A predictive energy management method for hybrid unmanned aerial vehicles based on a pilot model

Abstract

This invention relates to a hybrid-powered unmanned aerial vehicle (UAV) energy management method, and more particularly to a UAV energy management method based on a pilot model, belonging to the field of UAV energy management technology. Based on the analysis of pilot operation characteristics, this invention constructs a power prediction model. According to the prediction time domain length, the weight matrix of the problem-solving framework in the model, and the constraints reflecting the pilot's operating style and the predicted impact on operation, a hybrid power system energy allocation strategy that balances range and power system lifespan is obtained. This achieves an energy management system that adapts to the pilot's needs, improving flight economy and personalization.

Description

Technical Field

[0001] This invention relates to a hybrid-powered unmanned aerial vehicle (UAV) energy management method, and more particularly to a UAV energy management method based on a pilot model, belonging to the field of UAV energy management technology. Background Technology

[0002] To address crises such as energy shortages and global warming, hybrid power technology has become a key focus in the transportation sector. Conventional transportation equipment uses internal combustion engines, which consume large amounts of fossil fuels and emit greenhouse gases, a major cause of energy shortages and global warming. Optimizing power technologies in the transportation sector is a viable way to alleviate these crises. Hybrid power vehicles, using lithium batteries, internal combustion engines, or fuel cells as energy sources, reduce fuel consumption and improve range to some extent, and have attracted widespread attention as a new type of power-driven transportation equipment.

[0003] Currently, energy management technologies for hybrid-powered vehicles are mainly focused on new energy vehicles, with relatively little research on energy management for aircraft. However, aircraft operate under more complex conditions and have greater operational freedom, and simply transferring vehicle energy management technologies has limited effectiveness in optimizing power source energy conservation. Therefore, suitable energy management methods are needed to reduce the energy consumption of hybrid aircraft.

[0004] In existing research on energy management strategies for hybrid power systems, the parameters are set fixed, and the optimization strategies are usually designed from the perspective of optimal energy consumption, while ignoring the role of the pilot in controlling the aircraft. Experienced pilots will assess the future impact of their actions on the aircraft and apply this judgment to actual maneuvers. In addition, pilots will constrain the use of different power sources according to mission requirements to achieve optimal energy-saving effects. Summary of the Invention

[0005] To address the aforementioned technical problems in this field, the main objective of this invention is to provide a predictive energy management method based on a pilot model. This method constructs a power prediction model based on the analysis of pilot operation characteristics. By considering the prediction time domain length, the weight matrix of the problem-solving framework in the model, and the constraints reflecting the pilot's operating style and the predicted impact on operation, a hybrid power system energy allocation strategy that balances range and power system lifespan is obtained. This achieves an energy management system that adapts to the pilot's needs, improving the economy and personalization of flight.

[0006] The objective of this invention is achieved through the following technical solution.

[0007] This invention discloses a predictive energy management method based on a driver model, comprising the following steps:

[0008] Step 1: Acquiring Flight Status; Acquiring pilot information and drone status information to control the drone's flight; Pilot information includes throttle position. Throttle movement speed The drone's status information includes flight speed. Flight acceleration Flight attitude angle Flight attitude angular velocity Power Requirement and the state of charge of the power battery The above-mentioned driver information and UAV status information data commands are referred to as raw status data;

[0009] Step 2: Calculation of driving characteristic parameters; The original state data obtained in Step 1 over a period of time is processed to obtain driving characteristic parameters;

[0010] Table 1 Driving Characteristic Parameters

[0011]

[0012] Step 3: Normalize the driving feature parameters obtained in Step 2 to obtain normalized driving feature parameters, which are named normalized feature parameters:

[0013] (1)

[0014] In the formula Represents the feature parameters before normalization. Represents the normalized feature parameters. The maximum value of the characteristic parameter. This represents the minimum value of the characteristic parameter;

[0015] Step 4: Use principal component analysis to screen the normalized feature parameters obtained in Step 3, selecting the combination that best represents the driver's characteristics, i.e., the combination with a cumulative contribution rate exceeding [a certain percentage]. The former Principal Components The parameters used to characterize driver feature classification are named feature principal components;

[0016] Step 5: Cluster the principal components of the features from Step 4 using the GMM clustering algorithm. The clusters are divided into... The driving style of a driver is defined based on the clustering results; the principal components of the clustered features are labeled with classification categories, and are called labeled data.

[0017] Step 6: Taking into account the principal components of the driver's features, labeled data, cluster categories, and UAV dynamic equations, a power prediction model based on the driver model is established using the K-Nearest Neighbors (KNN) algorithm and Model Predictive Control (MPC) to complete the power prediction considering the driver's driving style.

[0018] S61. Divide the labeled data from step five into a training set and a test set. The dataset format is as follows:

[0019] (2)

[0020] in The feature vector is composed of principal component information, which contains the first feature vector selected in step five. Principal Components , As a style category, , To mark the total number of data;

[0021] S62. Use the KNN classification method to classify driver style. The KNN classification method requires supervised learning on the training set data to obtain a suitable classification model, and use the test set to test the classification effect.

[0022] The established KNN classification model is as follows:

[0023] (3)

[0024] In the formula, This represents the N driver categories generated by clustering in step five;

[0025] S63. Combining the UAV state information obtained in step one, ignoring the aerodynamic forces of the fuselage, establish the UAV dynamics model and attitude kinematics model:

[0026] (4) (5) (6)

[0027] In the formula Indicates the quality of the drone. Indicates the angular velocity of the machine body. Represents the gravitational acceleration vector. Represents the tension vector. Represents the inertia matrix. Indicates the angular acceleration of the body. This indicates the torque acting on the body. A matrix representing the relationship between attitude angular velocity and body angular velocity;

[0028] S64. Based on the principle of small perturbation and the Euler difference method, equations (4), (5) and (6) in S63 are transformed into discrete linear state-space equations for solving the model prediction optimization problem:

[0029] (7) (8)

[0030] State quantity in the formula Control quantity , Represents output quantity. Represents the state matrix. Represents the control matrix. Represents the output matrix. represent time, represent The moment One discrete period;

[0031] S65. Establishing the objective function of the model prediction framework This is to illustrate the impact of experienced pilots' predictive abilities on the mission performance of drones and the pilot's driving style:

[0032] (9)

[0033] In the formula Indicates the desired operating condition. Indicates the increment of the control quantity. The weight matrix represents the expected tracking effect of the operating conditions. The weight matrix represents the increment of the control quantity. It is the predicted step size. It controls the step size; it adjusts... The size of the indicator reflects the driver's predictive ability; adjustments can be made accordingly. , The values ​​of the matrix can reflect the driver's style;

[0034] S66. Based on the clustering results from step five, set the corresponding parameters. The model predictive controller parameters are set; the model predictive controller parameters include constraints, weight matrix, and prediction step size; based on the driver style identified in step S62. Select the appropriate model to predict the controller parameters for problem-solving and optimization;

[0035] When the recognition result is Class time:

[0036] (10)

[0037] in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size;

[0038] When the recognition result is Class time:

[0039] (11)

[0040] in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size;

[0041] When the recognition result is Class time:

[0042] (12)

[0043] in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size;

[0044] S67. Solve the optimization problems in steps S65 and S66 using dynamic programming algorithm to obtain the optimal control quantity. Establish a connection between the efficiency matrix and rotor speed:

[0045] (13)

[0046] In the formula The control efficiency matrix for a quadcopter drone is related to the drone's size and rotor type. Given the rotor speed; using the power demand mapping relationship, the predicted power demand is obtained:

[0047] (14)

[0048] In the formula For the predicted power demand, The rotor power coefficient, This is the rotor serial number.

[0049] Step 7: Based on the integrated UAV status information and predicted power demand, use Model Predictive Control (MPC) to establish a hybrid UAV energy management method, and achieve optimal allocation of power between the aero-engine and the power battery. The establishment principle of Model Predictive Control and the solution of the optimization problem are the same as in Step 6.

[0050] S71. Based on the way the battery current is expressed, the rate of change of the power battery's state of charge (SOC) is used as the system state equation:

[0051] (15)

[0052] In the formula, For the power battery current, This is the open-circuit voltage of the power battery. The internal resistance of the power battery, This refers to the engine's output power.

[0053] Equation (15) is transformed into a discrete linear state-space equation:

[0054] (16)

[0055] In the formula, cons is a constant term, and the state variable is... The control quantity is , Represents the state matrix. Represents the control matrix. Represents the interference term matrix, represent time, represent The moment One discrete period;

[0056] S72. To ensure the power battery's performance while meeting the power requirements of the drone, the hybrid system... Follow the global optimum The curve ensures that the engine operates at its optimal fuel consumption curve. The objective function of the model prediction framework is established as follows. :

[0057] (17)

[0058] In the formula As a weight for fuel consumption, For fuel consumption rate, To predict the step size, To control the step size, Represents global optimality The weight matrix for curve tracking effect. A weight matrix representing the optimal fuel consumption curve tracking performance;

[0059] S73. Set the following constraints:

[0060] (18)

[0061] in Indicates the minimum size of the power battery , Indicates the maximum power battery capacity , Indicates the minimum power of the power battery. Indicates the maximum power of the power battery. Indicates the minimum current of the power battery. Indicates the maximum current of the power battery. Indicates the engine's minimum power. Engine maximum power;

[0062] S74. Repeat steps one through seven to solve equation (17) to obtain the real-time optimal engine output power. This enables optimized energy management for hybrid drones to meet the needs of pilots until the end of the flight mission.

[0063] Beneficial effects:

[0064] 1. This invention constructs a driver model to predict power, which can compensate the driver's prediction of the operation effect in energy management, adjust the performance of the UAV, and improve the ability to cope with complex working conditions and make multi-degree-of-freedom action decisions. Compared with the direct transfer of vehicle energy management technology, it has stronger adaptability.

[0065] 2. This invention can meet the individual needs of the pilot in UAV energy management by recognizing the pilot's style and adjusting the model predictive control parameters, thereby achieving the goal of adapting the UAV to the pilot.

[0066] 3. This invention adopts a two-layer model predictive control architecture. The upper layer predicts the required power through a pilot model, and the lower layer optimizes energy allocation through predictive energy management, which can improve the accuracy of power prediction and the economy of flight.

[0067] 4. This invention uses the GMM clustering algorithm to classify the driving style of drone pilots. Compared with the commonly used K-means clustering method, it has better discrimination for different types of samples, can output more classification information, and is more suitable for the complex working conditions of drones. Attached Figure Description

[0068] Figure 1 This is a flowchart of a predictive energy management method for hybrid unmanned aerial vehicles based on a pilot model, according to an embodiment of the present invention.

[0069] Figure 2 This is a schematic diagram of the clustering results of the GMM model based on driving features.

[0070] Figure 3 Model predictive control principle block diagram

[0071] Figure 4 This is a flowchart illustrating the parameter selection process for the driver model based on driving style, as described in an embodiment of the present invention.

[0072] Figure 5 A schematic diagram for solving dynamic programming problems.

[0073] Figure 6 This is an embodiment of the present invention. Consumption and Global Optimum Comparison diagram of reference curves Detailed Implementation

[0074] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0075] A predictive energy management method for hybrid unmanned aerial vehicles based on a pilot model, the method flow is as follows: Figure 1 As shown, it includes the following steps:

[0076] Step 1: Acquiring Flight Status; Acquiring pilot information and drone status information to control the drone's flight; Pilot information includes throttle position. Throttle movement speed The drone's status information includes flight speed. Flight acceleration Flight attitude angle Attitude angular velocity Power Requirement and the state of charge of the power battery The aforementioned pilot information and UAV status information data commands are referred to as raw status data; power demand. and the state of charge of the power battery Based on the physical model, the power requirement of the rotary-wing UAV is mapped as follows:

[0077] (1)

[0078] In the formula The rotor power coefficient, The rotor speed, This refers to the rotor serial number. The power battery model uses the Rint model, and its output power is expressed as:

[0079] (2)

[0080] In the formula For the output power of the power battery, This is the open-circuit voltage of the power battery. For the power battery current, The internal resistance of the power battery and the battery current. It can be represented as:

[0081] (3)

[0082] The power battery is obtained using the integral method. :

[0083] (4)

[0084] In the formula For the initial battery value, This refers to the battery capacity.

[0085] Step 2: Calculation of driving characteristic parameters. The raw state data obtained in Step 1 over a certain period is processed to obtain the driving characteristic parameters.

[0086] Table 1 Driving Characteristic Parameters

[0087]

[0088] Step 3: Normalize the driving feature parameters obtained in Step 2 to obtain normalized driving feature parameters, which are named normalized feature parameters:

[0089] (5)

[0090] In the formula Represents the feature parameters before normalization. Represents the normalized feature parameters. The maximum value of the characteristic parameter. This represents the minimum value of the characteristic parameter;

[0091] Step 4: Principal component analysis is used to screen the 18 normalized feature parameters obtained in Step 3, selecting the combination that best represents the driver's characteristics, i.e., the combination with a cumulative contribution rate exceeding [a certain percentage]. The former Principal Components The parameters used to characterize driver feature classification are named feature principal components;

[0092] S41. Calculate the correlation coefficient matrix. :

[0093] (6)

[0094] In the formula Indicates the first The and the first The correlation coefficient of each normalized feature parameter and Indicates the first The and the first Sample values ​​of each normalized feature parameter. and The sample mean of the normalized feature parameters. Given the total number of samples, the normalized feature parameter sample mean is calculated as follows:

[0095] (7)

[0096] S42. Calculate the correlation coefficient matrix. From the eigenvalues ​​and eigenvectors, we obtain the eigenvalues. and corresponding feature vectors The feature transformation matrix is ​​represented as:

[0097] (8)

[0098] Principal component data matrix The calculation method is as follows:

[0099] (9)

[0100] In the formula Represents the normalized feature parameter data matrix ;

[0101] S43. Calculate the contribution rate of the principal components. Contribution rate The calculation method is as follows:

[0102] (10)

[0103] forward Principal Components The cumulative contribution rate is:

[0104] (11)

[0105] The cumulative contribution rate exceeded The former Principal Components As a parameter characterizing driver features for classification;

[0106] Step 5: Cluster the principal components of the features from Step 4 using the GMM clustering algorithm. The clusters are divided into... The driving style of a driver is defined based on the clustering results; the principal components of the clustered features are labeled with classification categories, and are called labeled data.

[0107] S51, to obey The probability density function of the GMM composed of multiple Gaussian distributions Represented as:

[0108] (12)

[0109] In the formula For the first The weight of each principal component For the first The mean of each principal component, For the first The covariance of each principal component For the first Gaussian distribution density function of each principal component;

[0110] S52, Set unknown parameters , and The initial value is , and The Expectation-Maximization (EM) algorithm is used iteratively until the parameters converge, yielding the desired result. , and That is, the GMM model based on driving characteristics, the clustering results are as follows Figure 2 As shown;

[0111] Step Six: Taking into account the driver's principal components, labeled data, cluster categories, and UAV dynamic equations, a power prediction model based on the driver model is established using the K-Nearest Neighbors (KNN) algorithm and Model Predictive Control (MPC). This completes the power prediction considering the driver's driving style. The principle of Model Predictive Control is as follows: Figure 3 As shown, the parameter selection process based on driving style is as follows: Figure 4 As shown;

[0112] S61. Divide the labeled data from step five into a training set and a test set. The dataset format is as follows:

[0113] (13)

[0114] in The feature vector is composed of principal component information, which contains the first feature vector selected in step five. Principal Components , As a style category, , To mark the total number of data;

[0115] S62. Use the KNN classification method to classify driver style. The KNN classification method requires supervised learning on the training set data to obtain a suitable classification model, and use the test set to test the classification effect.

[0116] The established KNN classification model is as follows:

[0117] (14)

[0118] In the formula, This represents the N driver categories generated by clustering in step five;

[0119] S63. Combining the UAV state information obtained in step one, and ignoring the aerodynamic forces of the fuselage, establish the UAV dynamics model and attitude kinematics model in the vertical plane:

[0120] (15) (16) (17) (18)

[0121] In the formula This represents the velocity along the lower x-axis of the body coordinate system. This represents the acceleration along the x-axis in the body coordinate system. This represents the velocity along the lower z-axis of the body coordinate system. This represents the acceleration along the z-axis in the body coordinate system. This represents the angular velocity of rotation about the y-axis in the body coordinate system. Indicates the quality of the drone. Represents gravitational acceleration. Indicates tension. This represents the torque about the y-axis in the body coordinate system. Represents the principal product of inertia about the y-axis, where ;

[0122] S64. Based on the small perturbation principle and the Euler difference method, equations (15), (16), (17), and (18) in S63 are transformed into discrete linear state-space equations for solving the model prediction optimization problem:

[0123] (19)

[0124] State quantity in the formula Control quantity , Represents output quantity. Represents the state matrix. Represents the control matrix. Represents the output matrix. represent time, represent The moment One discrete period;

[0125] S65. Establishing the objective function of the model prediction framework This is to illustrate the impact of experienced pilots' predictive abilities on the mission performance of drones and the pilot's driving style:

[0126] (20)

[0127] In the formula Indicates the desired operating condition. Indicates the increment of the control quantity. The weight matrix represents the expected tracking effect of the operating conditions. The weight matrix represents the increment of the control quantity. It is the predicted step size. It controls the step size; it adjusts... The size of the indicator reflects the driver's predictive ability; adjustments can be made accordingly. , The values ​​of the matrix can reflect the driver's style;

[0128] S66. Based on the clustering results from step five, set the corresponding parameters. The model predictive controller parameters are set; the model predictive controller parameters include constraints, weight matrix, and prediction step size; based on the driver style identified in step S62. Select the appropriate model to predict the controller parameters for problem-solving and optimization;

[0129] When the recognition result is Class time:

[0130] (21)

[0131] in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size;

[0132] When the recognition result is Class time:

[0133] (22)

[0134] in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size;

[0135] When the recognition result is Class time:

[0136] (23)

[0137] in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size;

[0138] The optimization problems described in steps S67, S65, and S66 are solved using a dynamic programming algorithm. This transforms the multi-time-stage, multi-objective solution in the prediction model into a solution for multiple single-time-stage problems. After the inverse search is completed, the optimal control sequence is obtained through forward iteration. The first set of control inputs is then output to the system. The solution process is as follows: Figure 5 As shown, the solved optimal control quantity Establish a connection between the efficiency matrix and rotor speed:

[0139] (24)

[0140] In the formula The control efficiency matrix for a quadcopter UAV is related to the UAV size and rotor type. The predicted power is obtained using the power demand mapping relationship from equation (1) in step one.

[0141] Step 7: Based on the integrated UAV status information and predicted power demand, use Model Predictive Control (MPC) to establish a hybrid UAV energy management method, and achieve optimal allocation of power between the aero-engine and the power battery. The establishment principle of Model Predictive Control and the solution of the optimization problem are the same as in Step 6.

[0142] S71. According to the battery current expression (3) in step one, the power battery Rate of change as the system state equation:

[0143] (25)

[0144] In the formula, For the power battery current, This is the open-circuit voltage of the power battery. The internal resistance of the power battery, For the engine output power; transform equation (25) into a discrete linear state-space equation:

[0145] (26)

[0146] In the formula The constant term is and the state variable is . The control quantity is , Represents the state matrix. Represents the control matrix. Represents the interference term matrix, represent time, represent The moment One discrete period;

[0147] S72. In order to enable the hybrid system to meet the power requirements of the drone while allowing the power battery to... Try to follow the global optimal The curve ensures that the engine operates at its optimal fuel consumption curve. The objective function of the model prediction framework is established as follows. :

[0148] (27)

[0149] In the formula As a weight for fuel consumption, For fuel consumption rate, To predict the step size, To control the step size, Represents global optimality The weight matrix for curve tracking effect. A weight matrix representing the optimal fuel consumption curve tracking performance;

[0150] S73. Set the following constraints:

[0151] (28)

[0152] in Indicates the minimum size of the power battery , Indicates the maximum power battery capacity , Indicates the minimum power of the power battery. Indicates the maximum power of the power battery. Indicates the minimum current of the power battery. Indicates the maximum current of the power battery. Indicates the engine's minimum power. Engine maximum power;

[0153] S74. Repeat steps one through seven to solve equation (27) to obtain the real-time optimal engine output power. This invention enables energy management of hybrid drones to adapt to the pilot's needs until the end of the flight mission. It is particularly useful for energy management of drones controlled by pilots with more aggressive flying styles. Consumption and Global Optimum The comparison results of the curves are as follows Figure 6 As shown. Because the power output response of the battery in a hybrid electric vehicle pack is faster, the battery adjusts its power first to cope with sudden changes in operating conditions. In this control method, the battery, while ensuring economy, can cope with sudden changes in operating conditions. The energy consumption is more timely, meeting the needs of drivers with aggressive driving styles for rapid changes in instantaneous energy, thus verifying the effectiveness of the method.

[0154] This invention establishes an energy management method that considers the pilot model within a model prediction framework. It performs feature parameter analysis and principal component analysis on the collected UAV state data to reduce data dimensionality and uncover personalized features reflecting the pilot's style. These personalized features are then assigned category labels using the GMM clustering method, providing a training dataset for the pilot model built based on the KNN feature classification method and the UAV kinematics model. After training, a power prediction model based on the pilot model is obtained, which is ultimately combined with the underlying energy prediction and management method to achieve pilot model-based energy management.

[0155] The above-described specific details are a further detailed explanation of the purpose, technical solution, and beneficial effects of the invention. It should be understood that the above description is only a part of the embodiments of the present invention flying in the vertical plane of a quadcopter UAV, and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A predictive energy management method based on a driver model, comprising the following steps: Step 1: Acquiring Flight Status; Acquiring pilot information and drone status information to control the drone's flight; Pilot information includes throttle position. Throttle movement speed The drone's status information includes flight speed. Flight acceleration Flight attitude angle Flight attitude angular velocity Power Requirement and the state of charge of the power battery The above-mentioned driver information and UAV status information data commands are referred to as raw status data; Step 2: Calculation of driving characteristic parameters; The original state data obtained in Step 1 over a period of time is processed to obtain driving characteristic parameters, as shown in the table below; Step 3: Normalize the driving feature parameters obtained in Step 2 to obtain normalized driving feature parameters, which are named normalized feature parameters: In the formula Represents the feature parameters before normalization. Represents the normalized feature parameters. The maximum value of the characteristic parameter. This represents the minimum value of the characteristic parameter; Step 4: Use principal component analysis to screen the normalized feature parameters obtained in Step 3, selecting the combination that best represents the driver's characteristics, i.e., the combination with a cumulative contribution rate exceeding [a certain percentage]. The former Principal Components The parameters used to characterize driver feature classification are named feature principal components; Step 5: Cluster the principal components of the features from Step 4 using the GMM clustering algorithm. The clusters are divided into... The driving style of a driver is defined based on the clustering results; the principal components of the clustered features are labeled with classification categories, and are called labeled data. Step 6: Taking into account the principal components of the driver's features, labeled data, cluster categories, and UAV dynamic equations, a power prediction model based on the driver model is established using the K-Nearest Neighbors (KNN) algorithm and Model Predictive Control (MPC) to complete the power prediction considering the driver's driving style. S61. Divide the labeled data from step five into a training set and a test set. The dataset format is as follows: in The feature vector is composed of principal component information, which contains the first feature vector selected in step five. Principal Components , As a style category, , To mark the total number of data; S62. Use the KNN classification method to classify driver style. The KNN classification method requires supervised learning on the training set data to obtain a suitable classification model, and use the test set to test the classification effect. The established KNN classification model is as follows: In the formula, The clustering generated in step five Types of drivers; S63. Combining the UAV state information obtained in step one, ignoring the aerodynamic forces of the fuselage, establish the UAV dynamics model and attitude kinematics model: In the formula This represents the velocity along the lower x-axis of the body coordinate system. This represents the acceleration along the x-axis in the body coordinate system. This represents the velocity along the lower z-axis of the body coordinate system. This represents the acceleration along the z-axis in the body coordinate system. This represents the angular velocity of rotation about the y-axis in the body coordinate system. Indicates the quality of the drone. Represents gravitational acceleration. Indicates tension. This represents the torque about the y-axis in the body coordinate system. Represents the principal product of inertia about the y-axis, where ; S64. Based on the principle of small perturbation and the Euler difference method, equations (4), (5), (6) and (7) in S63 are transformed into discrete linear state-space equations for solving the model prediction optimization problem: State quantity in the formula Control quantity , Represents output quantity. Represents the state matrix. Represents the control matrix. Represents the output matrix. represent time, represent The moment One discrete period; S65. Establishing the objective function of the model prediction framework This is to illustrate the impact of experienced pilots' predictive abilities on the mission performance of drones and the pilot's driving style: In the formula Indicates the desired operating condition. Indicates the increment of the control quantity. The weight matrix represents the expected tracking effect of the operating conditions. The weight matrix represents the increment of the control quantity. It is the predicted step size. It controls the step size; it adjusts... The size of the indicator reflects the driver's predictive ability; adjustments can be made accordingly. , The values ​​of the matrix can reflect the driver's style; S66. Based on the clustering results from step five, set the corresponding parameters. The model predictive controller parameters are set; the model predictive controller parameters include constraints, weight matrix, and prediction step size; based on the driver style identified in step S62. Select the appropriate model to predict the controller parameters for problem-solving and optimization; When the recognition result is Class time: in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size; When the recognition result is Class time: in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size; When the recognition result is Class time: in, express Class minimum speed, express Maximum speed of class express Minimum body angular velocity of a class express Maximum body angular velocity of the class Indicates the minimum attitude angle. Indicates the maximum attitude angle. express Class tracking effect weight matrix, express Class control incremental weight matrix, express Prediction step size, Indicates the control step size; S67. Solve the optimization problems in steps S65 and S66 using dynamic programming algorithm to obtain the optimal control quantity. Establish a connection between the efficiency matrix and rotor speed: In the formula The control efficiency matrix for a quadcopter drone is related to the drone's size and rotor type. Given the rotor speed; using the power demand mapping relationship, the predicted power demand is obtained: In the formula For the predicted power demand, The rotor power coefficient, The rotor serial number; Step 7: Based on the integrated UAV status information and predicted power demand, use Model Predictive Control (MPC) to establish a hybrid UAV energy management method, and achieve optimal allocation of power between the aero-engine and the power battery. The establishment principle of Model Predictive Control and the solution of the optimization problem are the same as in Step 6. S71. Based on the way the battery current is expressed, the rate of change of the power battery's state of charge (SOC) is used as the system state equation: In the formula, For the power battery current, This is the open-circuit voltage of the power battery. The internal resistance of the power battery, This refers to the engine's output power. Equation (15) is transformed into a discrete linear state-space equation: In the formula, #imgpt157# is a constant term, #imgpt158# is the state variable, #imgpt159# is the control variable, #imgpt160# represents the state matrix, #imgpt161# represents the control matrix, #imgpt162# represents the disturbance term matrix, #imgpt163# represents time #imgpt164#, and #imgpt165# represents the #imgpt167#th discrete period at time #imgpt166#. S72. To ensure that the hybrid system meets the power requirements of the UAV, guarantees that the power battery follows the globally optimal curve, and simultaneously ensures that the engine operates on the optimal fuel consumption curve, the objective function of the following model prediction framework is established: In the formula, #imgpt173# represents the fuel consumption weight, #imgpt174# represents the fuel consumption rate, #imgpt175# represents the prediction step size, #imgpt176# represents the control step size, #imgpt177# represents the weight matrix of the globally optimal #imgpt178# curve tracking effect, and #imgpt179# represents the weight matrix of the best fuel consumption curve tracking effect; S73. Set the constraints as follows: #imgpt180#=#imgpt181# Where #imgpt183# represents the minimum power battery, #imgpt184# represents the maximum power battery, #imgpt185# represents the maximum power battery, #imgpt186# represents the minimum power battery, #imgpt187# represents the minimum power battery, #imgpt188# represents the maximum power battery, #imgpt189# represents the minimum power battery, #imgpt190# represents the maximum power battery, #imgpt191# represents the minimum engine power, and #imgpt192# represents the maximum engine power. S74. Repeat steps one to seven to solve equation (17) to obtain the real-time optimal engine output power #imgpt193#, thereby achieving energy management optimization of the hybrid UAV to meet the needs of the pilot until the flight mission ends.

Images