Hybrid power city area motor train unit configuration and control deep collaborative optimization method

Abstract

The invention discloses a deep collaborative optimization method for configuration and control of a hybrid power city-area motor train unit, and the method comprises the steps: building constraint conditions, and constructing a constraint penalty model of a system; constructing a state-action pair value function of the system; under the physical constraint, a flexible action-evaluation algorithm based on direction guidance is adopted to estimate the value of a system state-action pair and reduce the strategy entropy, the magnitude of a combined value is constructed to evaluate the advantages and disadvantages of actions, and finally the optimal output power distribution scheme of the system at the moment is selected; analyzing a constraint condition satisfied by the system, system parameter configuration cost and an energy management control result by adopting a three-target non-dominated sorting genetic algorithm with a telescopic factor, and obtaining a parameter configuration non-dominated condition of the system according to a three-target non-dominated sorting rule; and finally, determining a final parameter matching scheme according to different expected targets. The performance coordination of the multi-stack fuel cell can be improved, the service life is effectively prolonged while the hydrogen consumption is reduced, the durability is improved, and the cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of hybrid urban EMUs, and in particular relates to a method for deep collaborative optimization of configuration and control of a hybrid urban EMU. Background Art

[0002] Proton exchange membrane fuel cells have been used in automobiles, trucks, shunting locomotives and other fields due to their advantages such as zero pollution, low noise and high efficiency. In order to solve the problem that fuel cells cannot recover braking energy and the energy conversion efficiency is low, fuel cells are usually equipped with batteries to form a fuel cell hybrid system to achieve flexible energy conversion. Nowadays, a complete fuel cell hybrid system usually includes a fuel cell system, a battery system, a unidirectional DC-DC converter, a bidirectional DC-DC converter, a signal collection system, and an energy management and control system. With the application of fuel cell hybrid systems in high-power scenarios (such as urban EMUs), hybrid systems with only a single fuel cell can no longer meet the high power requirements of system operation. Therefore, people connect multiple fuel cell monomers in parallel to form a multi-stack fuel cell hybrid system to solve the above problems. At present, the multi-stack fuel cell hybrid system has the following main defects that need to be further optimized:

[0003] The parameter matching process affects the energy management results: The traditional urban EMU parameter design process first determines the hybrid system parameter configuration based on the vehicle design parameters and line conditions according to the power and energy constraints, and then performs energy management. It is a sequential process, considers few constraints, ignores some practical factors, and does not consider the impact of parameter matching results on the energy management process. If the fuel cell or battery is too powerful, it may cause the output power of the fuel cell or battery to change frequently during the operation of the hybrid system, affecting its service life and reducing system stability.

[0004] Low energy conversion efficiency: Traditional multi-stack fuel cell systems distribute the required power of urban EMUs to each stack through average distribution or chain distribution strategies. These methods do not take into account the differences between each stack and cannot fine-tune the output power of a single stack. The performance degradation of the stack seriously affects its energy conversion efficiency. The energy conversion efficiency of the stack with severe performance degradation is low, which in turn leads to a low average energy conversion efficiency of the multi-stack fuel cell system.

[0005] Poor system durability: The short life and poor durability of fuel cells are one of the important reasons that hinder their large-scale application. The performance degradation of fuel cells is largely affected by their operating conditions. Excessive degradation of fuel cell performance will affect the safe and stable operation of the system. Therefore, it is necessary to model and analyze the performance degradation of fuel cells. Unreasonable operating conditions will cause a significant reduction in the life of the fuel cell, thereby reducing the durability of the system.

[0006] Poor internal balance of multi-stack fuel cell systems: Current research on multi-stack fuel cell systems focuses on the performance changes of each fuel cell within the system, and usually optimizes the system with the goal of reducing the performance degradation of a single fuel cell. It does not take into account that as a whole, excessive use of any fuel cell in the multi-stack fuel cell system or premature degradation of its life to the end will destroy the balance of the entire system, thereby causing system instability and affecting the safe operation of the system. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a hybrid urban EMU configuration and control deep collaborative optimization method, which considers the influence of the hybrid system power source configuration on the energy management control result, optimizes the parameter matching process and the multi-stack fuel cell hybrid system energy management control process as a whole, reduces the fuel cell performance attenuation while reducing the system fuel consumption, ensures the performance coordination of the multi-stack fuel cells, extends the service life of the multi-stack fuel cells, and improves the system durability. In addition, the method also helps to reduce the power source configuration cost.

[0008] To achieve the above-mentioned purpose, the technical scheme adopted by the present invention is: a method for deep collaborative optimization of configuration and control of hybrid urban EMUs, wherein the fuel cell hybrid power system for urban EMUs includes a multi-stack fuel cell system, a battery system, a DC bus, a signal collection system, and an energy management and control system; wherein the multi-stack fuel cell system is composed of a plurality of fuel cell monomers, each fuel cell monomer is connected to a unidirectional DC converter and then connected to the DC bus in parallel with the remaining monomers; the battery is connected to a bidirectional DC converter and then connected to the DC bus; the signal collection system is used to collect information such as output voltage and output current of multiple fuel cells and batteries, so as to provide a basis for subsequent energy management and control systems to calculate hydrogen consumption and performance attenuation of fuel cells and batteries; the energy management and control system is designed based on a DSP controller, and the corresponding energy management program is loaded to distribute the required power of the urban EMU to multiple fuel cells and batteries, and indirectly control the power output of each power source by controlling the corresponding DC converter;

[0009] Based on the above-mentioned fuel cell hybrid power system for urban EMU, the hybrid urban EMU configuration and control deep collaborative optimization method includes the following steps:

[0010] S100 analyzes the constraints that the EMU power system needs to meet based on the design parameters of the urban EMU and the basic conditions of the operating line, and builds a constraint penalty model for the system;

[0011] S200 analyzes and establishes a system hydrogen consumption model, a fuel cell and battery performance attenuation model, a multi-stack fuel cell system performance coordination evaluation model, a fuel cell output power fluctuation limitation model, and a battery state of charge constraint model, and constructs the system state-action value function based on the above models;

[0012] Under the physical constraints of system operation, S300 uses a flexible action-evaluation algorithm based on direction guidance to estimate the value of the system state-action pair and reduce the policy entropy. The value of the state-action pair and the policy entropy construct a combined value to evaluate the quality of the action, and finally selects the optimal output power allocation scheme of the system at that moment until the system operation process ends, completing the whole process energy management.

[0013] S400 uses a three-objective non-dominated sorting genetic algorithm with a scaling factor to analyze the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results. According to the three-objective non-dominated sorting rules, the system parameter configuration non-dominated situation is obtained, and finally the final parameter matching solution is determined according to different expected goals.

[0014] Furthermore, in step S100, the constraints of the urban EMU power system are analyzed to construct a constraint penalty model of the system, including the following steps:

[0015] S101, extracting parameters used for working condition construction from the design parameters of the urban EMU; extracting route parameters from the running route of the urban EMU; performing stress analysis on the urban EMU in operation, performing traction calculations on multiple special working conditions of the urban EMU according to the stress analysis results, obtaining the maximum power and maximum energy consumption under each working condition, and further comparing and obtaining the maximum power and maximum energy consumption required during the operation of the urban EMU;

[0016] S102, power and energy constraints: The output power and energy consumption of the hybrid system should meet the maximum power and maximum energy consumption required during the operation of the urban EMU;

[0017] S103, Mass and volume constraints The hybrid system should meet the axle weight and space volume requirements of the EMU;

[0018] S104, bus voltage constraint: ensure that the maximum operating voltage of the fuel cell is lower than the rated voltage of the DC bus; at the same time, the maximum output voltage of the battery is lower than the rated voltage of the DC bus;

[0019] S105, economic constraint: the hybrid power system purchased should not exceed the maximum budget;

[0020] S106, based on the analysis of the above constraints, the penalty value is increased by 1 for each constraint that the power system configuration fails to satisfy. When the power system configuration satisfies all constraints, the penalty value is 0. The smaller the penalty value, the better.

[0021] Furthermore, in step S200:

[0022] The hydrogen consumption model during the operation of the system is expressed as:

[0023]

[0024] in,

[0025] In the formula, C fc is the hydrogen consumption of the multi-stack fuel cell system, obtained by adding the hydrogen consumption of the fuel cell monomers; k is the charge state correction coefficient of the battery; C bat is the equivalent hydrogen consumption of the battery; n is the number of fuel cell monomers in the multi-stack fuel cell system; m fc is the hydrogen consumption of the multi-stack fuel cell system; m bat is the equivalent hydrogen consumption of the battery system; C fci is the fuel cell single hydrogen consumption;

[0026] The fuel cell performance attenuation model is:

[0027] D fc =D on / off +D low +D high +D chg ;

[0028] Among them, D on / off , D low , D high and D chg They represent the performance losses caused by start-stop conditions, high-power operation, low-power operation, and variable load conditions respectively;

[0029] The multi-stack fuel cell performance coordination evaluation model is:

[0030] m vs =k vs V s ;

[0031] in,

[0032]

[0033] In the formula, k VS D is the performance coordination coefficient of multiple fuel cell stacks; fci is the real-time attenuation degree of the i-th fuel cell; R fci_initrepresents the initial residual performance of the i-th fuel cell before it is put into use; L fc Coordinate the average degree of attenuation for fuel cell performance; V s is the coefficient of variation of the performance coordination of multiple fuel cells; σ is the standard deviation of the performance coordination of multiple fuel cells; μ is the expected attenuation of fuel cell performance coordination;

[0034] The fuel cell output power fluctuation limitation model is:

[0035]

[0036] Among them, k Dfc is the fuel cell performance attenuation coefficient; P fci,t is the fuel cell output power at time t; P fci,t-1 is the fuel cell output power at time t-1;

[0037] The battery state of charge constraint model is:

[0038] m soc =k soc (SOC-SOC init );

[0039] Among them, k soc SOC is the battery state of charge constraint coefficient; SOC represents the battery state of charge; SOC init Indicates the set battery target state of charge.

[0040] Furthermore, in order to quantitatively evaluate the performance degradation of the fuel cell, it is necessary to evaluate the remaining performance of the fuel cell. The remaining performance evaluation formula of the fuel cell is:

[0041]

[0042] Among them, ΔU rated_max It is the maximum voltage drop allowed when the fuel cell is running at rated current. It is defined that when the rated output voltage of the fuel cell drops by 10%, it is regarded as the end of its life. rated_degraded is the pressure drop caused by the performance degradation of the fuel cell; U rated_init is the rated voltage of the fuel cell.

[0043] Furthermore, the state-action pair value function of the system includes fuel cell and battery hydrogen consumption, fuel cell output power fluctuation constraint, battery state of charge constraint, and multi-stack fuel cell performance coordination constraint;

[0044] The comprehensive evaluation function of the system is:

[0045] J=m fc +m bat +mvs +m Dfc +m soc .

[0046] Furthermore, in step S300, a flexible action-evaluation algorithm based on direction guidance is used to estimate the value of the system state-action pair and reduce the policy entropy. The flexible action-evaluation algorithm based on direction guidance includes an actor network and four critic networks, namely, state value estimation v and Target v networks; action-state value estimation Q0 and Q1 networks; the input of the actor network is state s t , the output is the probability of the action π(a t |s t ), the input of the critic network is the state, the output of the vcritic network is the estimate of the state value v(s), and the output of the Qcritic network is the estimate of the action-state value q(s,a).

[0047] Furthermore, before the direction-guided flexible action-evaluation algorithm is applied to energy management, the network needs to be trained, including the following steps:

[0048] (1) Generate experience pool: a state s is known t , the probability of all actions is obtained through the actor network π(a|s t ), and then sample the action a according to probability t , then a t Input into the environment and get s t+1 and r t+1 , so you get an experience: (s t ,a t ,s t+1 ,r t+1 ), and then put the experience into the experience pool. When training the network, a batch of experience is selected from the experience pool to eliminate the strong correlation between the experiences.

[0049] (2) Qcritic network update: extract data from the experience pool (s t ,a t ,s t+1 ,r t+1 ) The Qcritic network is updated. Based on the optimal Bellman equation, the output of the Target vcritic network is used for the true value estimation. The mean square error loss is constructed together with the output of the Qcritic network before the update as the predicted value estimation as the loss function to train the Qcritic network;

[0050] (3) vcritic network update: extract data from the experience pool (s t ,at ,s t+1 ,r t+1 ) Update the vcritic network, use the entropy-containing formula to estimate the state value as the true value of the vcritic network output, use the output of the vcritic network as the predicted value to construct the mean square error loss as the loss function, and train the vcritic network;

[0051] (4) Actor network update: Combined strategy in state s t Take action a t The probability π(a t |s t ), action-state pair value q(s t ,a t ), temperature adjustment coefficient α, state value v(s t ) etc. to construct the combined value as the loss function to perform gradient descent training on the actor;

[0052] (5) Repeat the above steps until the network converges.

[0053] Furthermore, the flexible action-evaluation algorithm based on direction guidance guides the network training process through three behaviors: real-time data cleaning of the experience pool, evolutionary direction sorting, and adaptive temperature coefficient adjustment, including the following steps:

[0054] (1) Real-time data cleaning of the experience pool: Instead of processing the entire experience pool, the real-time extracted experience is processed. During the real-time training of the network, the selected experience for training needs to be identified and processed according to the constraints of the urban EMU operation process, and the data that does not meet the constraints is removed from the experience pool, and additional experience is extracted from the experience pool;

[0055] (2) Sorting by evolutionary direction: The important samples are sorted after normalization based on the error between the predicted value and the true value during sample training;

[0056] (3) Adaptive temperature adjustment coefficient: The temperature adjustment coefficient is outputted after fuzzy control using the average error of training experience, so that the temperature adjustment coefficient can be adaptively modified as the training process progresses.

[0057] Furthermore, the physical constraints of the system operation include maximum and minimum limits on the output power of the fuel cell and battery, maximum and minimum limits on the battery state of charge, the fuel cell output power change rate not exceeding 12.5% ​​of its rated power, and matching of the power source output power and the load power.

[0058] Furthermore, in step S400, the role of the scaling factor is to gradually tighten the restrictions on the parameter matching constraints as the iteration process gradually proceeds, which helps to expand the search range in the initial stage and find a solution that meets the constraints in the later stage of the iteration;

[0059] It is necessary to build a three-objective fast non-dominated sorting rule for the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results. Define a scaling factor ε. The maximum value of the Γ function is 7, so the initial value of ε is set to 6. The construction rule is:

[0060] (1) If the Γ function value of x1 in two parameter configurations x1 and x2 is less than or equal to the scaling factor ε, and the Γ function value of x2 is greater than the scaling factor ε, then x1 dominates x2;

[0061] (2) When the Γ function values ​​of x1 and x2 are both greater than the scaling factor ε, if the Γ function value of x1 is less than the Γ function value of x2, then x1 dominates x2;

[0062] (3) When the Γ function values ​​of x1 and x2 are both less than or equal to the scaling factor ε, the dominance of x1 and x2 is determined based on the system parameter configuration cost and the energy management control result: when there is a configuration between the two that makes the system parameter configuration cost and the energy management control result the same as the other and one of them is better, then the configuration dominates the other; if x1 dominates x2, then the relationship is expressed as:

[0063]

[0064] n represents the nth objective function; f1(x) and f2(x) represent the parameter configuration cost and energy management result of the system under the parameter configuration x, respectively.

[0065] The beneficial effects of adopting this technical solution are:

[0066] The present invention discloses a method for deep collaborative optimization of hybrid urban EMU configuration and intelligent control. The method fully considers the influence of hybrid power system power source configuration on energy management control results, optimizes the parameter matching process and the energy management control process of multi-stack fuel cell hybrid power system as a whole, and adopts a flexible action-evaluation algorithm based on direction guidance to estimate the value of system state-action pairs and reduce policy entropy. The value of the state-action pair and the size of the combined value constructed by the policy entropy are used to evaluate the quality of the action, select the optimal output power allocation scheme of the system, and complete the effective allocation of the power required by the urban EMU; a three-objective non-dominated sorting genetic algorithm with a telescopic factor is used to analyze the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results, and a three-objective non-dominated sorting rule is set. In the solution process, the telescopic factor is continuously reduced so that the system satisfies the constraints, and the non-dominated situation of the parameter configuration solution of the system is obtained. Finally, the final parameter matching scheme is determined according to different expected goals, and the deep collaborative optimization of parameter matching and energy management control of the fuel cell hybrid urban EMU is completed. This method can improve the performance coordination of multiple fuel cell stacks, effectively extend the life of the multiple fuel cell stack system while reducing hydrogen consumption, improve system durability, and also help reduce the configuration cost of the fuel cell hybrid power system.

[0067] The deep collaborative optimization method for hybrid urban EMU configuration and intelligent control proposed in the present invention constructs a minimum optimization target for the system state-action pair value function. This target comprehensively considers multiple system indicators, including system hydrogen consumption, fuel cell performance degradation, multi-stack fuel cell performance coordination, fuel cell output power fluctuation, and battery charge state, thereby achieving comprehensive optimization of system operation.

[0068] The present invention takes into account the integrity of the multi-stack fuel cell system and incorporates the coordination of the performance of the multi-stack fuel cells into one of the optimization goals, effectively reducing the situation where any fuel cell in the multi-stack fuel cell system is overused or its life is prematurely decayed to the end. The life of the multi-stack fuel cells reaches the end at the same time, which makes the life of the multi-stack fuel cell system no longer limited to any single stack, effectively extending the durability of the system and improving the stability of the system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic flow chart of a hybrid urban EMU configuration and control depth collaborative optimization method of the present invention;

[0070] Figure 2 Schematic diagram of a fuel cell hybrid power system for a city train in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings.

[0072] In this embodiment, see Figure 2 As shown, the present invention proposes a method for deep collaborative optimization of hybrid urban EMU configuration and control, and the fuel cell hybrid power system for the urban EMU includes a multi-stack fuel cell system, a battery system, a DC bus, a signal collection system and an energy management and control system; wherein the multi-stack fuel cell system is composed of a plurality of fuel cell monomers, each fuel cell monomer is connected to a unidirectional DC converter and then connected to the DC bus in parallel with the remaining monomers; the battery is connected to a bidirectional DC converter and then connected to the DC bus; the signal collection system is used to collect information such as output voltage and output current of the multi-stack fuel cells and batteries, so as to provide a basis for subsequent energy management and control systems to calculate the hydrogen consumption and performance attenuation of the fuel cells and batteries; the energy management and control system is designed based on a DSP controller, and the corresponding energy management program is loaded to distribute the required power of the urban EMU to the multi-stack fuel cells and batteries, and indirectly controls the power output of each power source by controlling the corresponding DC converter, taking into account the fuel economy and durability of the system.

[0073] This method fully considers the influence of the power source configuration of the hybrid power system on the energy management control results, and optimizes the parameter matching process and the energy management control process of the multi-stack fuel cell hybrid power system as a whole. Firstly, according to the design parameters of the urban EMU and the basic conditions of the running line, the constraints are analyzed to meet the system's power performance, safe operation and economic budget, and the constraint penalty model of the system is constructed through symbolic functions. The state-action pair value function of the system is established on the basis of fully considering the hydrogen consumption of fuel cells and batteries, fuel cell performance degradation, battery state of charge constraints, fuel cell output power fluctuation limits and the coordination of multi-stack fuel cell performance. The flexible action-evaluation algorithm based on direction guidance is used to estimate the value of the system state-action pair and reduce the policy entropy. The combination value constructed by the value of the state-action pair and the policy entropy is used to evaluate the pros and cons of the action. Finally, the optimal output power allocation scheme of the system is selected to realize energy management control. The three-objective non-dominated sorting genetic algorithm with a scaling factor is used to analyze the constraints satisfied by the system, the system parameter configuration cost and the energy management control results. The three-objective non-dominated sorting rule is set. In the solution process, the scaling factor is continuously reduced to make the system meet the constraints, and the non-dominated situation of the system parameter configuration solution is obtained. Finally, the final parameter matching scheme is determined according to different expected goals, and the parameter matching of the urban EMU and the deep collaborative optimization of the energy management control of the multi-stack fuel cell hybrid system are completed.

[0074] In addition, during the implementation of the proposed method, the operation of the fuel cell hybrid system is limited by the power source output characteristics and physical conditions, including the output power of the fuel cell and the battery, the range of the battery state of charge, the rate of change of the fuel cell power, and the matching of the power source output power and the load power. In the application of the proposed method, it is necessary to solve the problem on the basis of satisfying the above-mentioned constraints.

[0075] Based on the above-mentioned fuel cell hybrid power system for urban EMU, the hybrid urban EMU configuration and control deep collaborative optimization method is proposed. Figure 1 As shown, the steps include:

[0076] S100 analyzes the constraints that the EMU power system needs to meet based on the design parameters of the urban EMU and the basic conditions of the operating line, and builds a constraint penalty model for the system;

[0077] S200 analyzes and establishes a system hydrogen consumption model, a fuel cell and battery performance attenuation model, a multi-stack fuel cell system performance coordination evaluation model, a fuel cell output power fluctuation limitation model, and a battery state of charge constraint model, and constructs the system state-action value function based on the above models;

[0078] Under the physical constraints of system operation, S300 uses a flexible action-evaluation algorithm based on direction guidance to estimate the value of the system state-action pair and reduce the policy entropy. The value of the state-action pair and the policy entropy construct a combined value to evaluate the quality of the action, and finally selects the optimal output power allocation scheme of the system at that moment until the system operation process ends, completing the whole process energy management.

[0079] S400 uses a three-objective non-dominated sorting genetic algorithm with a scaling factor to analyze the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results. According to the three-objective non-dominated sorting rules, the system parameter configuration non-dominated situation is obtained, and finally the final parameter matching solution is determined according to different expected goals.

[0080] The specific operation of the deep collaborative optimization method of hybrid urban EMU configuration and intelligent control includes three steps.

[0081] The first step is to determine the fuel cell hybrid system parameter matching constraints, as follows:

[0082] S101, power and energy consumption calculation, extracting parameters used for working condition construction from the design parameters of the urban EMU, such as load, maximum continuous traction, maximum speed, etc.; extracting route parameters from the operating route of the urban EMU, such as slope, line length, maximum climbing distance, and average station spacing parameters; performing force analysis on the urban EMU in operation, and performing traction calculations on multiple special operating conditions of the urban EMU according to the force analysis results, to obtain the maximum power and maximum energy consumption under each operating condition, and further comparing to obtain the maximum power and maximum energy consumption required during the operation of the urban EMU;

[0083] S102, power and energy constraints: The output power and energy consumption of the hybrid system should meet the maximum power and maximum energy consumption required during the operation of the urban EMU;

[0084] The power and energy constraints that the hybrid system needs to meet are expressed as follows:

[0085]

[0086] n fc and n bat are the number of fuel cells and batteries, respectively, and are independent variables in the optimization process; P fc and P bat are the rated output power of the fuel cell and battery respectively; η fcDC , η batDC and η DC / AC They are the unidirectional DC / DC converter efficiency, bidirectional DC / DC converter efficiency and inverter efficiency; E bat Energy storage for battery cells; P loadmax and E loadmax They respectively represent the maximum power and maximum energy consumption required during the operation of the urban EMU.

[0087] S103, build mass and volume constraints: When building a hybrid power system for a city EMU, it is necessary to consider the limitations of actual conditions; the axle weight and internal space of the EMU are limited, so the hybrid power system should meet the axle weight and space volume requirements of the EMU;

[0088] The constraints on the mass and volume of the power source of the urban EMU are as follows:

[0089]

[0090] Additional mass K of the fuel cell mfc , Volume coefficient K Vfc ; Additional mass K of the battery mbat , Volume coefficient K Vbat ; The mass of the fuel cell monomer M fc and volume V fc; The mass of the battery cell M bat and volume V bat ; and the maximum mass M of the power system allowed by the EMU max , maximum volume V max .

[0091] S104, bus voltage constraint: ensure that the maximum operating voltage of the fuel cell is lower than the rated voltage of the DC bus; and at the same time, the maximum output voltage of the battery is lower than the rated voltage of the DC bus;

[0092] The bus voltage constraints of the urban EMU are as follows:

[0093]

[0094] U batmax and U fcmax are the maximum output voltages of the battery and fuel cell respectively.

[0095] S105, economic constraint: the hybrid power system purchased should not exceed the maximum budget;

[0096] The economic constraints of the urban EMU are as follows:

[0097] C tra <C sys ;

[0098] C tra and C sys are the cost of the purchased hybrid power system and the maximum purchase budget of the system respectively.

[0099] S106, determining the feasible domain of parameter matching: Based on the analysis of the above constraints, the penalty value is increased by 1 for each constraint condition that the power system configuration does not meet. When the power system configuration meets all the constraints, the penalty value is 0. The smaller the penalty value, the better.

[0100] The constraint penalty model of the system is constructed as follows:

[0101]

[0102] P sys 、E sys 、M sys 、V sys and C sys They represent the maximum required power, maximum energy consumption, maximum allowable mass, maximum allowable volume and maximum purchase budget of the system respectively; P tra 、E tra 、M tra 、V tra and C traThey represent the maximum required power, maximum energy consumption, maximum mass, maximum volume and purchase budget under the current power system configuration; U fcmax , U batmax and U bus They represent the maximum voltage of the fuel cell, the maximum voltage of the battery, and the bus voltage respectively.

[0103] The second step is to construct the state-action value function of the urban EMU and use the flexible action-evaluation algorithm based on direction guidance to obtain the optimal output power allocation plan of the system. The implementation process of this step is as follows:

[0104] (1) Constructing a hydrogen consumption model for the system: including the hydrogen consumption of the fuel cell and the equivalent hydrogen consumption of the battery. Since the hybrid system cannot be supplied with energy from the outside during operation, all energy must be obtained by consuming hydrogen through the fuel cell. Therefore, it is necessary to equate the electric energy consumption of the battery to the hydrogen consumption of the fuel cell. The hydrogen consumption model during the operation of the system is expressed as:

[0105]

[0106] in,

[0107] In the formula, C fc is the hydrogen consumption of the multi-stack fuel cell system, obtained by adding the hydrogen consumption of the fuel cell monomers; k is the charge state correction coefficient of the battery; C bat is the equivalent hydrogen consumption of the battery; n is the number of fuel cell monomers in the multi-stack fuel cell system; m fc is the hydrogen consumption of the multi-stack fuel cell system; m bat is the equivalent hydrogen consumption of the battery system; C fci is the hydrogen consumption of a fuel cell cell.

[0108] (2) Constructing a fuel cell performance degradation model: To achieve the control of fuel cell durability, it is first necessary to evaluate the fuel cell performance degradation in real time. The fuel cell performance degradation is caused by start-stop conditions, high power operation, low power operation and variable load conditions. The fuel cell performance degradation model is:

[0109] D fc =D on / off +D low +D high +D chg ;

[0110] Among them, D on / off , D low , D high and D chg They respectively represent the performance losses caused by start-stop conditions, high-power operation, low-power operation and variable load conditions.

[0111] (3) Constructing a fuel cell residual performance evaluation formula: Constructing a fuel cell residual performance evaluation formula: In order to quantitatively evaluate the performance degradation of the fuel cell, it is necessary to evaluate the residual performance of the fuel cell. The fuel cell residual performance evaluation formula is:

[0112]

[0113] Among them, ΔU rated_max It is the maximum voltage drop allowed when the fuel cell is running at rated current. It is defined that when the rated output voltage of the fuel cell drops by 10%, it is regarded as the end of its life. rated_degraded is the pressure drop caused by the performance degradation of the fuel cell; U rated_init is the rated voltage of the fuel cell.

[0114] (4) Constructing a multi-stack fuel cell performance coordination evaluation model: To control any of the multi-stack fuel cells from being overused or prematurely decaying to the end of their lifespan, so as to extend the service life of the multi-stack fuel cell system, the multi-stack fuel cell performance coordination evaluation model is:

[0115] m vs =k vs V s ;

[0116] in,

[0117]

[0118] In the formula, k VS D is the performance coordination coefficient of multiple fuel cell stacks; fci is the real-time attenuation degree of the i-th fuel cell; R fci_init represents the initial residual performance of the i-th fuel cell before it is put into use; L fc Coordinate the average degree of attenuation for fuel cell performance; V s is the coefficient of variation of the performance coordination of multiple fuel cells; σ is the standard deviation of the performance coordination of multiple fuel cells; μ is the expected attenuation of fuel cell performance coordination.

[0119] (5) Constructing a fuel cell output power fluctuation limitation model: Frequent fluctuations in fuel cell output power are an important cause of fuel cell performance degradation. Therefore, limiting the output power fluctuation of the fuel cell can effectively reduce the fuel cell performance degradation. The fuel cell output power fluctuation limitation model is:

[0120]

[0121] Among them, k Dfc is the fuel cell performance attenuation coefficient; P fci,t is the fuel cell output power at time t; Pfci,t-1 is the fuel cell output power at time t-1.

[0122] (6) Constructing a battery state of charge constraint model: The battery state of charge will affect the power performance of the system when it is running at high power. Keeping the battery state of charge in a small range is conducive to the system having sufficient power to support when facing possible high power operation. The battery state of charge constraint model is:

[0123] m soc =k soc (SOC-SOC init );

[0124] Among them, k soc SOC is the battery state of charge constraint coefficient; SOC represents the battery state of charge; SOC init Indicates the set battery target state of charge.

[0125] (7) Constructing the system state-action pair value function: Considering the overall optimization of the system, the system state-action pair value function includes the fuel cell and battery hydrogen consumption, fuel cell output power fluctuation constraint, battery charge state constraint and multi-stack fuel cell performance coordination constraint. The system state-action pair value function is:

[0126] J=m fc +m bat +m vs +m Dfc +m soc .

[0127] (8) Analysis and solution constraints: The fuel cell hybrid system is subject to physical constraints during actual operation, including the maximum and minimum limits on the output power of the fuel cell and battery, the maximum and minimum limits on the battery state of charge, the fuel cell output power change rate not exceeding 12.5% ​​of its rated power, and the matching of the power source output power and the load power. The relationship can be expressed as:

[0128]

[0129] (9) Train the network:

[0130] A flexible action-evaluation algorithm based on direction guidance is used to estimate the value of the system state-action pair and reduce the policy entropy. The flexible action-evaluation algorithm based on direction guidance includes an actor network and four critic networks, namely the state value estimation v and Target v networks; the action-state value estimation Q0 and Q1 networks; the input of the actor network is the state s t , the output is the probability of the action π(a t |st ), the input of the critic network is the state, the output of the vcritic network is the estimate of the state value v(s), and the output of the Qcritic network is the estimate of the action-state value q(s,a); in order to make the strategy more balanced and diversified, the concept of entropy is added to the direction-guided flexible action-evaluation algorithm to measure the uncertainty of the strategy.

[0131] Before the direction-guided flexible action-evaluation algorithm is applied to energy management, the network needs to be trained, including the following steps: (1) Generate an experience pool: a state s is known. t , the probability of all actions is obtained through the actor network π(a|s t ), and then sample the action a according to probability t , then a t Input into the environment and get s t+1 and r t+1 , so you get an experience: (s t ,a t ,s t+1 ,r t+1 ), and then put the experience into the experience pool. When training the network, a batch of experience is selected from the experience pool to eliminate the strong correlation between each experience; (2) Qcritic network update: take data from the experience pool (s t ,a t ,s t+1 ,r t+1 ) Update the Qcritic network. Based on the optimal Bellman equation, the output of the Target vcritic network is used to estimate the true value. Together with the output of the Qcritic network before the update as the predicted value estimate, the mean square error loss is constructed as the loss function to train the Qcritic network; (3) vcritic network update: take data (s) from the experience pool t ,a t ,s t+1 ,r t+1 ) Update the vcritic network, use the entropy-containing formula to estimate the state value as the true value of the vcritic network output, use the output of the v critic network as the predicted value to construct the mean square error loss as the loss function, and train the vcritic network; (4) actor network update: Combine the strategy in state s t Take action a t The probability π(a t |s t ), action-state pair value q(s t ,a t ), temperature adjustment coefficient α, state value v(s t) etc. construct the combined value as the loss function to perform gradient descent training on the actor; (5) repeat the above steps until the network converges.

[0132] The flexible action-evaluation algorithm based on direction guidance guides the network training process through three behaviors: real-time data cleaning in the experience pool, evolutionary direction sorting, and adaptive temperature coefficient adjustment. It ensures that the algorithm has good search capabilities while improving the convergence of the algorithm. The steps include:

[0133] ① Real-time data cleaning of the experience pool: In a specific system state, there are many experiences about the system state, and the data quality needs to be improved to ensure the accuracy and consistency of the data. In order to improve the speed and efficiency of data cleaning, instead of processing the entire experience pool, the real-time extracted experience is processed. During the real-time training of the network, the selected experience for training needs to identify and process the data that does not meet the constraints according to the constraints during the operation of the urban EMU, remove it from the experience pool, and extract additional experience from the experience pool to ensure sufficient training data;

[0134] ② Sorting of evolutionary directions: After the evolutionary direction is determined during the algorithm training process, it needs to be converged as quickly as possible to improve the convergence speed of the algorithm. To this end, it is necessary to adjust the algorithm's search capabilities and focus the algorithm's training process more on the convergence process. Therefore, based on the error between the predicted value and the true value during sample training, the important samples are normalized and sorted so that the experience with small errors has a greater probability of being adopted to accelerate network convergence;

[0135] ③ Adaptive temperature adjustment coefficient: The temperature adjustment coefficient is used to balance the algorithm's search ability and convergence ability. In the early stage of algorithm training, sufficient search ability needs to be guaranteed to find a more suitable convergence direction, and in the later stage, the convergence of the algorithm needs to be accelerated as much as possible. Therefore, a fixed temperature adjustment coefficient is not conducive to balancing the search ability and convergence ability. The temperature adjustment coefficient is output after fuzzy control using the average error of training experience, so that the temperature adjustment coefficient can be adaptively modified as the training process progresses.

[0136] (10) The flexible action-evaluation algorithm based on direction guidance is used to solve:

[0137] The third step is to deeply coordinate the optimization of parameter matching and energy management control of multi-stack fuel cell hybrid power systems. A three-objective non-dominated sorting genetic algorithm with a scaling factor is used to analyze the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results, and the non-dominated parameter configuration of the system is obtained. Finally, the final parameter matching scheme is determined based on different expected goals, such as focusing on energy management results or focusing on power source configuration costs or a weighted combination of the two as judgment criteria. The role of the scaling factor is to gradually tighten the restrictions on parameter matching constraints as the iterative process progresses, which helps to expand the search range at the beginning and find solutions that meet the constraints in the later stages of the iteration.

[0138] When solving the problem using a three-objective non-dominated sorting genetic algorithm with a scaling factor, it is necessary to construct a three-objective fast non-dominated sorting rule for the constraints satisfied by the system, the system parameter configuration cost, and the energy management control result, and define a scaling factor ε. Since the maximum value of the Γ function in claim 3 is 7, the initial value of ε is set to 6. The construction rule is as follows:

[0139] (1) If the Γ function value of x1 in two parameter configurations x1 and x2 is less than or equal to the scaling factor ε, and the Γ function value of x2 is greater than the scaling factor ε, then x1 dominates x2;

[0140] (2) When the Γ function values ​​of x1 and x2 are both greater than the scaling factor ε, if the Γ function value of x1 is less than the Γ function value of x2, then x1 dominates x2;

[0141] (3) When the Γ function values ​​of x1 and x2 are both less than or equal to the scaling factor ε, the dominance of x1 and x2 is determined based on the system parameter configuration cost and the energy management control result: when there is a configuration between the two that makes the system parameter configuration cost and the energy management control result the same as the other and one of them is better, then the configuration dominates the other; if x1 dominates x2, then the relationship is expressed as:

[0142]

[0143] n represents the nth objective function; f1(x) and f2(x) represent the parameter configuration cost and energy management result of the system under the parameter configuration x, respectively.

[0144] The invented hybrid urban EMU configuration and intelligent control deep collaborative optimization method can fully consider the impact of parameter matching results on energy management control results, comprehensively optimize the operation of the system, take into account the fuel economy and durability of the system, balance the differences within the multi-stack fuel cell system to extend the service life of the multi-stack fuel cell system, and reduce the operating cost of the system. In addition, the method also maintains the battery charge state in a small range of fluctuations, which is conducive to the system having sufficient power to support the operation of the system when facing possible high power demand.

[0145] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A hybrid urban EMU configuration and control deep collaborative optimization method, characterized in that: The fuel cell hybrid power system for urban EMUs includes a multi-stack fuel cell system, a battery system, a DC bus, a signal collection system, and an energy management and control system; wherein the multi-stack fuel cell system is composed of a plurality of fuel cell monomers, each fuel cell monomer is connected to a unidirectional DC converter and then connected to the DC bus in parallel with the remaining monomers; the battery is connected to a bidirectional DC converter and then connected to the DC bus; the signal collection system is used to collect information such as the output voltage and output current of the multi-stack fuel cells and batteries, and provide a basis for the subsequent energy management and control system to calculate the hydrogen consumption and performance attenuation of the fuel cells and batteries; the energy management and control system is designed based on a DSP controller, and the corresponding energy management program is loaded to distribute the required power of the urban EMU to the multi-stack fuel cells and batteries, and indirectly control the power output of each power source by controlling the corresponding DC converter; Based on the above-mentioned fuel cell hybrid power system for urban EMU, the hybrid urban EMU configuration and control deep collaborative optimization method includes the following steps: S100 analyzes the constraints that the EMU power system needs to meet based on the design parameters of the urban EMU and the basic conditions of the operating line, and builds a constraint penalty model for the system; S200 analyzes and establishes a system hydrogen consumption model, a fuel cell and battery performance attenuation model, a multi-stack fuel cell system performance coordination evaluation model, a fuel cell output power fluctuation limitation model, and a battery state of charge constraint model, and constructs the system state-action value function based on the above models; Under the physical constraints of system operation, S300 uses a flexible action-evaluation algorithm based on direction guidance to estimate the value of the system state-action pair and reduce the policy entropy. The value of the state-action pair and the policy entropy construct a combined value to evaluate the quality of the action, and finally selects the optimal output power allocation scheme of the system at that moment until the system operation process ends, completing the whole process energy management. S400 uses a three-objective non-dominated sorting genetic algorithm with a scaling factor to analyze the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results. According to the three-objective non-dominated sorting rules, the system parameter configuration non-dominated situation is obtained, and finally the final parameter matching solution is determined according to different expected goals.

2. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 1, characterized in that: In the step S100, the constraints of the urban EMU power system are analyzed to construct a constraint penalty model of the system, including the following steps: S101, extracting parameters for working condition construction from the design parameters of the urban EMU; extracting route parameters from the operating route of the urban EMU; Conduct stress analysis on the urban EMU in operation, perform traction calculations on multiple special operating conditions of the urban EMU based on the stress analysis results, obtain the maximum power and maximum energy consumption under each operating condition, and further compare and obtain the maximum power and maximum energy consumption required during the operation of the urban EMU; S102, power and energy constraints: The output power and energy consumption of the hybrid system should meet the maximum power and maximum energy consumption required during the operation of the urban EMU; S103, Mass and volume constraints The hybrid system should meet the axle weight and space volume requirements of the EMU; S104, bus voltage constraint: ensure that the maximum operating voltage of the fuel cell is lower than the rated voltage of the DC bus; at the same time, the maximum output voltage of the battery is lower than the rated voltage of the DC bus; S105, economic constraint: the hybrid power system purchased should not exceed the maximum budget; S106, based on the analysis of the above constraints, the penalty value is increased by 1 for each constraint condition that the power system configuration fails to meet. When the power system configuration meets all the constraints, the penalty value is 0. The smaller the penalty value, the better.

3. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 1, characterized in that: In step S200: The hydrogen consumption model during the operation of the system is expressed as: in, In the formula, C fc is the hydrogen consumption of the multi-stack fuel cell system, obtained by adding the hydrogen consumption of the fuel cell monomers; k is the charge state correction coefficient of the battery; C bat is the equivalent hydrogen consumption of the battery; n is the number of fuel cell monomers in the multi-stack fuel cell system; m fc is the hydrogen consumption of the multi-stack fuel cell system; m bat is the equivalent hydrogen consumption of the battery system; C fci is the fuel cell single hydrogen consumption; The fuel cell performance attenuation model is: D fc =D on / off +D low +D high +D chg ; Among them, D on / off , D low , D high and D chg They represent the performance losses caused by start-stop conditions, high-power operation, low-power operation, and variable load conditions respectively; The multi-stack fuel cell performance coordination evaluation model is: m vs =k vs In s ; in, In the formula, k VS D is the performance coordination coefficient of multiple fuel cell stacks; fci is the real-time attenuation degree of the i-th fuel cell; R fci_init represents the initial residual performance of the i-th fuel cell before it is put into use; L fc Coordinate the average degree of attenuation for fuel cell performance; V s is the coefficient of variation of the performance coordination of multiple fuel cells; σ is the standard deviation of the performance coordination of multiple fuel cells; μ is the expected attenuation of fuel cell performance coordination; The fuel cell output power fluctuation limitation model is: Among them, k Dfc is the fuel cell performance attenuation coefficient; P fci,t is the fuel cell output power at time t; P fci,t-1 is the fuel cell output power at time t-1; The battery state of charge constraint model is: m soc =k soc (SOC-SOC init ); Among them, k soc SOC is the battery state of charge constraint coefficient; SOC represents the battery state of charge; SOC init Indicates the set battery target state of charge.

4. A hybrid urban EMU configuration and control depth collaborative optimization method according to claim 3, characterized in that: In order to quantitatively evaluate the performance attenuation of the fuel cell, it is necessary to evaluate the remaining performance of the fuel cell. The remaining performance evaluation formula of the fuel cell is: Among them, ΔU rated_max It is the maximum voltage drop allowed when the fuel cell is running at rated current. It is defined that when the rated output voltage of the fuel cell drops by 10%, it is regarded as the end of its life. rated_degraded is the pressure drop caused by the performance degradation of the fuel cell; U rated_init is the rated voltage of the fuel cell.

5. A hybrid urban EMU configuration and control depth collaborative optimization method according to any one of claims 1-4, characterized in that: The system's state-action pair value function includes fuel cell and battery hydrogen consumption, fuel cell output power fluctuation constraints, battery state of charge constraints, and multi-stack fuel cell performance coordination constraints; The system comprehensive evaluation function is: J=m fc +m bat +m vs +m Dfc +m soc 。 6. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 1, characterized in that: In step S300, a flexible action-evaluation algorithm based on direction guidance is used to estimate the value of the system state-action pair and reduce the policy entropy. The flexible action-evaluation algorithm based on direction guidance includes an actor network and four critic networks, namely, state value estimation v and Target v networks; action-state value estimation Q0 and Q1 networks; the input of the actor network is the state s t , the output is the probability of the action π(a t |s t ), the input of the critic network is the state, the output of the vcritic network is the estimate of the state value v(s), and the output of the Qcritic network is the estimate of the action-state value q(s,a).

7. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 6, characterized in that: Before the direction-guided flexible action-evaluation algorithm is applied to energy management, the network needs to be trained, including the following steps: (1) Generate experience pool: a state s is known t , the probability of all actions is obtained through the actor network π(a|s t ), and then sample the action a according to probability t , then a t Input into the environment and get s t+1 and r t+1 , so you get an experience: (s t ,a t ,s t+1 ,r t+1 ), and then put the experience into the experience pool. When training the network, a batch of experience is selected from the experience pool to eliminate the strong correlation between the experiences. (2) Qcritic network update: extract data from the experience pool (s t ,a t ,s t+1 ,r t+1 ) The Qcritic network is updated. Based on the optimal Bellman equation, the output of the Target vcritic network is used for the true value estimation. The mean square error loss is constructed together with the output of the Qcritic network before the update as the predicted value estimation as the loss function to train the Qcritic network; (3) vcritic network update: extract data from the experience pool (s t ,a t ,s t+1 ,r t+1 ) Update the vcritic network, use the entropy-containing formula to estimate the state value as the true value of the vcritic network output, use the output of the vcritic network as the predicted value to construct the mean square error loss as the loss function, and train the vcritic network; (4) Actor network update: Combined strategy in state s t Take action a t The probability π(a t |s t ), action-state pair value q(s t ,a t ), temperature adjustment coefficient α, state value v(s t ) etc. to construct the combined value as the loss function to perform gradient descent training on the actor; (5) Repeat the above steps until the network converges.

8. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 6, characterized in that: The flexible action-evaluation algorithm based on direction guidance guides the network training process through three behaviors: real-time data cleaning in the experience pool, evolutionary direction sorting, and adaptive temperature coefficient adjustment. The steps include: (1) Real-time data cleaning of the experience pool: Instead of processing the entire experience pool, the real-time extracted experience is processed. During the real-time training of the network, the selected experience for training needs to be identified and processed according to the constraints of the urban EMU operation process, and the data that does not meet the constraints is removed from the experience pool, and additional experience is extracted from the experience pool; (2) Sorting by evolutionary direction: The important samples are sorted after normalization based on the error between the predicted value and the true value during sample training; (3) Adaptive temperature adjustment coefficient: The temperature adjustment coefficient is outputted after fuzzy control using the average error of training experience, so that the temperature adjustment coefficient can be adaptively modified as the training process progresses.

9. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 1, characterized in that: The physical constraints of the system operation include maximum and minimum limits on the output power of the fuel cell and battery, maximum and minimum limits on the battery state of charge, the fuel cell output power change rate not exceeding 12.5% ​​of its rated power, and matching of the power source output power and load power.

10. A hybrid urban EMU configuration and control deep collaborative optimization method according to claim 1, characterized in that: In step S400, the role of the scaling factor is to gradually tighten the restrictions on the parameter matching constraints as the iteration process gradually proceeds, which helps to expand the search range in the initial stage and find solutions that meet the constraints in the later stage of the iteration; It is necessary to build a three-objective fast non-dominated sorting rule for the constraints satisfied by the system, the system parameter configuration cost, and the energy management control results. Define a scaling factor ε. The maximum value of the Γ function is 7, so the initial value of ε is set to 6. The construction rule is: (1) If the Γ function value of x1 in two parameter configurations x1 and x2 is less than or equal to the scaling factor ε, and the Γ function value of x2 is greater than the scaling factor ε, then x1 dominates x2; (2) When the Γ function values ​​of x1 and x2 are both greater than the scaling factor ε, if the Γ function value of x1 is less than the Γ function value of x2, then x1 dominates x2; (3) When the Γ function values ​​of x1 and x2 are both less than or equal to the scaling factor ε, the dominance of x1 and x2 is determined based on the system parameter configuration cost and the energy management control result: when there is a configuration between the two that makes the system parameter configuration cost and the energy management control result the same as the other and one of them is better, then the configuration dominates the other; if x1 dominates x2, then the relationship is expressed as: n represents the nth objective function; f1(x) and f2(x) represent the parameter configuration cost and energy management result of the system under the parameter configuration x, respectively.

Images