A fuel cell locomotive real-time energy management method

Abstract

This invention discloses a real-time energy management method for fuel cell locomotives, comprising: acquiring historical typical driving condition data of the fuel cell hybrid locomotive and constructing a Markov state transition probability matrix; acquiring the current driving condition and predicting the vehicle speed based on the Markov state transition probability matrix; establishing a whole-vehicle power system model of the fuel cell locomotive based on the power system structure of the fuel cell hybrid locomotive; acquiring the relationship between costate variables and SOC changes offline using the Pontryagin minimum principle; obtaining the optimal costate variables using the relationship between the costate variables and SOC changes; correcting the Hamiltonian function online based on the predicted vehicle speed; updating the corrected Hamiltonian function based on the optimal costate variables; minimizing the updated Hamiltonian function to obtain the optimal power allocation among the power batteries, thereby realizing real-time energy management of the fuel cell locomotive.

Description

Technical Field

[0001] This invention belongs to the field of new energy locomotive technology, and in particular relates to a real-time energy management method for fuel cell locomotives. Background Technology

[0002] After years of research and development, hydrogen energy, as the ultimate energy source for human society, has gained widespread global recognition and high regard for its value, and its large-scale construction and application have entered a period of explosive growth. Fuel cells, as one of the most important forms of hydrogen energy utilization, are a technology with significant research significance and development potential in the hydrogen energy industry. With the advancement of proton exchange membrane fuel cell technology, the application of fuel cells in the field of vehicle power has developed rapidly. Fuel cells have also begun to be used in locomotives. Fuel cell locomotives, equipped with fuel cell systems, hydrogen storage systems, and other equipment, completely eliminate the traction power supply system, achieving energy conservation and environmental protection while also significantly reducing line construction costs. This makes the overall system investment cost lower than that of the overhead contact line traction power supply system, thus giving it a strong market competitiveness.

[0003] Energy management is one of the key technologies for fuel cell vehicles. Currently, global optimization energy management strategies (EMS) include dynamic programming (DP) and Pontryagin's Minimum Principle (PMP). DP can obtain a globally optimal solution, but it requires global prior information and is computationally intensive, making real-time application impossible. Compared to DP, PMP has lower computational requirements and can obtain a globally optimal solution or approach it, but it also requires global prior information, thus making real-time application difficult as well.

[0004] Therefore, considering the relatively fixed nature of locomotive routes, historical operating conditions and SOC feedback can be combined to realize the real-time application of the PMP algorithm, while simultaneously satisfying the real-time and optimal nature of the energy management strategy. Summary of the Invention

[0005] To address the aforementioned technical challenges, this invention proposes a real-time energy management method for fuel cell locomotives, which ensures near-global optimal control performance while enabling real-time application. Furthermore, based on speed prediction technology and SOC feedback, the adaptability of the energy management strategy is further enhanced.

[0006] To achieve the above objectives, the present invention provides a real-time energy management method for fuel cell locomotives, comprising:

[0007] Obtain historical typical driving condition data of fuel cell hybrid electric vehicles, and construct Markov state transition probability matrices for speed and acceleration under different driving conditions;

[0008] Obtain the current driving conditions and predict the vehicle speed based on the Markov state transition probability matrix;

[0009] Based on the power system structure of the fuel cell hybrid locomotive, a model of the entire vehicle power system of the fuel cell locomotive is established;

[0010] Based on the Pontryagin minimum principle, the historical typical driving condition data, and the vehicle power system model of the fuel cell hybrid locomotive, the relationship between costate variables and SOC changes was obtained offline using the target shooting method.

[0011] Based on the difference between the target SOC and the current SOC and the current driving conditions, and by utilizing the relationship between the co-state variable and the change in SOC, the optimal co-state variable is obtained.

[0012] The Hamiltonian function is corrected online based on the predicted vehicle speed.

[0013] The corrected Hamiltonian function is updated based on the optimal costate variables.

[0014] The updated Hamiltonian function is minimized, and the optimal power allocation between the fuel cell and the power battery is obtained based on the fuel cell hybrid locomotive power system model to achieve real-time energy management of the fuel cell locomotive.

[0015] Optionally, obtaining the current driving conditions includes:

[0016] Based on the historical typical driving condition data, a dataset is constructed, and the learning vector quantization neural network is trained using the dataset to obtain a driving condition recognition model.

[0017] Real-time feature parameters of the fuel cell hybrid locomotive are extracted, and the current driving conditions are obtained by identifying the real-time feature parameters through the operating condition identification model.

[0018] Optionally, based on the aforementioned historical typical driving conditions, the dataset is constructed as follows:

[0019] The historical typical driving condition data is divided into several segments, and the feature parameters of each segment are extracted. The feature parameters include average vehicle speed, average acceleration, average deceleration, proportion of high-speed time, proportion of low-speed time, and proportion of constant speed time.

[0020] The dataset is constructed using the feature parameters and the corresponding operating condition categories.

[0021] Optionally, constructing the Markov state transition probability matrix for speed and acceleration under different driving conditions includes:

[0022] Extract acceleration and speed information from the historical typical driving condition data;

[0023] The acceleration information and the velocity information are divided to obtain equally spaced acceleration state sequences and equally spaced velocity sequences, and a state number index matrix is ​​constructed.

[0024] The position number of the velocity and acceleration state at each moment is assigned a position number using the state number index matrix, wherein the position numbers of the velocity and acceleration states at the same moment constitute a state quantity.

[0025] A state transition frequency empty matrix is ​​constructed using the state number index matrix; based on the state transition frequency empty matrix, the state quantities at two adjacent time points are obtained and 1 is incremented at the corresponding positions of the state transition frequency empty matrix to construct the state transition frequency matrix.

[0026] Based on the state transition frequency matrix, construct the Markov state transition probability matrix;

[0027] The method for constructing the Markov state transition probability matrix is ​​as follows:

[0028]

[0029] Where P(i,j) is the state transition probability matrix, and T i,j Let be the element in the i-th row and j-th column of the state transition frequency matrix, where i and j are the state variables Fi and Fj corresponding to times t and t+1, respectively, and M and N are the discrete state numbers of acceleration and velocity, respectively.

[0030] Optionally, obtaining the equally spaced acceleration state sequence and the equally spaced velocity sequence includes:

[0031] Obtain the minimum and maximum acceleration values ​​from the acceleration information, and obtain the interval between the minimum and maximum acceleration values ​​as the first interval;

[0032] The acceleration information is divided into an equally spaced acceleration state sequence with a length of a first interval by the first interval;

[0033] Obtain the maximum speed from the speed information, and obtain the distance from the speed of zero to the maximum speed as the second distance;

[0034] The velocity information is divided into equally spaced velocity sequences of a second length by the second spacing.

[0035] Optionally, the fuel cell locomotive power system model includes a locomotive longitudinal dynamics model, a fuel cell hydrogen consumption and efficiency model, and a power battery system model; the required driving power of the whole vehicle, the output power of the fuel cell system, and the output power of the power battery are obtained through the locomotive longitudinal dynamics model, the fuel cell system hydrogen consumption and efficiency model, and the power battery system model, respectively.

[0036] Optionally, based on the Pontryagin minimum principle, the historical typical driving condition data, and the fuel cell hybrid locomotive's overall power system model, the relationship between costate variables and SOC changes can be obtained offline using a target-shooting method, including:

[0037] The total hydrogen consumption of the fuel cell is used as a cost function;

[0038] Establish constraints on the output power of the fuel cell system, the output power of the power battery, and the range of SOC variation;

[0039] Based on the Pontryagin minimum principle, a Hamiltonian function is established according to the cost function and the constraints.

[0040] By solving the Hamiltonian function using different costate variables, the relationship between the costate variables and the change in SOC under historical driving conditions can be obtained.

[0041] Optionally, online correction of the Hamiltonian function includes:

[0042] The average and standard deviation of the predicted vehicle speed are obtained based on the predicted vehicle speed;

[0043] The Hamiltonian function is modified by introducing a penalty term to limit the output power fluctuation of the fuel cell system;

[0044] The penalty for limiting the output power fluctuation of the fuel cell system is:

[0045]

[0046] Where λ1 is the penalty term for limiting the output power fluctuation of the fuel cell system, P fc P represents the output power of the fuel cell system, where t is the time interval. fc,max P represents the minimum output power of the fuel cell system. fc,min This represents the minimum output power of the fuel cell system.

[0047] The Hamiltonian function is modified based on the penalty term for maintaining SOC according to the mean and the standard deviation;

[0048] The penalty term for maintaining SOC is:

[0049]

[0050] Where λ2 is the penalty term for maintaining SOC, v std v is the standard deviation of the predicted velocity. ave This represents the average value of the predicted speed.

[0051] Optionally, the method for updating the corrected Hamiltonian function based on the optimal costate variables is as follows:

[0052]

[0053] Wherein, H(SOC(t),P fc Let (t), λ(t) be the Hamiltonian function, and ΔSOC be the difference between the target SOC and the current SOC. For the hydrogen consumption rate of the fuel cell system, Let λ(t) be the rate of change of SOC, and let λ(t) be the optimal costate variable.

[0054] Optionally, the method for obtaining the optimal power allocation between the fuel cell and the power battery is as follows:

[0055]

[0056] In the formula, P fc P is the output power of the fuel cell system. b P is the output power of the power battery system. req For the required power of the whole vehicle, H(SOC(t),P) fc (t),λ(t)) are Hamiltonian functions; Let λ be the rate of change of SOC, and λ(t) be the optimal costate variable.

[0057] Technical effects of the invention:

[0058] (1) This invention utilizes the relationship between the co-state variable and the battery SOC change in the locomotive driving conditions obtained by offline processing, and matches typical historical driving conditions according to the working condition identification algorithm, and considers the real-time feedback of the SOC of the power battery pack to obtain the optimal co-state variable λ(t). The optimization result is close to the global optimum.

[0059] (2) In solving for optimal power allocation, the present invention only updates the Hamiltonian function obtained offline, which reduces the amount of computation and can be used for real-time control.

[0060] (3) The present invention adds a penalty term for fuel cell power fluctuation and power battery SOC change during the updating of Hamiltonian function, which improves the service life of fuel cell and maintains the battery pack in the high-efficiency operating range.

[0061] (4) In the process of updating the Hamiltonian function, this invention takes into account the short-term driving speed of the locomotive in the future and the impact of future information on the optimization results, which further improves the ability of the power battery pack to maintain the SOC target value. Attached Figure Description

[0062] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0063] Figure 1 This is a flowchart illustrating an energy management strategy for a fuel cell locomotive according to an embodiment of the present invention;

[0064] Figure 2 This is a schematic diagram showing typical historical driving conditions and segment divisions selected for embodiments of the present invention;

[0065] Figure 3 This is a schematic diagram of the structure for LVQ neural network-based condition recognition in an embodiment of the present invention;

[0066] Figure 4 This is a schematic diagram illustrating the state numbering based on velocity and acceleration according to an embodiment of the present invention;

[0067] Figure 5 This is a schematic diagram of the velocity and acceleration Markov state transition probability matrix for a typical working condition selected in an embodiment of the present invention.

[0068] Figure 6 This is a structural diagram of the fuel cell locomotive power system in an example of the present invention;

[0069] Figure 7 This is a graph showing the relationship between the costate variable and the SOC variable under three selected driving conditions in this invention. Detailed Implementation

[0070] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0071] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0072] like Figure 1 As shown, this embodiment provides a real-time energy management method for fuel cell locomotives, including:

[0073] Step 1: Obtain typical historical driving conditions of the fuel cell hybrid electric vehicle and divide them into several segments, such as... Figure 2 As shown, feature parameters for each segment are extracted, including: average speed, average acceleration, average deceleration, proportion of high-speed time, proportion of low-speed time, and proportion of constant-speed time. A learned vector quantization (LVQ) neural network is then used for training in condition recognition, as shown below. Figure 3 As shown, the feature parameters and working condition categories are combined to form the training data for that working condition. Each feature parameter vector [x1; x2; ...; x...] n The input layer is defined as [[]], and the output layer is defined as [[]], where the number of hidden layers is set to 20. The LVQ neural network function built into MATLAB is used, and its calling format is as follows: net = newlvq(PR, S1, PC, LR, LF); PR is a matrix representing the range of feature parameter vector values ​​after segmentation [x...]. min x max Sl represents the number of neurons in the hidden layer; PC represents the percentage of each class in the total training data; LR represents the learning rate, with a default value of 0.01; LF represents the learning function, with a default value of learnlv1.

[0074] Step 2: Construct the velocity and acceleration Markov probability transition matrix for each typical driving condition. This includes the following steps:

[0075] Step 2.1: Extract acceleration and velocity information under typical historical working conditions, and divide the acceleration into segments of length M, representing the minimum acceleration value a from the working conditions. min to the maximum acceleration a max The equally spaced acceleration state sequence is divided into velocity ranges of length N, from 0 to the maximum velocity V during the working condition. max The equally spaced velocity sequences are given, where M and N are the discrete state numbers for acceleration and velocity, respectively. An M-row, N-column state index matrix is ​​constructed to number the velocity and acceleration states at each moment, such as... Figure 4 As shown.

[0076] Step 2.2: Obtain the velocities V(k), V(k+1), and accelerations a(k), a(k+1) for different typical driving conditions. Also, obtain the discrete values ​​Vi and Vj corresponding to V(k) and V(k+1) in the velocity sequence, and the discrete values ​​ai and aj corresponding to a(k) and a(k+1) in the acceleration sequence. Each pair of Vi and ai constitutes a state variable Fi, and Vj and aj constitute a state variable Fj. Iterate through the driving data at each time step, and find the velocity and acceleration position numbers corresponding to times t and t+1 in the state index matrix constructed in Step 2.1. Simultaneously, an empty state transition frequency matrix T0(i,j) with M*N rows and columns is constructed to store the transition frequency of each state, which facilitates the acquisition of the state quantity at each adjacent time. Here, the row represents the state quantity Fi at time t, and the column represents the state quantity Fj at time (t+1). Based on the state quantities Fi and Fj, the corresponding positions of the empty state transition frequency matrix T0(i,j) are cumulatively incremented by 1 to construct the state transition frequency matrix, i.e., T(i,j).

[0077] Step 2.3: Calculate the state transition probability matrix based on the frequency matrix, such as... Figure 5 As shown:

[0078]

[0079] Where P(i,j) is the state transition probability matrix, and T i,j The element in the i-th row and j-th column of the state transition frequency matrix constructed in step 2.2 is Fi and Fj, which are the state variables Fi and Fj at times t and t+1, respectively.

[0080] Step 3: Based on the power system structure of the fuel cell hybrid locomotive, establish a complete power system model of the fuel cell locomotive. In this embodiment of the invention, the fuel cell is indirectly connected to the DC bus through a unidirectional DC / DC converter, and the power battery is directly connected to the DC bus, as shown in the example. Figure 6 As shown, the vehicle controller can communicate bidirectionally with the power battery system, fuel cell system, and DC / DC converter via the CAN bus to control the power distribution between the power battery and the fuel cell, thereby realizing the energy management strategy proposed in this invention.

[0081] Step 3.1, the longitudinal dynamics model of the whole vehicle is as follows:

[0082]

[0083] F i =mgsinα

[0084]

[0085]

[0086] In the formula, F t F i F j F ω These represent locomotive traction force, gradient resistance, acceleration resistance, and basic resistance, respectively; m is the total mass of the train; m e ω is the total mass of the train, including its rotational mass; γ is the slewing mass coefficient; g is the gravitational acceleration; α is the gradient angle; v is the train speed; and ω0 is the unit basic resistance.

[0087] The basic resistance is a quadratic function of the vehicle's speed:

[0088] ω0=A ω +B ω v+C ω v 2

[0089] In the formula: A ω B ω C ω This represents the basic operating resistance coefficient of the locomotive and rolling stock. Further, the locomotive power balance equation can be derived:

[0090]

[0091] In the formula, P req Driven by the locomotive's required power; η t For transmission system efficiency.

[0092] Step 3.2, construct the fuel cell hydrogen consumption and efficiency model as follows:

[0093] P fc =P stack -P BOP

[0094]

[0095]

[0096] Among them, P fc For the output power of the fuel cell system, P stack P represents the stack power of the fuel cell. BOP Power consumed by the accessories of the fuel cell system; n fc I represents the number of fuel cell cells. fc M is the fuel cell current. H2 n is the molar mass of hydrogen. e η is the number of electrons transferred in the electrochemical reaction, F is the Faraday constant, and η is the number of electrons transferred in the reaction. fc For fuel cell system efficiency, E H2 Hydrogen has a low calorific value (120 kJ / g). This represents the hydrogen consumption rate.

[0097] Step 3.3, construct the power battery model as follows:

[0098] P b =U oc I b -I b 2 R b

[0099] Among them, U oc R is the open-circuit voltage of the power battery. b P is the internal resistance of the power battery. b This represents the output power of the power battery. Therefore, the current I... b This can be expressed as:

[0100]

[0101]

[0102] In the formula, Q is the rate of change of SOC. b This refers to the capacity of the power battery.

[0103] Step 4: Using total hydrogen consumption as the cost function and establishing a Hamiltonian function, simultaneously determine the constraints for control variables and state variables. Based on the Pontryagin minimum principle and historical driving condition information, a "shooting method" is used—that is, continuously trying different costate variables—to solve the problem offline, obtaining the relationship between the costate variable λ(t) and the change in SOC under historical driving conditions. Specifically, this includes the following steps:

[0104] Step 4.1, using the total hydrogen consumption of the fuel cell as the cost function, the expression is:

[0105]

[0106] In the formula, J is the cost function. This represents the hydrogen consumption rate of the fuel cell.

[0107] Step 4.2, create the constraints for the power and SOC variation range of the fuel cell and power battery as follows:

[0108]

[0109] Step 4.3, based on the Pontryagin minimum principle, the Hamiltonian function established by minimizing the cost function is:

[0110]

[0111] In the formula, H(SOC(t),P fc (t),λ(t)) are Hamiltonian functions.

[0112] Step 4.4: Using different costate variables λ(t), offline calculations are performed to determine the initial and final changes in SOC under various operating conditions when the Hamiltonian function reaches its minimum, thus obtaining the relationship curve between costate variable λ(t) and SOC change. For example... Figure 7 As shown.

[0113] Step 5.1, set the sampling period for the vehicle speed sequence, in t k Time will pass. k - The data from the H1 driving time is used for real-time vehicle feature extraction data, and the same feature parameters as in step 1 are extracted to match the current driving conditions with historical conditions.

[0114] Step 5.2: Based on the matched working condition, select the Markov state transition probability matrix corresponding to the working condition in Step 2 and calculate the vehicle speed for the next 5 seconds based on the real-time collected speed.

[0115] Step 6: Consider the real-time SOC feedback of the power battery pack. Based on the difference between the target SOC and the current SOC, i.e., ΔSOC = SOC target -SOC(t), and based on the identified driving conditions, according to the relationship curve between the costate variable λ(t) and the change in SOC in step 4, the optimal costate variable λ(t) is obtained. At the same time, based on the speed prediction algorithm in step 5, the locomotive speed in the next 1s, 2s, 3s, 4s, and 5s is predicted and the average value and standard deviation of the speed are obtained, so as to realize the online correction of the Hamiltonian function.

[0116] To improve the durability of fuel cells, this invention addresses the issue of suppressing power fluctuations in the fuel cell by introducing a penalty term λ1 into the original Hamiltonian function to limit the output power fluctuations of the fuel cell system.

[0117]

[0118] In addition, to ensure the battery's ability to maintain its charge, i.e., the State of Charge (SOC) fluctuates around the target value throughout the entire driving cycle, a penalty term for maintaining SOC is introduced into the original Hamiltonian function:

[0119] ΔSOC=SOC target -SOC(t)

[0120] Its penalty weight is set as follows:

[0121]

[0122] In the formula, v std v is the standard deviation of the predicted velocity. ave This represents the average value of the predicted speed.

[0123] Step 7: Apply the optimal costate variables obtained in Step 4 to the Hamiltonian function to update its function body; that is:

[0124]

[0125] in, Hydrogen consumption rate of fuel cell system

[0126] Step 8: Minimize the Hamiltonian function to obtain the optimal power allocation between the fuel cell and the power battery, which can be applied to future applications. k +H P At any given time, the power required by the locomotive is rationally allocated to the fuel cell system and the power battery system to achieve energy management of the hybrid fuel cell locomotive. The optimal power allocation between the fuel cell and the power battery is as follows:

[0127]

[0128] In the formula, P fc P is the output power of the fuel cell system. b P is the output power of the power battery system. req This is the power required for the entire vehicle.

[0129] Steps 1 through 4 can be performed offline, while steps 5 through 8 can be performed online.

[0130] This invention discloses a real-time energy management method for fuel cell locomotives based on the Pontryagin minimum principle. The method comprises offline and online components. The offline component involves obtaining the globally optimal solution for historical driving conditions using the Pontryagin minimum principle, based on the locomotive's historical driving data. The online component involves obtaining the optimal costate variable based on SOC feedback and further refining the Hamiltonian function using speed prediction technology. This achieves online energy management of the fuel cell locomotive based on the Pontryagin minimum principle, balancing fuel economy, fuel cell durability, and the SOC maintenance capability of the power battery.

[0131] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A real-time energy management method for fuel cell locomotives, characterized in that, include: Obtain historical typical driving condition data of fuel cell hybrid electric vehicles, and construct Markov state transition probability matrices for speed and acceleration under different driving conditions; Obtain the current driving conditions and predict the vehicle speed based on the Markov state transition probability matrix; Based on the power system structure of the fuel cell hybrid locomotive, a model of the entire vehicle power system of the fuel cell locomotive is established; Based on the Pontryagin minimum principle, the historical typical driving condition data, and the vehicle power system model of the fuel cell hybrid locomotive, the relationship between costate variables and SOC changes is obtained offline by the target shooting method. Based on the difference between the target SOC and the current SOC and the current driving conditions, and by utilizing the relationship between the co-state variable and the change in SOC, the optimal co-state variable is obtained. Based on the predicted vehicle speed, the Hamiltonian function is corrected online, including: The average and standard deviation of the predicted vehicle speed are obtained based on the predicted vehicle speed; The Hamiltonian function is modified by introducing a penalty term to limit the output power fluctuation of the fuel cell system; The penalty for limiting the output power fluctuation of the fuel cell system is: ； in, To limit the penalty term for output power fluctuations in fuel cell systems, P fc For the output power of the fuel cell system, For a moment, This represents the minimum output power of the fuel cell system. This represents the minimum output power of the fuel cell system. The Hamiltonian function is modified based on the average value and the standard deviation, using a penalty term for maintaining SOC. The penalty term for maintaining SOC is: ； in, This is a penalty for maintaining SOC. The standard deviation of the predicted speed, This represents the average of the predicted speeds; The corrected Hamiltonian function is updated based on the optimal costate variables, as follows: ； in, For Hamiltonian functions, The difference between the target SOC and the current SOC. For the hydrogen consumption rate of the fuel cell system, The rate of change of SOC For optimal costate variables; The updated Hamiltonian function is minimized, and the optimal power allocation between the fuel cell and the power battery is obtained based on the fuel cell hybrid locomotive power system model to achieve real-time energy management of the fuel cell locomotive.

2. The real-time energy management method for fuel cell locomotives as described in claim 1, characterized in that, Obtaining the current driving conditions includes: Based on the historical typical driving condition data, a dataset is constructed, and the learning vector quantization neural network is trained using the dataset to obtain a driving condition recognition model. Real-time feature parameters of the fuel cell hybrid locomotive are extracted, and the current driving conditions are obtained by identifying the real-time feature parameters through the operating condition identification model.

3. The real-time energy management method for fuel cell locomotives as described in claim 2, characterized in that, Based on the aforementioned historical typical driving conditions, the dataset includes: The historical typical driving condition data is divided into several segments, and the feature parameters of each segment are extracted. The feature parameters include average vehicle speed, average acceleration, average deceleration, proportion of high-speed time, proportion of low-speed time, and proportion of constant speed time. The dataset is constructed using the feature parameters and the corresponding operating condition categories.

4. The real-time energy management method for fuel cell locomotives as described in claim 1, characterized in that, Constructing the Markov state transition probability matrices for speed and acceleration under different driving conditions includes: Extract acceleration and speed information from the historical typical driving condition data; The acceleration information and the velocity information are divided to obtain equally spaced acceleration state sequences and equally spaced velocity sequences, and a state number index matrix is ​​constructed. The velocity and acceleration states at each moment are numbered using the state number index matrix, where the position numbers of the velocity and acceleration states at the same moment constitute a state quantity. Construct a state transition frequency empty matrix using the state number index matrix; Based on the state transition frequency empty matrix, the state variables at two adjacent time points are obtained and 1 is incremented at the corresponding positions of the state transition frequency empty matrix to construct the state transition frequency matrix. Based on the state transition frequency matrix, construct the Markov state transition probability matrix; The method for constructing the Markov state transition probability matrix is ​​as follows: ； in, The state transition probability matrix is... Let be the element in the i-th row and j-th column of the state transition frequency matrix, where i and j are the state variables Fi and Fj corresponding to times t and t+1, respectively, and M and N are the discrete state numbers of acceleration and velocity, respectively.

5. The real-time energy management method for fuel cell locomotives as described in claim 4, characterized in that, Obtaining the equally spaced acceleration state sequence and equally spaced velocity sequence includes: Obtain the minimum and maximum acceleration values ​​from the acceleration information, and obtain the interval between the minimum and maximum acceleration values ​​as the first interval; The acceleration information is divided into an equally spaced acceleration state sequence with a length of a first interval by the first interval; Obtain the maximum speed from the speed information, and obtain the distance from the speed of zero to the maximum speed as the second distance; The velocity information is divided into equally spaced velocity sequences of a second length by the second spacing.

6. The real-time energy management method for fuel cell locomotives as described in claim 1, characterized in that, The fuel cell locomotive power system model includes a locomotive longitudinal dynamics model, a fuel cell hydrogen consumption and efficiency model, and a power battery system model. The required driving power of the whole vehicle, the output power of the fuel cell system, and the output power of the power battery are obtained through the locomotive longitudinal dynamics model, the fuel cell hydrogen consumption and efficiency model, and the power battery system model, respectively.

7. The real-time energy management method for fuel cell locomotives as described in claim 6, characterized in that, Based on the Pontryagin minimum principle, the historical typical driving condition data, and the fuel cell hybrid locomotive powertrain model, the relationship between costate variables and SOC changes is obtained offline using a target-shooting method, including: The total hydrogen consumption of the fuel cell is used as a cost function; Establish constraints on the output power of the fuel cell system, the output power of the power battery, and the range of SOC variation; Based on the Pontryagin minimum principle, a Hamiltonian function is established according to the cost function and the constraints. By solving the Hamiltonian function using different costate variables, the relationship between the costate variables and the change in SOC under historical driving conditions can be obtained.

8. The real-time energy management method for fuel cell locomotives as described in claim 6, characterized in that, The method for obtaining the optimal power allocation between the fuel cell and the power battery is as follows: ； In the formula, P fc P is the output power of the fuel cell system. b P is the output power of the power battery system. req For the power required by the whole vehicle, It is the Hamiltonian function; These are the optimal costate variables.

Images