A fuel cell drone energy management method and a computer-readable storage medium

Abstract

This invention provides an energy management method for a fuel cell unmanned aerial vehicle (UAV) and a computer-readable storage medium. The method includes the following steps: S1, hybrid power system state perception and load; S2, upper-level optimization layer generating the optimal reference trajectory; S3, multi-objective coordination and dynamic weight decision-making; S4, lower-level tracking layer implementing power allocation; and S5, constraint management and feedback correction. This invention decouples the computationally complex long-term multi-objective optimization problem from the real-time power allocation problem requiring rapid response by combining "upper-level global optimization + lower-level local tracking." The upper layer solves for the global reference trajectory at a lower frequency, relying on a simplified convex optimization model for efficient solution; the lower layer performs rolling tracking control at a high frequency. Under limited onboard computing resources, this approach simultaneously considers the global optimization potential and the system's real-time requirements, achieving an effective balance between optimization performance and real-time performance, improving the system's overall economy and durability, and enhancing its adaptability and robustness under various operating conditions.

Description

Technical Field

[0001] This invention belongs to the field of energy management technology for fuel cell hybrid power systems, specifically relating to an energy management method for fuel cell unmanned aerial vehicles and a computer-readable storage medium. Background Technology

[0002] As a crucial energy solution for unmanned aerial vehicles (UAVs), fuel cell hybrid power systems require energy management strategies that balance dynamic load demands with factors such as range, economy, and equipment lifespan. Existing methods primarily include rule-based, optimization-based, and predictive control strategies, each with its own limitations.

[0003] Rule-based control strategies (such as state machines and fuzzy logic) offer good real-time performance, but their reliance on fixed rules makes them ill-suited to the complex and ever-changing flight conditions of UAVs, preventing them from achieving global optimization and resulting in poor energy efficiency and durability. Optimization-based control strategies (such as dynamic programming) can achieve theoretical global optimization, but they depend on complete prior operating condition information and involve significant computational loads, limiting their application to offline analysis and benchmark setting, and making online applications impossible.

[0004] Model predictive control (MPC) provides a feasible framework for online energy management through rolling optimization and feedback correction. However, traditional MPC still faces significant challenges when applied to fuel cell drones: First, it is mostly a single-objective optimization (such as minimizing hydrogen consumption), making it difficult to simultaneously balance multiple objectives such as system economy, fuel cell durability, and battery life; second, the short prediction time domain limits its long-term optimization capability, while the long prediction time domain causes real-time issues; third, the operating conditions of drones are complex and variable, and fixed-parameter MPC strategies lack the ability to adaptively adjust to load fluctuations and energy state, resulting in insufficient robustness.

[0005] In addition, although some studies have attempted to introduce hierarchical control or intelligent algorithms, these often have complex structures, heavy computational burdens, or poor interpretability, making it difficult to meet the stringent requirements of airborne systems for reliability, lightweight design, and real-time performance.

[0006] Therefore, there is an urgent need to develop an energy management method that combines real-time performance with global optimization potential, can dynamically coordinate multi-objective conflicts, and adapts to uncertain operating conditions, so as to promote the overall improvement of fuel cell drone performance. Summary of the Invention

[0007] To address the aforementioned technical problems in the existing technology, this invention provides a fuel cell unmanned aerial vehicle (UAV) energy management method, comprising the following steps:

[0008] Step S1, Hybrid power system state awareness and load management:

[0009] The system acquires real-time state parameters of the UAV hybrid power system and, based on UAV flight motion control information and historical load data, predicts the load power demand sequence for a future time domain using a load condition fusion prediction model. Where k is the current time, To predict the time-domain step size;

[0010] Step S2: The upper optimization layer generates the optimal reference trajectory.

[0011] Establish a system that includes the cost of hydrogen consumption. Equivalent cost of fuel cell lifespan degradation and equivalent cost of lithium battery life degradation The economic objective function JE;

[0012] Inputting the load power demand sequence and the state parameters of the hybrid power system within the prediction time domain, and under the condition of satisfying system constraints, the optimization algorithm is used to solve online within the prediction time domain to obtain the globally optimal lithium battery state of charge reference trajectory. ;

[0013] Step S3, Multi-objective Coordination and Dynamic Weight Decision:

[0014] Construct a multi-objective optimization function that includes economic efficiency, fuel cell durability, and lithium battery health status. ;

[0015] A fuzzy logic algorithm is used to dynamically adjust the weight coefficients of each sub-objective in the multi-objective optimization function based on the real-time normalized load power and the real-time state of charge of the lithium battery.

[0016] Step S4, the lower tracking layer implements power allocation:

[0017] The Model Predictive Control (MPC) method is adopted, with the globally optimal lithium battery state of charge reference trajectory as the tracking target, the load power demand sequence in the predicted time domain as the feedforward information, and the multi-objective optimization function with dynamic weights as the optimization index. Rolling optimization and feedback correction are performed to solve and output the power control command of the fuel cell in real time to complete the power distribution.

[0018] Step S5, Constraint Management and Feedback Correction:

[0019] During the optimization process in steps S2 and S4, physical constraints of the system are continuously applied and satisfied; feedback correction is performed based on the deviation between the actual output and the predicted output of the system; and when the replanning trigger condition is met, step S2 is triggered to be re-executed, and a new optimal lithium battery state of charge reference trajectory is output to ensure the overall optimality of the system.

[0020] Furthermore, the state parameters of the UAV hybrid power system described in step S1 constitute a state vector. It includes at least: the state of charge (SOC) value of the lithium battery. Current output power of fuel cells or output current and DC bus voltage .

[0021] Furthermore, the load condition fusion prediction model described in step S1 is a hybrid predictor that combines feedforward motion information and feedback historical data, and its workflow is as follows:

[0022] ① Feedforward prediction channel: Receives real-time motion command sequences (such as target acceleration and climb angle) from the UAV flight control unit, and calculates the future short-time domain through a pre-calibrated UAV power consumption-dynamics mapping model. Internal load power reference prediction .

[0023] ② Feedback Prediction Channel: Receives historical load power time series data, inputs it into a trained time series prediction network (such as a GRU network), and outputs a longer future time domain. Load power trend forecast within .

[0024] ③ Adaptive fusion module: This module weights and fuses the outputs of the two channels in the overlapping prediction time domain to generate the final unified load prediction sequence. .

[0025] Fusion weights The following principles are followed: the closer the prediction time step is to the current time, the higher the weight of feedforward information; as the prediction time step lengthens, the weight of historical data trends gradually increases, that is, the weight adjustment satisfies... Follow Increases and decreases.

[0026] Furthermore, the economic objective function J mentioned in step S2 E It is the sum of the total expected operating costs within the prediction time domain, and its general formula is:

[0027]

[0028] Among them, the predicted hydrogen consumption cost in the time domain , This is the unit hydrogen price coefficient. Hydrogen consumption rate of fuel cells;

[0029] Predicting fuel cell lifespan degradation costs over the time domain , To convert the fuel cell life degradation into an equivalent factor for economic cost, its determination is based on fuel cell aging experiments or average replacement cycle data. It is a measure of instantaneous lifespan degradation and is strongly correlated with the operating state of the fuel cell. , The rate of change weighting coefficient. To represent the 0-1 variables for start-up and shutdown, This refers to the start / stop weighting coefficient. This represents the absolute value of the change in the fuel cell's output power.

[0030] Lithium battery life degradation cost ,in, To convert the lithium battery life degradation into an equivalent factor for economic cost, its determination is based on the lithium battery cycle life test or statistical life data. As a measure of instantaneous lifetime decay, Used to punish high-current charging and discharging stress, among which A stress acceleration factor greater than 1 This is the battery's rated current.

[0031] Specifically, fuel cell hydrogen consumption rate Its output power can be fitted by fuel cell experimental data. The function, , where the coefficient This was obtained by fitting the hydrogen consumption-power characteristic curve of the fuel cell; Determine data based on aging tests or average replacement cycles of fuel cells; Used to punish drastic changes in output power and frequent start-stop events.

[0032] Furthermore, the system power balance relationship The discrete integral dynamics model of lithium battery SoC, together with the economic objective function JE and physical inequality constraints such as fuel cell power limit, battery SoC working window, and charge / discharge current limit, constitute a constrained optimization problem with the future SoC trajectory as the decision variable.

[0033] To enable online engineering applications, the optimization problem is reasonably simplified: nonlinear components such as fuel cell hydrogen consumption characteristics and external characteristics are fitted into polynomial or linear relationships, so that, given a load prediction sequence, the aforementioned economic objective function... This can be primarily represented as a convex function (e.g., a quadratic form) concerning the future lithium-ion battery SoC sequence. Therefore, the "optimization algorithm" specifically refers to an algorithm suitable for quickly solving such sequential decision convex optimization problems online, including but not limited to sequential quadratic programming, quadratic programming, or efficient one-dimensional search algorithms (such as the golden section method), thereby achieving periodic solutions for the upper optimization layer under limited onboard computing resources.

[0034] Furthermore, a multi-objective optimization function integrating economic efficiency, fuel cell durability, and lithium battery health status is constructed. It consists of three linearly weighted sub-objective functions with clear physical meanings, and uses a fuzzy logic algorithm to dynamically adjust the weight of each sub-objective according to the real-time flight load and battery energy status, so as to achieve adaptive switching of optimization focus.

[0035] Specifically, the multi-objective optimization function mentioned in step S3 ,in, Let the economic objective function be... For the sake of fuel cell durability, The cost to the health of lithium batteries;

[0036]

[0037] In the formula, The weighting factor for high output current. The weighting coefficient for the drastic rate of change of current;

[0038]

[0039] In the formula, For ideal state of charge The weighting coefficients, This is a weighting factor for high charge / discharge currents.

[0040] in, The immediate economic cost is directly related to the current hydrogen consumption, and its expression is consistent with or a simplified form of the local economic cost defined in step S2. This function is used to suppress operating modes that are detrimental to lifespan by penalizing high output current ( (as weighting coefficients) and drastic current change rate ( (as a weighting coefficient) to smooth fuel cell output and reduce its performance degradation. This function is used to maintain the battery within its ideal operating range and limit stress by penalizing the ideal state of charge. deviation ( (as weighting coefficients) and high charge / discharge current ( (as a weighting factor) to protect the battery and extend its cycle life.

[0041] Furthermore, the phrase "using fuzzy logic algorithm based on real-time load and battery" is used. "Dynamically adjusting the weights of each objective" specifically refers to:

[0042] ①System Construction: Establish a fuzzy inference system with two inputs and three outputs.

[0043] ② Input definition:

[0044] Input 1: Normalized real-time load power ,in This represents the system's rated power. Its domain covers the typical operating range, for example, [0, 1.2].

[0045] Input 2: Current state of charge of the lithium battery Its domain of discourse covers the permitted working range. .

[0046] ③ Output definition: Dynamic weight coefficients corresponding to the three sub-objective functions

[0047] As output, the universe of discourse for each weight coefficient is typically [0, 1], and must satisfy a certain normalization relation ( ).

[0048] ④ Rule-making and reasoning:

[0049] Based on a deep understanding of the system's operating characteristics (such as expert knowledge) or a large amount of offline simulation data, a fuzzy rule base is developed to describe the mapping relationship between inputs and outputs. The rule format is: "IF ( is A) AND (SoC isB), THEN ( is C, is D, "is E"), where A and B are input fuzzy subsets, and C, D, and E are output fuzzy subsets.

[0050] The core logic of the rules aims to make the strategy adaptable to different operating conditions: for example, during high load or climb phases, the focus is on system dynamics and economy (improving efficiency). When the lithium battery SoC is low, the focus is on battery protection and energy recovery (improvement). When the load is stable and the SoC is moderate, the focus is on smoothing fuel cell operation to improve its durability (enhancing...). ).

[0051] In each control cycle, the precise input value is obtained through fuzzy inference (such as the Mamdani method) and defuzzification (such as the center-of-gravity method). Mapped to precise weight values This enables online adaptive adjustment of multiple objective priorities.

[0052] Furthermore, the Model Predictive Control (MPC) method described in step S4 employs a quadratic programming solver to solve the rolling optimization problem online quickly. The specific process is as follows:

[0053] Build optimization metrics Includes tracking the upper layer. The deviation term of the reference trajectory is incorporated into a multi-objective optimization function. ;

[0054] An efficient solver is used for online calculation to obtain the optimal fuel cell power command at the current moment. And send it to the power actuator;

[0055] Using an internal prediction model, based on and predicted load Calculate the predicted system state value for the next moment, which will be used for feedback correction in subsequent control cycles.

[0056] This invention employs a model predictive control method to track the optimal output of the upper layer. Based on the reference trajectory, the fuel cell output power is adjusted in real time through rolling optimization and feedback correction to complete the power distribution.

[0057] Furthermore, step S4 is implemented through a model prediction controller, which includes the following core elements and technical means:

[0058] ① Internal prediction model: Establish a simplified prediction model that can reflect the key dynamics of the system to predict the impact of control actions on the future state of the system (mainly lithium battery SoC).

[0059] The model can be a discrete state-space model or a transfer function model obtained based on mechanism analysis or system identification.

[0060] A typical simplified form focuses on the energy balance of the lithium battery, modeling the dynamic changes of the SoC as the integral process of its power input. For example, in discrete time, it can be expressed as:

[0061]

[0062] in, The charge / discharge efficiency coefficient. The rated capacity of the lithium battery, To control the cycle, the model explicitly establishes the fuel cell output power. The dynamic relationship between (control variables) and battery SoC (state variables).

[0063] The model may also include a first-order dynamic element that reflects the inertia of the fuel cell output power change, in order to constrain its rate of change.

[0064] ② Rolling optimization: at each control moment :

[0065] Problem formulation: To track the optimal SoC reference trajectory in the finite time domain provided by the upper layer. The primary objective is to simultaneously incorporate the multi-objective function output from step S3, which integrates dynamic weights. As optimization terms, they together constitute the objective function. .

[0066] Optimization solution: in the future finite control time domain Within the system, a set of optimal fuel cell power control command sequences is solved. This optimization problem is carried out under the simplified prediction model and system physical constraints, and can generally be expressed as a minimization problem of a quadratic objective function.

[0067] Command output: The first element of the obtained control sequence. This serves as the actual control command output at the current moment.

[0068] ③ Feedback correction: In At each time step, the actual output of the measurement system (such as the SoC) is compared with the model prediction value of the previous time step. The output error is used to compensate for subsequent predictions, so as to correct model mismatch and suppress unknown disturbances, thus forming closed-loop control.

[0069] Furthermore, the model predictive controller can be implemented using specific methods such as model algorithm control (MAC), generalized predictive control (GPC), or predictive control based on a state-space model. Since rolling optimization problems can typically be constructed as convex optimization problems (such as quadratic programming), their solvers can employ efficient quadratic programming algorithms to meet the high real-time requirements of the lower-level control.

[0070] Furthermore, the physical operating constraints mentioned in step S5 include: upper and lower limits of fuel cell output power and its rate of change, upper and lower limits of lithium battery state of charge operating range, upper and lower limits of lithium battery charging and discharging current, and system power balance constraints.

[0071] The feedback correction mentioned in step S5 includes the immediate correction of output error by the lower-level MPC and the replanning trigger mechanism;

[0072] The replanning trigger mechanism includes setting a deviation threshold. When the actual Trajectory and upper-level optimal reference trajectory The deviation continues to exceed If the mean square error of the load prediction exceeds a certain threshold, the mechanism will trigger the upper optimization layer to re-execute step S2 and generate a new global reference trajectory based on the latest system state and prediction information.

[0073] Specifically, step S5 defines a global mechanism to ensure the safe operation of the system and the robustness of the strategy:

[0074] ① Full-process constraint management: The system's physical constraints are explicitly handled as hard constraints or soft constraints with severe penalties in both optimization stages S2 (upper-level optimization) and S4 (lower-level tracking) of the entire strategy. These constraints are uniformly expressed as:

[0075]

[0076]

[0077]

[0078]

[0079] ② Hierarchical Feedback and Adaptation: Feedback correction is not only reflected in the immediate correction of output errors by the lower-level MPC, but also defines a replanning triggering mechanism for the upper-level reference trajectory. Specifically, a deviation threshold is set. When the actual Trajectory and upper-level optimal reference trajectory The deviation continues to exceed If the mean square error of the load prediction exceeds a certain threshold, the mechanism will trigger the upper optimization layer to re-execute step S2, generate a new global reference trajectory based on the latest system state and prediction information, thereby ensuring the overall optimality of the strategy during long-term operation and when facing significant operating condition deviations.

[0080] The beneficial effects of this invention are as follows:

[0081] 1. Achieving an effective balance between performance optimization and real-time performance: This invention decouples the computationally complex long-term multi-objective optimization problem from the real-time power allocation problem requiring rapid response through a hierarchical structure of "upper-layer global optimization + lower-layer local tracking". The upper layer solves the global reference trajectory at a lower frequency (e.g., several seconds) and relies on a simplified convex optimization model (e.g., quadratic programming) for efficient solution; the lower layer performs rolling tracking control at a higher frequency (e.g., per second). This structure allows for both global optimization potential and system real-time requirements to be considered within limited onboard computing resources.

[0082] 2. Improved overall system economy and durability: This invention, for the first time in UAV energy management, unifies the modeling of hydrogen consumption costs, fuel cell lifespan degradation costs, and lithium battery lifespan degradation costs into a single comprehensive economic objective function. Furthermore, the equivalent coefficients of each cost item are determined based on experimental data or lifespan statistics, making the model reasonable. Combined with a multi-objective dynamic coordination mechanism, the strategy can intelligently balance the objectives of "immediate energy saving" and "long-term lifespan extension" under different flight phases and load conditions, thereby effectively reducing the total lifespan cost.

[0083] 3. Enhanced adaptability and robustness of the strategy: A multi-objective weight online adjustment mechanism based on fuzzy logic is introduced. This mechanism automatically switches optimization priorities based on real-time load intensity and battery energy status, adhering to core rules such as "high load prioritizing economy, low SoC prioritizing battery protection, and stable operating conditions prioritizing smooth fuel cell operation." This enables the management strategy to dynamically adapt to complex and uncertain flight conditions, significantly improving overall robustness.

[0084] 4. Clear structure and strong engineering applicability: The "perception-optimization-coordination-tracking-correction" framework proposed in this invention has a clear structure and well-defined functions for each module. The model predictive control, fuzzy logic, and convex optimization methods used are mature and reliable. Furthermore, the online computational complexity is reduced through model simplification (such as fuel cell hydrogen consumption fitting and battery SoC integral model), eliminating the need for extensive offline data training and facilitating engineering applications and deployment on UAV onboard computing platforms. Attached Figure Description

[0085] Figure 1 This is a flowchart of the core steps of the method of the present invention;

[0086] Figure 2 This is a diagram of the energy management architecture of the method of the present invention. Detailed Implementation

[0087] The following specific embodiments illustrate the implementation of the invention. Those skilled in the art can easily understand other advantages and effects of the invention from the content disclosed in this specification.

[0088] Exemplary embodiments of the invention are now described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention.

[0089] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0090] Example 1

[0091] This embodiment provides an energy management method for fuel cell drones, including the following steps:

[0092] Step S1, Hybrid power system state awareness and load management:

[0093] The system acquires real-time state parameters of the UAV hybrid power system and, based on UAV flight motion control information and historical load data, predicts the load power demand sequence for a future time domain using a load condition fusion prediction model. Where k is the current time, To predict the time-domain step size;

[0094] Step S2: The upper optimization layer generates the optimal reference trajectory.

[0095] Establish a system that includes the cost of hydrogen consumption. Equivalent cost of fuel cell lifespan degradation and equivalent cost of lithium battery life degradation The economic objective function JE;

[0096] Inputting the load power demand sequence and the state parameters of the hybrid power system within the prediction time domain, and under the condition of satisfying system constraints, the optimization algorithm is used to solve online within the prediction time domain to obtain the globally optimal lithium battery state of charge reference trajectory. ;

[0097] Step S3, Multi-objective Coordination and Dynamic Weight Decision:

[0098] Construct a multi-objective optimization function that includes economic efficiency, fuel cell durability, and lithium battery health status. ;

[0099] A fuzzy logic algorithm is used to dynamically adjust the weight coefficients of each sub-objective in the multi-objective optimization function based on the real-time normalized load power and the real-time state of charge of the lithium battery.

[0100] Step S4, the lower tracking layer implements power allocation:

[0101] The Model Predictive Control (MPC) method is adopted, with the globally optimal lithium battery state of charge reference trajectory as the tracking target, the load power demand sequence in the predicted time domain as the feedforward information, and the multi-objective optimization function with dynamic weights as the optimization index. Rolling optimization and feedback correction are performed to solve and output the power control command of the fuel cell in real time to complete the power distribution.

[0102] Step S5, Constraint Management and Feedback Correction:

[0103] During the optimization process in steps S2 and S4, physical constraints of the system are continuously applied and satisfied; feedback correction is performed based on the deviation between the actual output and the predicted output of the system; and when the replanning trigger condition is met, step S2 is triggered to be re-executed, and a new optimal lithium battery state of charge reference trajectory is output to ensure the overall optimality of the system.

[0104] This invention combines long-term global optimization with short-term real-time tracking through a hierarchical structure of "upper-level global optimization + lower-level local tracking" and introduces a multi-objective dynamic optimization mechanism. This enables comprehensive optimization of system endurance, operational economy and durability of key components under limited airborne computing resources, thereby improving the overall performance and mission adaptability of the UAV.

[0105] Example 2

[0106] Based on Example 1, this example uses a fuel cell drone as an application scenario to illustrate the specific implementation process of the method of the present invention.

[0107] 1. System Configuration and Initialization

[0108] The method of this invention is deployed on the onboard computing unit of an unmanned aerial vehicle (UAV) to manage a hybrid power system consisting of a fuel cell (main power source) and a lithium battery (auxiliary energy storage). Parameter initialization is required before implementation.

[0109] ① Economic model parameters: including hydrogen price coefficient Fuel cell lifespan degradation equivalent cost coefficient Lithium battery life degradation equivalent cost coefficient These coefficients convert hydrogen consumption, the physical lifespan degradation of fuel cells and lithium batteries into a unified economic cost, with specific values ​​determined based on hydrogen market prices and the average replacement cycle of components obtained through aging experiments or long-term statistics.

[0110] ② Fuel cell performance model parameters: These are parameters obtained by fitting experimental data of the fuel cell and describing its hydrogen consumption rate. With output power The coefficients of the functional relationship between them. For example, using a quadratic function. When performing fitting, the coefficients that need to be initialized are... , , .

[0111] ③ System operation constraint boundary values: determined according to the hardware safety specifications of the hybrid power system, including: upper and lower limits of fuel cell power. and power change rate limit lithium batteries Allow working window Maximum charging current of lithium battery With maximum discharge current .

[0112] ④ Control and Prediction Parameters: Set the basic control cycle (e.g., 1 second), the call cycle of the upper-level global optimization layer (usually several times longer). Predicted time domain length (e.g., 10 steps), control the length of the time domain (e.g., 3 steps) to match the needs of a hierarchical control architecture.

[0113] ⑤ Intelligent Decision Parameters: Load the membership functions and rule base preset by the fuzzy logic controller. Specifically, this includes defining the normalized load. and lithium batteries Input universe of discourse and fuzzy subset; definition of economic weights Fuel cell durability weight Lithium battery health weight The output domain of discourse; and based on the principle that "high load emphasizes economy, low load..." Based on the core logic of "emphasizing battery protection and smooth operation of fuel cells under stable conditions", fuzzy reasoning rules were formulated.

[0114] 2. Execute the steps using a flowchart.

[0115] The following combination Figure 1 and Figure 2 The flowchart shown illustrates the specific implementation of each step within a complete control cycle:

[0116] Step S1: System Status Awareness and Load Prediction

[0117] Data acquisition: At the beginning of each control cycle, real-time state vectors are acquired through a sensor network. It includes at least the state of charge of the lithium battery. and the output power of fuel cells .

[0118] Information fusion: Receives real-time motion commands (such as acceleration and climb angle) from the flight control system.

[0119] Prediction calculation:

[0120] a) Feedforward prediction: Input the motion command into a pre-calibrated motion-power consumption mapping model to calculate the future... Step reference load power sequence .

[0121] b) Time Series Prediction: Input a window of historical load data into a trained time series prediction network (such as GRU), and output the future... Step trend load power sequence .

[0122] c) Adaptive fusion: through a designed weighting function right and The data is fused within the overlapping time domains to generate a unified load prediction sequence for subsequent optimization. The weighting principle is: the closer to the current time, the higher the weight of the feedforward prediction.

[0123] Step S2: The upper optimization layer generates the optimal reference trajectory.

[0124] Triggering condition: When the preset upper-level optimization cycle is reached (e.g., every...) Execute when a basic control cycle is reached, or when a replanning trigger signal is received from step S5.

[0125] Problem construction: based on the current state and predicted load sequence As input, build a future-oriented Step This is a constrained optimization problem where the sequence is the decision variable. The objective function is... The goal is to minimize the total operating cost (the sum of hydrogen consumption cost, fuel cell and lithium battery lifespan degradation cost) within the prediction time domain. Constraints include all physical constraints of the system and the power balance equations.

[0126] Online Solution: A fast optimization algorithm suitable for embedded platforms is used for online solution. Due to reasonable model simplification, this problem can typically be transformed into a convex optimization problem (such as quadratic programming), thus allowing for solution using sequential quadratic programming, quadratic programming, or efficient one-dimensional search algorithms. The output is a globally optimal lithium battery solution. Reference trajectory And pass it to the lower-level controller.

[0127] Step S3: Multi-objective coordination and dynamic weight decision-making

[0128] Input calculation: Calculate the instantaneous load rate in each control cycle. And current lithium batteries .

[0129] Fuzzy reasoning: and Input a pre-defined fuzzy logic system. This system reasons based on a pre-defined fuzzy rule base. The core logic of the rules is: during high load or ramp-up phases, the strategy prioritizes system dynamics and economy; in lithium batteries... When the load is low, the strategy focuses on battery protection and energy recovery; when the load is stable and... When moderate, the strategy focuses on smoothing fuel cell operation to improve its durability.

[0130] Output Solution: Through defuzzification calculation (such as the centroid method), a set of normalized real-time weight coefficients is output: economic weights. Fuel cell durability weight Lithium battery health weight These weights determine the priority of each optimization objective at the current moment.

[0131] Step S4: The lower tracking layer implements power allocation.

[0132] Rolling optimization: in each control cycle The lower-level model prediction controller (MPC) performs the following process:

[0133] a) Predictive Model: A simplified model is used to predict the impact of control actions. A typical model is a lithium battery. Integral model: ,in For efficiency coefficient, For rated capacity, To control the cycle.

[0134] b) Problem Formulation: In the Short Control Time Domain Within this framework, we construct a constrained optimization problem. Its objective function is... Mainly includes tracking the upper layer The deviation term of the reference trajectory is incorporated, along with the dynamically weighted multi-objective instantaneous cost output from step S3. .

[0135] c) Solution and Output: An efficient solver (such as a quadratic programming solver) is used for online calculation to obtain the optimal fuel cell power command at the current moment. And send it to the power actuator.

[0136] Feedback preparation: Utilizing internal prediction models, based on and predicted load Calculate the predicted system state value for the next moment, which will be used for feedback correction in subsequent control cycles.

[0137] Step S5: Constraint Management and Feedback Correction

[0138] Constraint Guarantee: During the optimization process in steps S2 and S4, all system hard constraints (such as power limits, etc.) are guaranteed. The range and current limit are strictly treated as constraints to ensure the feasibility of the solution and the safety of the system.

[0139] Feedback correction: In At any given time, obtain the actual system output (such as...) The error is compared with the model prediction from the previous time step to obtain the output error. This error is used to update the disturbance estimate of the internal model and perform feedforward compensation in the prediction of the next cycle to achieve closed-loop disturbance immunity.

[0140] Replanning Trigger: Continuous Monitoring of Actual Trajectory to upper reference trajectory The tracking error. If the error continues to exceed the preset threshold, or if there is a significant deviation between the load forecast and the actual value, the "upper-level replanning" will be triggered immediately, interrupting the current plan and forcibly returning to step S2 for global re-optimization based on the latest system status and forecast information.

[0141] Depend on Figure 1 As can be seen, the process starts from S1, passes through S2, S3, and S4, and finally reaches S5. The status feedback generated by S5 is sent back to S1 to update the information. Significant deviations detected by S5 can trigger replanning, directly jumping to S2 for global strategy re-optimization, forming a closed-loop process with adaptive capabilities.

[0142] Figure 2 The hierarchical system architecture of the method of this invention is shown, which consists of three layers:

[0143] Upper layer (global optimization decision layer): It includes three core modules: load control prediction Moqua, economic optimizer and dynamic multi-objective coordination, which are responsible for generating forward-looking optimal reference trajectory and dynamic strategy weights.

[0144] Lower layer (real-time tracking control layer): With the Model Predictive Controller (MPC) as the core, it receives instructions from the upper layer to achieve high-frequency, high-precision closed-loop power distribution.

[0145] Physical System and Feedback Layer: This layer includes the controlled object and the monitoring module. The monitoring module provides multi-channel feedback on the actual operating status of the system, which is used to update predictions, correct the controller model, and trigger upper-level replanning when necessary, thus forming a complete adaptive closed-loop control system.

[0146] 3. A step-by-step example of a complete application scenario

[0147] To more clearly demonstrate the execution logic of the method of the present invention in dynamic tasks, the following describes a simplified flight scenario segment from cruise to climb, and then to hover:

[0148] Scene activation: The drone is in a stable cruise state, lithium battery The load power is low and fluctuates little. The controller operates on a basic cycle. The system continuously runs steps S1, S3, S4, and S5 every 5 seconds, and calls step S2 every 5 seconds for global optimization.

[0149] Entering the climbing phase:

[0150] S1: The flight control system issues a climb command. Based on this motion command, the fusion predictive model predicts a significant increase in future load power.

[0151] S2 (triggered at the next 5-second cycle point): The upper-level optimizer, based on the new load prediction sequence, aims to minimize the total operating cost (while considering hydrogen consumption and component lifespan degradation) and solves for an allowable... A moderate decrease in the globally optimal reference trajectory to prioritize meeting high power demands. .

[0152] S3: As the actual load rate As the fuzzy logic system increases, it automatically raises the weight of economic efficiency based on the core rule that "high loads emphasize both economy and power efficiency". And fuel cell durability weight This allows the strategy to focus on ensuring power output and operational economy.

[0153] S4: Lower-level Model Predictive Controller (MPC) based on the new And the adjusted weights, combined with a simplified integral prediction model (such as...) The power distribution command between the fuel cell and the lithium battery is solved in a rolling manner to ensure a smooth power transition and accurate tracking of the reference trajectory.

[0154] S5: Monitors and strictly enforces physical constraints such as power and current throughout the entire process. If the climb rate exceeds expectations, resulting in actual... Deviation If the preset threshold is exceeded, replanning will be triggered immediately, forcibly jumping to S2 to regenerate the global trajectory based on the latest state.

[0155] Entering the hovering phase:

[0156] S1: The sequence of load reduction output by the prediction model.

[0157] S3: Load Rate Reduce, and if battery It has dropped to a low level, and the fuzzy logic system is based on "low" The rule of "emphasizing battery protection" automatically increases the weight of lithium battery health. This shifts the strategy towards focusing on battery energy recovery and lifespan protection.

[0158] S2 (Cycle Trigger): Based on stable load forecasting, and with economic efficiency and battery recovery as comprehensive objectives, a new route is planned to enable... A global reference trajectory showing a slow recovery.

[0159] S4: The lower-level MPC smoothly executes a power allocation strategy that prioritizes fuel cells and gently charges batteries, based on the new reference trajectory and a weight that emphasizes battery protection.

[0160] As can be seen from the above scenario examples, this invention constructs a hierarchical, forward-looking, and adaptive energy management system through a closed-loop collaborative process of five steps: "perception and prediction (S1) – global optimization (S2) – dynamic coordination (S3) – local tracking (S4) – constraint feedback (S5)". This method can proactively respond to complex and ever-changing flight missions, achieving dynamic optimization of multiple objectives while strictly ensuring system safety, thereby comprehensively improving the UAV's endurance, operational economy, and the durability of key components.

[0161] Example 3

[0162] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the fuel cell drone energy management method of Embodiment 2.

[0163] The examples above are merely illustrative of the invention and do not constitute a limitation on the scope of protection of the invention. Any design that is the same as or similar to the invention falls within the scope of protection of the invention.

Claims

1. A method for energy management of a fuel cell unmanned aerial vehicle, characterized in that, Includes the following steps: Step S1, Hybrid power system state awareness and load management: The system acquires real-time state parameters of the UAV hybrid power system and, based on UAV flight motion control information and historical load data, predicts the load power demand sequence for a future time domain using a load condition fusion prediction model. Where k is the current time, To predict the time-domain step size; Step S2: The upper optimization layer generates the optimal reference trajectory. Establish a system that includes the cost of hydrogen consumption. Equivalent cost of fuel cell lifespan degradation and equivalent cost of lithium battery life degradation The economic objective function JE; Inputting the load power demand sequence and the state parameters of the hybrid power system within the prediction time domain, and under the condition of satisfying system constraints, the optimization algorithm is used to solve online within the prediction time domain to obtain the globally optimal lithium battery state of charge reference trajectory. ; Step S3, Multi-objective Coordination and Dynamic Weight Decision: Construct a multi-objective optimization function that includes economic efficiency, fuel cell durability, and lithium battery health status. ; A fuzzy logic algorithm is used to dynamically adjust the weight coefficients of each sub-objective in the multi-objective optimization function based on the real-time normalized load power and the real-time state of charge of the lithium battery. Step S4, the lower tracking layer implements power allocation: The Model Predictive Control (MPC) method is adopted, with the globally optimal lithium battery state of charge reference trajectory as the tracking target, the load power demand sequence in the predicted time domain as the feedforward information, and the multi-objective optimization function with dynamic weights as the optimization index. Rolling optimization and feedback correction are performed to solve and output the power control command of the fuel cell in real time to complete the power distribution. Step S5, Constraint Management and Feedback Correction: During the optimization process in steps S2 and S4, physical constraints of the system are continuously applied and satisfied; feedback correction is performed based on the deviation between the actual output and the predicted output of the system; and when the replanning trigger condition is met, step S2 is triggered to be re-executed, and a new optimal lithium battery state of charge reference trajectory is output to ensure the overall optimality of the system.

2. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The state parameters of the UAV hybrid power system described in step S1 constitute a state vector. It includes at least: the state of charge (SOC) value of the lithium battery. Current output power of fuel cells or output current and DC bus voltage .

3. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The load condition fusion prediction model described in step S1 has the function of combining feedforward motion information and feedback historical data. Its construction method is as follows: The feedforward prediction results based on UAV motion control information and the prediction results of a time-series neural network trained on historical load data are adaptively weighted and fused in the overlapping prediction time domains to generate a unified future load power demand sequence. ; The weighting coefficients are dynamically adjusted as the prediction time step increases; the closer to the current time, the higher the weight of the feedforward prediction result.

4. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The economic objective function J mentioned in step S2 E It is the sum of the total expected operating costs within the prediction time domain, and its general formula is: Among them, the predicted hydrogen consumption cost in the time domain , This is the unit hydrogen price coefficient. Hydrogen consumption rate of fuel cells; Predicting fuel cell lifespan degradation costs over the time domain , To convert the fuel cell life degradation into an equivalent factor for economic cost, its determination is based on fuel cell aging experiments or average replacement cycle data. It is a measure of instantaneous lifespan degradation and is strongly correlated with the operating state of the fuel cell. , The rate of change weighting coefficient. To represent the 0-1 variables for start-up and shutdown, This refers to the start / stop weighting coefficient. This represents the absolute value of the change in the fuel cell's output power. Lithium battery life degradation cost ,in, To convert the lithium battery life degradation into an equivalent factor for economic cost, its determination is based on the lithium battery cycle life test or statistical life data. As a measure of instantaneous lifetime decay, Used to punish high-current charging and discharging stress, among which A stress acceleration factor greater than 1 This is the battery's rated current.

5. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The optimization algorithm mentioned in step S2 is a sequential quadratic programming algorithm, a quadratic programming algorithm, or an optimization algorithm capable of fast one-dimensional search.

6. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The multi-objective optimization function mentioned in step S3 ,in, Let the economic objective function be... For the sake of fuel cell durability, The cost to the health of lithium batteries; In the formula, The weighting factor for high output current. The weighting coefficient for the drastic rate of change of current; In the formula, For ideal state of charge The weighting coefficients, This is a weighting factor for high charge / discharge currents.

7. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The Model Predictive Control (MPC) method described in step S4 uses a quadratic programming solver to solve the rolling optimization problem online quickly. The specific process is as follows: Build optimization metrics Includes tracking the upper layer. The deviation term of the reference trajectory is incorporated into a multi-objective optimization function. ; An efficient solver is used for online calculation to obtain the optimal fuel cell power command at the current moment. And send it to the power actuator; Using an internal prediction model, based on and predicted load Calculate the predicted system state value for the next moment, which will be used for feedback correction in subsequent control cycles.

8. The energy management method for a fuel cell unmanned aerial vehicle according to claim 7, characterized in that, The internal prediction model is for lithium batteries. Integral model: ,in For efficiency coefficient, For rated capacity, To control the cycle.

9. The energy management method for a fuel cell unmanned aerial vehicle according to claim 1, characterized in that, The physical operating constraints mentioned in step S5 include: upper and lower limits of fuel cell output power and its rate of change, upper and lower limits of lithium battery state of charge operating range, upper and lower limits of lithium battery charging and discharging current, and system power balance constraints. The feedback correction mentioned in step S5 includes the immediate correction of output error by the lower-level MPC and the replanning trigger mechanism; The replanning trigger mechanism includes setting a deviation threshold. When the actual Trajectory and upper-level optimal reference trajectory The deviation continues to exceed If the mean square error of the load prediction exceeds a certain threshold, the mechanism will trigger the upper optimization layer to re-execute step S2 and generate a new global reference trajectory based on the latest system state and prediction information.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fuel cell drone energy management method as described in any one of claims 1 to 9.

Images