conditions are enhanced; Finally, the effectiveness and superiority of the optimization method are verified through simulation comparison and performance evaluation, and efficient and intelligent management of the energy distribution strategy of hybrid locomotives is achieved.

2. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 1 is characterized in that: The establishment of mathematical modeling of four-dimensional objective function in the construction of the multi-objective optimization system The composite model of hydrogen fuel cost and emission cost is established as follows: Regarding the economic goals of locomotives: Among them, FuelCost (P eng (t)) is the hydrogen fuel cost and the fuel cell output power P eng (t), EmissionCost(P eng (t)) is the hydrogen fuel cost and the fuel cell output power P eng (t) can be obtained through experiments or data provided by manufacturers; λ1 and λ2 are weight coefficients. The hydrogen fuel cost and emission cost are calculated by table lookup method. Through empirical analysis, λ1 is taken as 0.3 and λ2 is taken as 0.

7. N refers to the number of control strategy combinations. A health model is constructed based on the battery discharge percentage DOD and remaining capacity percentage SOC change rate after the locomotive is discharged, and the energy storage life target is obtained: Deep discharge penalty: suppress overcharge and overdischarge through the square term of DOD; Dynamic current suppression: The SOC change rate term limits the battery charge and discharge rate; Parameter calibration: Based on the accelerated aging test data of LFP batteries, γ and δ are weight coefficients. Generally, γ = 0.6 and δ = 0.4 are used to make the deviation between the model and the actual life less than 8%; By the required power P req (t) and actual output power P out The relative deviation of (t) measures the response speed and obtains the dynamic response target; Dynamic response threshold: 95% of the required power tracking is required within 0.5 seconds; Penalty mechanism: When the power deviation exceeds 10%, the objective function value decays exponentially.

3. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 2 is characterized in that: The comprehensive efficiency is calculated based on the product of the efficiencies of each subsystem to obtain the system efficiency target, where η eng (t) is the hydrogen fuel cell efficiency, η bat (t) is the battery efficiency, η sc (t) is the supercapacitor efficiency; Build constraints: Power balance constraints: P eng (t)+P bat (t)+P sc (t)=P req (t) (1.5) Energy storage SOC constraints: 0.2≤SOC bat (t)≤0.9,0.3≤SOC sc (t)≤0.8 (1.6) Hydrogen fuel cell power constraints: P eng,min ≤P eng (t)≤P eng,max (1.7)。 4. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 1 is characterized in that: The Arctic Puffin algorithm optimization includes the following steps: Modeling foraging behavior: Simulate the two movement modes of puffins when foraging: "flying in the air" and "foraging underwater," corresponding to the local search and global exploration phases of the algorithm, respectively. Before this, variables need to be initialized. Population initialization: In the original APO algorithm, the initial population of the Arctic Puffin algorithm is an N×D matrix, X=[X1,X2,…,X N ] T , where X i is a D-dimensional vector, that is, X i =[X i,1 ,X i,2 ,…,X i,D ], which represents a combination of all the undetermined control parameters of energy management. The population initialization process is described by the following formula: in, represents the current position of the i-th Atlantic puffin, rand generates a random number between 0 and 1, ub and lb represent the upper and lower limits of the search space respectively, and N is the population size; Air flight: Atlantic puffins typically fly in groups, which improves flight efficiency and the chances of cooperative hunting. While in the air, they first conduct an aerial search. During this phase, they focus on detecting potential prey while remaining alert to nearby potential predators. The position update formula for this phase is as follows: R=round(0.5×(0.05+rand))×α (1.10) α~Normal(0,1) (1.11) represents the position of the i-th Arctic puffin in the current population, that is, the i-th combination of all the undetermined parameters in the control strategy, where r is a random integer between 1 and N-1, excluding i, represents a randomly selected Atlantic puffin from the current population, L(D) represents a random number generated by Levy flight, D is the dimension, and α is a random sequence following a standard normal distribution.

5. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 4 is characterized in that: When an Atlantic puffin searches for suitable prey in the air, it will dive to catch it. At this time, it will quickly change its flight direction to speed up the dive. The position update formula at this stage is as follows: S=tan((rand-0.5)×π) (1.13) The Arctic puffin uses the speed coefficient S to adjust its displacement during the dive; In order to obtain better results in the exploration phase, the candidate solutions generated by the aerial search and diving predation processes are merged, and then these candidate solutions are sorted and the top N are selected to form a new population. The merging process is as follows: Among them, sort is to sort the new population from small to large according to the fitness value; Underwater foraging: When Atlantic puffins dive underwater, they enter the underwater foraging phase. First, they adopt a collective strategy, gathering around schools of fish and foraging together. The position update formula for this process is as follows: F is the cooperation factor, which is used to adjust the foraging behavior of Atlantic puffins. In the original APO algorithm, F = 0.5, where r1, r2 and r3 are random integers between 1 and N-1, and and are three candidate solutions randomly selected from the current population, and r1≠r2≠r3, After gathering foraging, the Atlantic puffin will change its underwater position in order to find more prey. The position update formula of this process is as follows: Among them, t represents the current iteration number, and T is the total number of iterations.

6. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 5 is characterized in that: When the Atlantic puffin is hunting underwater, it may encounter other predators. At this time, the Atlantic puffin will quickly change its position to avoid danger. The following are the escape strategies of the Atlantic puffin: represents the position of the i-th Atlantic puffin in the current population, are two candidate solutions randomly selected from the current population, and β is a uniformly distributed random number between 0 and 1; In order to obtain the optimal results of the above three cases, the algorithm merges the candidate positions in the three different position equations into a new solution, then sorts these solutions according to their fitness values ​​and selects the top N individuals. The formula is described as follows: Behavior Conversion Factor In the early stages of the algorithm, the Atlantic puffin tends to conduct global exploration through frequent aerial flights; in the later stages, the Atlantic puffin focuses on underwater foraging for local development. The traditional Arctic puffin algorithm does not have a clear rule to switch between these two behavior modes. Therefore, a behavior conversion factor B is designed to achieve the transition from global exploration to local development. The specific definition of behavior conversion factor B is as follows: B=2×log(1 / rand)×(1-t / T)(1.24) Where: t represents the current iteration number, T is the total number of iterations, and rand is a random number between 0 and 1. The algorithm's search strategy during the iteration process is controlled based on the behavior conversion factor B and parameter C. When B is greater than C, the algorithm adopts a global exploration strategy. When B is less than or equal to C, the algorithm turns to local development.

7. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 1 is characterized in that: The training process based on test data is as follows: First, historical operating data is preprocessed and fed into the algorithm. Each individual puffin represents an energy allocation strategy, and iterative optimization is performed by simulating the group foraging behavior of puffins. In the early stages of the algorithm, individuals perform global exploration in an "aerial" mode, using the Levy flight mechanism to randomly jump around in the solution space to cover a wider area. At the same time, a diving strategy is used to quickly identify potential high-quality solutions. As the iteration progresses, the algorithm switches to "underwater foraging" mode, where individuals learn from the optimal strategy in the neighborhood and adjust collaboratively, while introducing dynamic perturbations to avoid falling into local optimality. To balance exploration and development, the algorithm uses a behavior conversion factor to automatically adjust the search intensity with the number of iterations, focusing on global scanning in the early stage and strengthening local optimization in the later stage; After each iteration, the individual adaptability is evaluated based on multiple objective functions such as energy consumption, battery life, and emission costs, and inferior strategies are eliminated while retaining the optimal solution; Finally, the optimal strategy parameters are output and their generalization ability is verified through the test set. Compared with traditional algorithms, they need to show better convergence speed and strategy stability. In actual application, the parameters are embedded in the locomotive control system to achieve real-time dynamic power distribution, and the training data can be periodically updated to adapt to long-term working condition changes.

8. The hybrid locomotive energy management method based on the Arctic Puffin optimization algorithm according to claim 1 is characterized in that: The optimization performance was verified by Pytorch+Matlab / Simulink joint simulation, and convergence analysis was performed on the four optimization objectives respectively.

Images