A power distribution method based on a heterogeneous dual-fuel cell system

Abstract

A power allocation method based on a heterogeneous dual fuel cell system is proposed. This method combines multi-frequency separation of load power with an improved particle swarm optimization (IPSO) algorithm and model predictive control (MPC) to achieve synergistic optimization of system performance. Simulation results demonstrate that the invention fully utilizes the efficiency characteristics of the heterogeneous dual-stack design to achieve intelligent power allocation, thereby significantly reducing hydrogen consumption per 100 kilometers. Simultaneously, through frequency domain decoupling and MPC constraint optimization, the dynamic stress of the fuel cell is effectively mitigated, and the standard deviation of the power change rate is significantly reduced. This lays a solid foundation for extending component lifespan and provides important theoretical basis and feasible engineering solutions for the design of future long-life, high-efficiency multi-stack fuel cell systems.

Description

Technical Field

[0001] This invention relates to the field of dual fuel cell system technology, and in particular to a power distribution method based on a heterogeneous dual fuel cell system. Background Technology

[0002] Against the backdrop of a growing global energy crisis and environmental pollution, the world's automotive industry is undergoing a profound transformation towards low-carbon and electrification. In this process, fuel cell hybrid electric vehicles (FCHEVs), with their superior performance including high energy conversion efficiency, zero emissions, long driving range, and quick refueling, are widely recognized as one of the most promising technological pathways for building a sustainable future transportation system. A typical FCHEV powertrain usually uses a fuel cell system as the primary energy source, supplemented by energy storage units such as batteries or supercapacitors, aiming to overcome the inherent limitations of fuel cells, such as slow dynamic response and the inability to recover braking energy. In this hybrid architecture, the energy management system (EMS) serves as the core control center, and its design directly determines the vehicle's operating economy, power responsiveness, and the service life of key components (especially the fuel cell).

[0003] Extensive and in-depth research has been conducted on energy management issues in fuel cell-grade batteries (FCHEVs), leading to three main technical approaches: rule-based strategies, optimization theory-based strategies, and learning-based strategies. Among these, frequency-decoupling-based energy management methods map time-domain power demand to the frequency domain and optimize power allocation based on the dynamic response characteristics of each power source, thereby improving system efficiency while ensuring real-time performance. Some existing technologies integrate wavelet transform, adaptive filtering, and fuzzy control to achieve rational power allocation, effectively reducing hydrogen consumption and extending fuel cell lifespan, laying a solid foundation for the coordinated control of multi-energy systems. Furthermore, existing technologies utilize wavelet transform to decompose demanded power and allocate it to fuel cells, batteries, and supercapacitors, effectively improving system efficiency.

[0004] However, current technologies focus on single fuel cells or homogeneous multi-energy systems. Under complex and variable real-world operating conditions, load power exhibits characteristics of multi-frequency and wide-amplitude fluctuations. A single fuel cell stack cannot guarantee efficient operation under all conditions. Especially under high power demand, a single stack system may be forced to operate in an inefficient region or suffer from drastic power transients, which not only worsens the overall vehicle economy but also accelerates the irreversible degradation of the stack. Although some existing technologies use dual fuel cell stack architectures to improve power levels and system redundancy, most simplify multi-stack systems into homogeneous models, thus neglecting the potential for synergistic optimization using the heterogeneous characteristics between stacks (such as differences in rated power, efficiency, and health status). This fails to fully explore its potential for efficiency enhancement and energy saving, and makes it difficult to achieve the optimal balance between mitigating dynamic stress in fuel cells and reducing hydrogen consumption. Summary of the Invention

[0005] The purpose of this invention is to provide a power distribution method based on a heterogeneous dual fuel cell system, which significantly reduces system hydrogen consumption while effectively suppressing the internal dynamic stress of different fuel cell stacks, thereby achieving a simultaneous improvement in operating economy and system durability.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a power distribution method based on a heterogeneous dual fuel cell system, comprising the following steps: Step 1: Build a longitudinal dynamics model of the heterogeneous dual fuel cell hybrid electric vehicle in the MATLAB / Simulink environment, and calculate the required power of the vehicle using the longitudinal dynamics equations. : Equation (1); In equation (1), v is the vehicle velocity, dv / dt is the vehicle acceleration, m is the vehicle's curb mass, g is the acceleration due to gravity, and f is the vehicle's acceleration due to gravity. r C is the rolling resistance coefficient; d Where A is the air resistance coefficient; A is the vehicle's frontal area. α is air density; α is road gradient; δ is vehicle rotational mass conversion factor. For the mechanical efficiency of the transmission system; Step 2: Build a hybrid power system model based on the vehicle's longitudinal dynamics model, including the main fuel cell stack, auxiliary fuel cell stack, battery pack, and supercapacitor pack. The four parts are connected to a common DC bus through a DC / DC converter to provide power to the drive motor. Step 3: Employ a frequency decoupling strategy based on wavelet packet transform (WPT) to determine the required power of the entire vehicle. Decomposed into high-frequency components Intermediate frequency components and low frequency components , to high frequency components The intermediate frequency component is allocated to the supercapacitor bank. Allocate to the battery pack; Step 4: Use low-frequency components As the total target power of the fuel cell system A two-layer power allocation strategy is adopted to distribute low-frequency components. It is allocated to the main fuel cell stack and the auxiliary fuel cell stack; S4.1 The upper-level optimizer uses IPSO to improve the particle swarm optimization algorithm to obtain the main heap optimization power P. fc1 And auxiliary stack optimization power P fc2 ; The process of IPSO's improved particle swarm optimization algorithm is as follows: first, the population is initialized, the position and velocity of each particle are set, then the fitness value of each particle is calculated, and the individual optimal position Pbest and the global optimal position Gbest are iteratively updated in turn. During multiple iterations, non-linearly decreasing inertial weights and asynchronously changing learning factors are used. The expression for the non-linearly decreasing inertia weight w(t) is: Equation (2); In equation (2), and These are the initial and final inertia weights, respectively. This represents the current iteration number. The maximum number of iterations, This is the curve shape control coefficient; Learning factors include cognitive factors social factors The formula for how the two change asynchronously with the number of iterations is: Equation (3); When the global optimal position Gbest fails to update in multiple consecutive iterations, and the number of iterations reaches a preset threshold, a random mutation mechanism is used to randomly select multiple particles in the population for mutation. The positions and velocities of the selected particles are then randomly re-initialized in the solution space, and the iteration continues until convergence. The resulting main heap optimization power P is then output. fc1 And auxiliary stack optimization power P fc2 ; S4.2 The upper-level optimizer uses a multi-objective optimization function that minimizes hydrogen consumption and aging rate to optimize the main reactor power P. fc1 And auxiliary stack optimization power P fc2 Optimization was performed to obtain a steady-state power distribution reference value. and ; The expression for the multi-objective optimization function is: Equation (4); In equation (4), and For the weighting coefficients, + =1, This represents the total hydrogen consumption rate of the two stacks. This represents the total aging rate of the system. S4.3 The lower-level coordinator performs real-time coordinated control based on Model Predictive Control (MPC). The prediction model adopts a first-order discrete system, and its expression is: Equation (5); In equation (5), A is the identity matrix, B is the diagonal matrix related to the sampling period, and C is the identity matrix; State variables: , representing the actual output power of the fuel cell main stack and fuel cell auxiliary stack at time k; Control variables: , representing the power change applied to the main fuel cell stack and the auxiliary fuel cell stack at time k; S4.4 The lower-level coordinator, in each control cycle k, based on the current system state... Solving for the optimal control function in the finite time domain yields the future optimal control sequence; The expression for the optimal control function in the finite-time domain is: Equation (6); In equation (6), To predict the time domain, To control the time domain, Q and R are symmetric positive definite weight matrices; , representing the control sequence starting from time k; , representing the reference trajectory from the upper optimizer; S4.5 The lower-level coordinator obtains the first control variable from the future optimal control sequence. As a steady-state power reference command for dual fuel cell stacks and This coordinates the power distribution between the main fuel cell stack and the auxiliary fuel cell stack.

[0007] Preferably, multiple constraints are applied during the solution process of the optimal control function in the finite time domain, including the rate of climb constraint: ; Physical upper and lower limits constraints: ; Conservative constraints on thermal management requirements: when When the power command exceeds 90% of the rated power, the power command is changed to avoid the thermal risk zone.

[0008] According to the above technical solution, the beneficial effects of the present invention are: This invention designs an energy management strategy that integrates multi-frequency separation and two-layer optimization. The upper layer of the strategy combines wavelet packet transform and an improved particle swarm optimization algorithm (IPSO) to achieve accurate decomposition of load power and optimized steady-state power allocation for heterogeneous dual-reactor systems. The lower layer employs model predictive control (MPC) to ensure smooth power tracking of the dual-reactor system and strictly constrain its dynamic stress. MPC's ability to handle system constraints and predict future load changes offers significant advantages in energy management for fuel cell hybrid electric vehicles. Extensive simulation verification demonstrates that, compared to traditional isomorphic schemes, this invention can significantly reduce system hydrogen consumption while effectively mitigating the dynamic aging of fuel cells, providing a new technical path for achieving high economic efficiency and long-life operation of fuel cell vehicles. Attached Figure Description

[0009] Figure 1 A flowchart of a two-layer power allocation strategy; Figure 2 Flowchart for improving the particle swarm optimization algorithm for IPSO; Figure 3 A comparison of hydrogen consumption per 100 kilometers for three different comparison strategies; Figure 4 A comparison of the power output of fuel cell stacks; Figure 5 For comparison of power change rates; Figure 6 The result is the three-frequency decomposition of the required power; Figure 7 The power distribution results for the heterogeneous dual fuel cell system; Figure 8 The state of charge (SOC) changes of batteries and supercapacitors; Figure 9 This describes the power allocation of the energy storage system. Figure 10 To compare the tracking performance between theoretical power output and actual power output; Figure 11 The statistical characteristics of power balance error; Figure 12 The SOC (State of Charge) rate of change is the dynamic response characteristic of the energy storage system. Figure 13 These are the key performance indicators of the system. Detailed Implementation

[0010] This embodiment is a power distribution method based on a heterogeneous dual fuel cell system, including the following steps: Step 1: Build a longitudinal dynamics model of the heterogeneous dual fuel cell hybrid electric vehicle in the MATLAB / Simulink environment, and calculate the required power of the vehicle using the longitudinal dynamics equations. : Equation (1); In equation (1), v is the vehicle velocity, dv / dt is the vehicle acceleration, m is the vehicle's curb mass, g is the acceleration due to gravity, and f is the vehicle's acceleration due to gravity. r C is the rolling resistance coefficient; d Where A is the air resistance coefficient; A is the vehicle's frontal area. α is air density; α is road gradient; δ is vehicle rotational mass conversion factor. For the mechanical efficiency of the transmission system.

[0011] Step 2: Build a hybrid power system model based on the vehicle's longitudinal dynamics model, including the main fuel cell stack, auxiliary fuel cell stack, battery pack, and supercapacitor pack. The four parts are connected to a common DC bus through a DC / DC converter to provide power to the drive motor.

[0012] Step 3: Employ a frequency decoupling strategy based on wavelet packet transform (WPT) to determine the required power of the entire vehicle. Decomposed into high-frequency components Intermediate frequency components and low frequency components , to high frequency components The intermediate frequency component is allocated to the supercapacitor bank. Allocate to the battery pack.

[0013] The frequency decoupling strategy of wavelet packet transform (WPT) selects the Db4 wavelet basis, which performs well in signal processing, and considers the total power demand of the vehicle. A three-level wavelet packet decomposition method is implemented. The core of this method lies in deconstructing the complex power demand signal into a series of nodal coefficients in different frequency bands, and selectively reconstructing these coefficients to precisely allocate power commands to power sources with different dynamic response characteristics. Within this framework, the lowest frequency band (approximately 0-1Hz) signal, representing the baseline trend of power demand, is reconstructed as the power command for the fuel cell system. This ensures that the system operates within a stable and efficient range. Correspondingly, the power components reconstructed from the mid-frequency band (approximately 1-10Hz) and the high-frequency band (approximately 10Hz and above) coefficients serve as the initial power commands for the battery and supercapacitor, respectively. To ensure the operational safety and cycle life of the energy storage components, the power commands allocated to the battery and supercapacitor also need to be dynamically limited and compensated based on their real-time state of charge (SOC) to ensure they always operate within a preset healthy range. Through calculation and analysis, this decomposition strategy effectively decouples the total power demand into three distinct frequency bands: the low-frequency steady-state power corresponding to vehicle constant speed or slow acceleration is allocated to the fuel cell system for stable supply; the mid-frequency dynamic power corresponding to frequent start-stop and urban driving conditions is handled by the battery; and the high-frequency transient power rapidly absorbed or released by the supercapacitor is for emergency acceleration, regenerative braking energy recovery peaks, and other conditions.

[0014] Step 4: Use low-frequency components As the total target power of the fuel cell system A two-layer power allocation strategy is adopted to distribute low-frequency components. It is allocated to the main fuel cell stack and the auxiliary fuel cell stack.

[0015] Two-layer power allocation strategy such as Figure 1 As shown, it includes an upper-level optimizer and a lower-level coordinator: S4.1 The upper-level optimizer uses IPSO to improve the particle swarm optimization algorithm to obtain the main heap optimization power P. fc1 And auxiliary stack optimization power P fc2 .

[0016] IPSO improves the particle swarm algorithm, such as Figure 2 As shown, the algorithm flow is as follows: first, the population is initialized, the position and velocity of each particle are set, then the fitness value of each particle is calculated, and the individual optimal position Pbest and the global optimal position Gbest are iteratively updated in turn.

[0017] During multiple iterations, non-linearly decreasing inertial weights and asynchronously changing learning factors are employed.

[0018] The expression for the non-linearly decreasing inertia weight w(t) is: Equation (2); In equation (2), and These are the initial and final inertia weights, respectively. This represents the current iteration number. The maximum number of iterations, This is the curve shape control coefficient.

[0019] Maintain a large size in the early stages of iteration The value decreases slowly, which is beneficial for extensive global exploration; in the later stages of iteration, The value decreases rapidly, which is beneficial for the algorithm to perform fine local development and accelerate convergence to the global optimum.

[0020] Learning factors include cognitive factors social factors The formula for how the two change asynchronously with the number of iterations is: Equation (3).

[0021] In the initial stages of the search, set a larger [target]. and smaller This encourages individual particle exploration, increasing population diversity; in the later stages of the search, smaller [scales / sets] are used. and larger This encourages particles to converge toward the group's historical best position, enhancing the algorithm's local exploitation capability.

[0022] When the global optimal position Gbest fails to update in multiple consecutive iterations, and the number of iterations reaches a preset threshold, a random mutation mechanism is used to randomly select multiple particles in the population for mutation. The positions and velocities of the selected particles are then randomly re-initialized in the solution space, and the iteration continues until convergence. The resulting main heap optimization power P is then output. fc1 And auxiliary stack optimization power P fc2 .

[0023] Random mutation mechanisms can inject new diversity into the population when the algorithm stagnates, helping the particle swarm escape local optima traps and thus increasing the probability of finding the global optimum.

[0024] S4.2 The upper-level optimizer uses a multi-objective optimization function that minimizes hydrogen consumption and aging rate to optimize the main reactor power P. fc1 And auxiliary stack optimization power P fc2 Optimization was performed to obtain a steady-state power distribution reference value. and .

[0025] The expression for the multi-objective optimization function is: Equation (4); In equation (4), and For the weighting coefficients, + =1, This represents the total hydrogen consumption rate of the two stacks. This represents the total aging rate of the system.

[0026] S4.3 The lower-level coordinator performs real-time coordinated control based on Model Predictive Control (MPC). The prediction model adopts a first-order discrete system, and its expression is: Equation (5); In equation (5), A is the identity matrix, B is the diagonal matrix related to the sampling period, and C is the identity matrix.

[0027] State variables: , representing the actual output power of the fuel cell main stack and fuel cell auxiliary stack at time k.

[0028] Control variables: , representing the power change applied to the main fuel cell stack and the auxiliary fuel cell stack at time k.

[0029] S4.4 The lower-level coordinator, in each control cycle k, based on the current system state... Solving for the optimal control function in the finite time domain yields the optimal control sequence for the future.

[0030] The expression for the optimal control function in the finite-time domain is: Equation (6); In equation (6), To predict the time domain, To control the time domain, Q and R are symmetric positive definite weight matrices.

[0031] , representing the control sequence starting from time k.

[0032] , representing the reference trajectory from the upper optimizer.

[0033] In the process of solving the optimal control function in the finite time domain, various constraints are imposed, including the rate of climb constraint: ; Physical upper and lower limits constraints: ; Conservative constraints on thermal management requirements: when When the power command exceeds 90% of the rated power, the power command is changed to avoid the thermal risk zone.

[0034] S4.5 The lower-level coordinator obtains the first control variable from the future optimal control sequence. As a steady-state power reference command for dual fuel cell stacks and .

[0035] That is, the first control variable of the optimal control sequence is used: As the final power distribution control quantity and This coordinates the power distribution between the main fuel cell stack and the auxiliary fuel cell stack.

[0036] Moving to the next moment Subsequently, the system obtains the latest state information through sensor measurements or state estimation. Using this as a new initial condition, we can re-predict the system dynamics and solve a new optimization problem.

[0037] This cyclical online rolling calculation allows for continuous refinement of control decisions using the latest feedback information, significantly enhancing its robustness to model uncertainties and external disturbances. This mechanism ensures that the power output of the dual fuel cells is always precisely constrained within a safe and stable operating range, mitigating dynamic stress on the fuel cell stack and extending its lifespan.

[0038] Simulation Results and Analysis A complete vehicle and powertrain model was built in the MATLAB / Simulink simulation platform, and typical cyclic operating conditions were selected for testing.

[0039] Simulation environment and parameters The simulation was conducted in the MATLAB R2024b / Simulink environment. The key parameters of the vehicle are shown in Table 1. The simulation test was the China Light Vehicle Driving Test Condition (CLTC-P), which comprehensively reflects the driving characteristics of urban, suburban, and high-speed driving, and lasted for 1800 seconds.

[0040] Table 1. Parameters of Fuel Cell Hybrid Electric Vehicles Comparison of scheme design To scientifically verify the effectiveness of this invention, all comparative tests were conducted on the same heterogeneous dual fuel cell system (75kW main + 25kW auxiliary) hardware platform. By maintaining a fixed architecture and varying only the control strategy, a fair performance comparison was achieved. The designed comparison strategies include the following four: 1. Heterogeneous Efficiency-First Rule-Based (HE-RB) Strategy: Based on the measured efficiency-power characteristics of the main and auxiliary reactors, the power allocation point that minimizes the instantaneous hydrogen consumption of the system is selected under the total power demand at each moment, without long-term optimization or power change rate constraints.

[0041] 2. Heterogeneous Dynamic Programming (H-DP) based strategy: Under the condition of a known complete driving cycle, dynamic programming is used to solve the globally optimal power allocation sequence. The optimization objective is consistent with that of this invention (minimizing total hydrogen consumption and aging cost).

[0042] 3. IPSO steady-state strategy without MPC coordination (IPSO-Only): The upper layer adopts the same improved particle swarm optimization (IPSO) algorithm as the present invention and outputs steady-state power allocation instructions; the lower layer cancels MPC and only performs simple ramp rate limiting.

[0043] 4. The heterogeneous two-layer optimization strategy (IPSO-MPC) proposed in this invention: the upper layer uses IPSO for global power allocation optimization, and the lower layer uses MPC to achieve rolling optimization and feedback correction, taking into account both economy and durability.

[0044] Economic Results Analysis To quantitatively evaluate the economics of the proposed energy management strategy for the heterogeneous dual fuel cell system, Figure 3 Table 2 compares the hydrogen consumption (kg / 100km) and statistical characteristics of the four heterogeneous control strategies under the same test cycle. The results show that there are significant differences in economic indicators among the different strategies, verifying the effectiveness of the proposed method.

[0045] Table 2. Economic Indicators of Energy Management Strategies for Heterogeneous Dual Fuel Cell Systems The average hydrogen consumption per 100 km for the four strategies was ranked as follows: HE-RB > IPSO-Only > IPSO-MPC > H-DP. Among them, the H-DP strategy based on dynamic programming achieved the lowest hydrogen consumption (0.9200 ± 0.0500 kg / 100 km), establishing the theoretically optimal benchmark. The HE-RB rule-based strategy, serving as the baseline, had the highest hydrogen consumption (1.2000 ± 0.0800 kg / 100 km), and its large standard deviation reflects the limited adaptability of fixed-rule control to operating condition fluctuations and its unstable economic performance.

[0046] The IPSO-MPC hierarchical strategy of this embodiment exhibits excellent overall economic performance, with an average hydrogen consumption of 0.9500 kg / 100km and a standard deviation of 0.0400 kg / 100km. Compared with the baseline HE-RB strategy, IPSO-MPC reduces hydrogen consumption by 20.83%, a significant improvement. More importantly, compared with the IPSO-Only strategy (0.9800 ± 0.0600 kg / 100km) which only uses IPSO for steady-state optimization, IPSO-MPC achieves further performance improvement by introducing a model predictive control layer to compensate for dynamic errors in real time, reducing hydrogen consumption by approximately 3.06%. This indicates that the rolling optimization of the MPC layer effectively improves the overall energy efficiency of the system in actual dynamic operation.

[0047] From the perspective of data stability, the standard deviation of the IPSO-MPC strategy (0.0400 kg / 100km) is the smallest among the four strategies, even lower than that of the theoretically optimal H-DP strategy (0.0500 kg / 100km), indicating that it has better robustness and consistency under different operating scenarios. The relatively high standard deviation of the IPSO-Only strategy, on the other hand, reveals the potential performance fluctuations that may arise when relying solely on steady-state optimization in the face of dynamic operating conditions.

[0048] Although the IPSO-MPC strategy did not fully achieve the theoretical optimal value of H-DP, with a difference of approximately 3.26%, this gap is within an acceptable range for engineering applications. Considering that H-DP requires prior knowledge of complete operating conditions and is a non-causal, offline algorithm, it cannot be applied in real-time in actual vehicles. In contrast, IPSO-MPC is a fully causal, online, and feasible strategy. Therefore, IPSO-MPC sacrifices minimum economic optimization for real-time performance and practicality, thus approximating the theoretical optimal performance.

[0049] Economic comparative analysis shows that the IPSO-MPC hierarchical energy management strategy proposed in this invention, by integrating the global steady-state optimization of IPSO with the local dynamic adjustment of MPC, significantly reduces the system hydrogen consumption while ensuring the real-time feasibility of the algorithm, and exhibits good robustness. It is an effective strategy that combines excellent economy and engineering practicality.

[0050] Durability results analysis To assess the potential impact of different energy management strategies on the durability of fuel cell systems, a comparative analysis was conducted from two dimensions: output power smoothness and dynamic stress. Figure 4The power output characteristics of the four strategies in response to dynamic loads are visually demonstrated. The original load power (gray dashed line) exhibits significant high-frequency fluctuations and step changes. The HE-RB strategy, serving as the baseline, shows a clear step adjustment in its output curve, indicating poor smoothness. Although the output of the IPSO-Only strategy has been optimized, some mid-to-high-frequency fluctuations still remain. The theoretically optimal H-DP strategy and the IPSO-MPC strategy proposed in this invention both demonstrate excellent smoothing characteristics. Among them, the output curve of IPSO-MPC is the most stable, almost perfectly filtering out high-frequency disturbances in the load and only gently tracking low-frequency trends. This indicates that the MPC layer effectively plays the role of a "dynamic buffer" in real-time rolling optimization.

[0051] Dynamic stress is a key factor affecting the lifespan of fuel cell stacks and is usually quantified by the power change rate (dP / dt). Figure 5 The curves show that the dynamic stress applied to the fuel cell stack by different strategies varies by orders of magnitude. The HE-RB strategy has the highest average power change rate, reaching 6.80 kW / s. Its drastic power fluctuations will cause continuous mechanical and chemical shocks to key components such as the membrane electrode assembly and catalyst, severely damaging the stack's lifespan. The IPSO-Only strategy reduces this index to 5.20 kW / s, demonstrating that steady-state optimization itself has a certain mitigating effect. The IPSO-MPC strategy achieves the lowest dynamic stress level, with an average power change rate of only 2.70 kW / s, which is significantly reduced by 60.3% and 48.1% compared to HE-RB and IPSO-Only, respectively. It is worth noting that the dynamic stress of IPSO-MPC is even slightly lower than that of the theoretically optimal strategy H-DP (3.50 kW / s), which requires knowledge of global information, demonstrating the superiority of its online control strategy.

[0052] Durability analysis reveals the significant potential of the IPSO-MPC hierarchical strategy in extending system lifespan. Its core advantage lies in the fact that the upper-level IPSO algorithm is responsible for establishing the globally optimal power allocation benchmark, while the lower-level MPC, through high-frequency rolling optimization, compensates for dynamic errors in real time and actively suppresses power fluctuations. This hierarchical architecture of "global optimization + local smoothing" fundamentally reduces the dynamic stress on the fuel cell stack, thereby potentially significantly slowing down the performance degradation rate and reducing the total lifespan maintenance cost. This characteristic, combined with economic advantages, further demonstrates the comprehensive value of the IPSO-MPC strategy in achieving optimal lifespan cost for fuel cell vehicles.

[0053] Operating condition adaptability analysis To verify the effectiveness of the energy management strategy of the heterogeneous dual fuel cell three-source hybrid system based on wavelet packet transform and IPSO-MPC in the embodiment, system simulation was conducted using the China Light Vehicle Test Cycle (CLTC-P) test cycle. The CLTC-P test cycle includes three typical driving modes: urban, suburban, and highway, with a total duration of 1800 seconds, which can comprehensively reflect the dynamic characteristics of the vehicle in actual operation.

[0054] Figure 6 The results of the three-frequency decomposition of the demand power are shown: the low-frequency component (0-0.01Hz) changes smoothly, mainly reflecting the basic power demand of the vehicle; the mid-frequency component (0.01-0.1Hz) contains moderate dynamic changes, corresponding to the normal acceleration and deceleration process; the high-frequency component (>0.1Hz) fluctuates sharply, reflecting transient power impact.

[0055] This frequency domain decomposition method can effectively separate power fluctuations at different time scales, laying the foundation for subsequent power allocation. By allocating power components of different frequencies to the most suitable energy components, system efficiency can be optimized and component lifespan extended.

[0056] Figure 7 The power allocation results of the heterogeneous dual fuel cell system are demonstrated. The main fuel cell stack (75kW) bears the majority of the base load power, while the auxiliary fuel cell stack (25kW) dynamically adjusts its output power according to operating conditions. This allocation strategy fully considers the different efficiency and aging characteristics of the two stacks, maximizing system efficiency.

[0057] from Figure 7 As can be seen, during periods of low power demand, the main reactor operates alone to maintain a high efficiency range; during periods of medium to high power demand, the two reactors work together, achieving the optimal power allocation ratio through the IPSO optimization algorithm. This heterogeneous architecture has better adaptability to operating conditions and greater room for efficiency optimization compared to a homogeneous architecture.

[0058] Changes in the state of an energy storage system directly reflect the effectiveness of energy management strategies. Figure 8 The display shows the SOC changes of the battery and supercapacitor. The battery SOC changes steadily within the range of 55%-65%, demonstrating effective handling of mid-frequency power components; the supercapacitor SOC exhibits rapid fluctuation characteristics, effectively coping with high-frequency power surges.

[0059] Figure 9 The power distribution of the energy storage system is further illustrated. The battery primarily handles power fluctuations over longer periods, while the supercapacitor focuses on rapid response to transient power. This collaborative division of labor effectively reduces the dynamic stress on the fuel cell.

[0060] The power balance accuracy of a system is an important indicator for evaluating energy management strategies. Figure 10 The tracking performance of theoretical power output and actual power output was compared, and the two were basically consistent, indicating that the system has good power tracking capability. Figure 11 The statistical characteristics of the power balance error are shown in detail. The maximum error does not exceed ±2.5kW and the RMS error is 0.85kW. These indicators prove that the MPC real-time control strategy proposed in this invention has good control accuracy.

[0061] The power balance error mainly originates from factors such as the dynamic response delay of the fuel cell, power conversion efficiency loss, and sensor measurement errors. Through reasonable control strategy design, the error is controlled within an acceptable range, ensuring the stable operation of the system.

[0062] Figure 12 The dynamic response characteristics of the energy storage system were further revealed through SOC change rate analysis. The SOC change rate of the battery was relatively smooth, reflecting its ability to smoothly handle mid-frequency power components; the SOC change rate of the supercapacitor fluctuated dramatically, fully demonstrating its rapid response capability to high-frequency power components. This differentiated response characteristic verifies the effectiveness of frequency domain power decomposition.

[0063] Figure 13 The key performance indicators of the system are summarized. The total hydrogen consumption of the system is 0.328 kg, the system efficiency reaches 48.7%, the standard deviation of power fluctuation is 12.35 kW for the fuel cell, 8.72 kW for the battery, and 15.63 kW for the supercapacitor. These indicators fully demonstrate the advantages of the energy management strategy proposed in this invention in improving system efficiency and reducing component stress.

[0064] The results directly verify the effectiveness of the multi-frequency separation and bi-layer optimization strategy proposed in this invention, and the system's operating mode meets the expected design.

Claims

1. A power distribution method based on a heterogeneous dual fuel cell system, characterized in that, Includes the following steps: Step 1: Build a longitudinal dynamics model of the heterogeneous dual fuel cell hybrid electric vehicle in the MATLAB / Simulink environment, and calculate the required power of the vehicle using the longitudinal dynamics equations. : Equation (1); In equation (1), v is the vehicle velocity, dv / dt is the vehicle acceleration, m is the vehicle's curb mass, g is the acceleration due to gravity, and f is the vehicle's acceleration due to gravity. r C is the rolling resistance coefficient; d Where A is the air resistance coefficient; A is the vehicle's frontal area. α is air density; α is road gradient; δ is vehicle rotational mass conversion factor. For the mechanical efficiency of the transmission system; Step 2: Build a hybrid power system model based on the vehicle's longitudinal dynamics model, including the main fuel cell stack, auxiliary fuel cell stack, battery pack, and supercapacitor pack. The four parts are connected to a common DC bus through a DC / DC converter to provide power to the drive motor. Step 3: Employ a frequency decoupling strategy based on wavelet packet transform (WPT) to determine the required power of the entire vehicle. Decomposed into high-frequency components Intermediate frequency components and low frequency components , to high frequency components The intermediate frequency component is allocated to the supercapacitor bank. Allocate to the battery pack; Step 4: Use low-frequency components As the total target power of the fuel cell system A two-layer power allocation strategy is adopted to distribute low-frequency components. It is allocated to the main fuel cell stack and the auxiliary fuel cell stack; S4.1 The upper-level optimizer uses IPSO to improve the particle swarm optimization algorithm to obtain the main heap optimization power P. fc1 And auxiliary stack optimization power P fc2 ; The process of IPSO's improved particle swarm optimization algorithm is as follows: first, the population is initialized, the position and velocity of each particle are set, then the fitness value of each particle is calculated, and the individual optimal position Pbest and the global optimal position Gbest are iteratively updated in turn. During multiple iterations, non-linearly decreasing inertial weights and asynchronously changing learning factors are used. The expression for the non-linearly decreasing inertia weight w(t) is: Equation (2); In equation (2), and These are the initial and final inertia weights, respectively. This represents the current iteration number. The maximum number of iterations, This is the curve shape control coefficient; Learning factors include cognitive factors social factors The formula for how the two change asynchronously with the number of iterations is: Equation (3); When the global optimal position Gbest fails to update in multiple consecutive iterations, and the number of iterations reaches a preset threshold, a random mutation mechanism is used to randomly select multiple particles in the population for mutation. The positions and velocities of the selected particles are then randomly re-initialized in the solution space, and the iteration continues until convergence. The resulting main heap optimization power P is then output. fc1 And auxiliary stack optimization power P fc2 ; S4.2 The upper-level optimizer uses a multi-objective optimization function that minimizes hydrogen consumption and aging rate to optimize the main reactor power P. fc1 And auxiliary stack optimization power P fc2 Optimization was performed to obtain a steady-state power distribution reference value. and ; The expression for the multi-objective optimization function is: Equation (4); In equation (4), and For the weighting coefficients, + =1, This represents the total hydrogen consumption rate of the two stacks. This represents the total aging rate of the system. S4.3 The lower-level coordinator performs real-time coordinated control based on Model Predictive Control (MPC). The prediction model adopts a first-order discrete system, and its expression is: Equation (5); In equation (5), A is the identity matrix, B is the diagonal matrix related to the sampling period, and C is the identity matrix; State variables: , representing the actual output power of the fuel cell main stack and fuel cell auxiliary stack at time k; Control variables: , representing the power change applied to the main fuel cell stack and the auxiliary fuel cell stack at time k; S4.4 The lower-level coordinator, in each control cycle k, based on the current system state... Solving for the optimal control function in the finite time domain yields the future optimal control sequence; The expression for the optimal control function in the finite-time domain is: Equation (6); In equation (6), To predict the time domain, To control the time domain, Q and R are symmetric positive definite weight matrices; , representing the control sequence starting from time k; , representing the reference trajectory from the upper optimizer; S4.5 The lower-level coordinator obtains the first control variable from the future optimal control sequence. As a steady-state power reference command for dual fuel cell stacks and This coordinates the power distribution between the main fuel cell stack and the auxiliary fuel cell stack.

2. The power distribution method based on a heterogeneous dual fuel cell system according to claim 1, characterized in that: In the process of solving the optimal control function in the finite time domain, various constraints are imposed, including the rate of climb constraint: ; Physical upper and lower limits constraints: ; Conservative constraints on thermal management requirements: when When the power command exceeds 90% of the rated power, the power command is changed to avoid the thermal risk zone.

Images